Noncommutative Geometry and Global Analysis

CONTEMPORARY
MATHEMATICS
546
Noncommutative Geometry
and Global Analysis
Conference in Honor of Henri Moscovici
June 29–July 4, 2009
Bonn, Germany
Alain Connes
Alexander Gorokhovsky
Matthias Lesch
Markus Pflaum
Bahram Rangipour
Editors
American Mathematical Society

and Global Analysis
H. Moscovici
CONTEMPORARY
MATHEMATICS
546
and Global Analysis
Conference in Honor of Henri Moscovici
June 29–July 4, 2009
Bonn, Germany
Alain Connes
Matthias Lesch
Markus Pflaum
Bahram Rangipour
Editors
American Mathematical Society

Providence, Rhode Island
Editorial Board
Dennis DeTurck, managing editor
George Andrews Abel Klein Martin J. Strauss
2010 Mathematics Subject Classification. Primary 13F35, 14B05, 16E40, 16T05, 19K56,
19D55, 22E46, 58B34, 53C24, 58J42.
Frontispiece photo courtesy of John R. Copeland.
Library of Congress Cataloging-in-Publication Data

Noncommutative geometry and global analysis : conference in honor of Henri Moscovici, June 29–
July 4, 2009, Bonn, Germany / Alain Connes . . . [et al.], editors.
p. cm. — (Contemporary mathematics ; v. 546)
Includes bibliographical references and index.
ISBN 978-0-8218-4944-6 (alk. paper)
1. Commutative rings—Congresses. 2. Noncommutative algebras—Congresses. 3. Global
analysis (Mathematics)—Congresses. I. Moscovici, Henri, 1944– II. Connes, Alain.
QA251.3.N664 2011
512.44—dc22
2011008749
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
Contents
Introduction vii
Conference Talks xi
List of Participants xix
Dirac cohomology and unipotent representations of complex groups
D. Barbasch and P. Pandžić 1
Algebraic index theorem for symplectic deformations of gerbes
P. Bressler, A. Gorokhovsky, R. Nest, and B. Tsygan 23
Index theory for basic Dirac operators on Riemannian foliations
J. Brüning, F.W. Kamber, and K. Richardson 39
The Witt construction in characteristic one and quantization
A. Connes 83
Lie prealgebras
M. Dubois-Violette and G. Landi 115
On the analogy between complex semisimple groups
and their Cartan motion groups
N. Higson 137
A survey on Hopf-cyclic cohomology and Connes-Moscovici characteristic map
A. Kaygun 171
A super version of Connes-Moscovici Hopf algebra
M. Khalkhali and A. Pourkia 181
Analytic torsion of Z2 -graded elliptic complexes
V. Mathai and S. Wu 199
The K-groups and the index theory of certain comparison C ∗ -algebras
B. Monthubert and V. Nistor 213
Relative pairings and the Atiyah-Patodi-Singer index formula for the
Godbillon-Vey cocycle
H. Moriyoshi and P. Piazza 225
Two exact sequences for lattice cohomology
A. Némethi 249
Cup products in Hopf cyclic cohomology with coefficients in contramodules
B. Rangipour 271
v
vi CONTENTS
Algebras of p-symbols, noncommutative p-residue, and the Brauer group

M. Wodzicki 283
Large scale geometry and its applications
G. Yu 305
Introduction
This volume represents the proceedings of the conference in honor of Henri

Moscovici held in Bonn, June 29 – July 4, 2009. Moscovici has made fundamental
contributions to Noncommutative Geometry, Global Analysis and Representation
Theory. Especially, his 30-year long collaboration with A. Connes has led to a num-
ber of foundational results. They obtained an L2 -index theorem for homogeneous
spaces of general Lie groups, generalizing the Atiyah-Schmid theorem for semisim-
ple Lie groups and a higher index theorem for multiply connected manifolds. The
latter played a key role in their proof of the Novikov conjecture for word-hyperbolic
groups. In the course of their work on the transverse geometry of foliations they
discovered the local index formula for spectral triples. The calculations based on the
latter formula provided the impetus for the development of the cyclic cohomology
theory of Hopf algebras.
In the words of Alain Connes: “I have always known Henri as a prince who
escaped from the dark days of the communist era in Romania. With his Mediter-
ranean charm and his intense intelligence, so often foresighted, but never taking
himself too seriously, with his inimitable wit, and his constant regard for others he
certainly stands out among mathematicians as a great and lovable exception.”
The present volume, which includes articles by leading experts in the fields
mentioned above, provides a panoramic view of the interactions of noncommutative
geometry with a variety of areas of mathematics. It contains several surveys as well
as high quality research papers. In particular, it focuses on the following themes:
geometry, analysis and topology of manifolds and singular spaces, index theory,
group representation theory, connections between noncommutative geometry and
number theory and arithmetic geometry, Hopf algebras and their cyclic cohomology.
We now give brief summaries of the papers appearing in this volume.
1. D. Barbasch and P. Pandžić “Dirac cohomology, unipotent representa-

tions.”
In this paper the authors study the problem of classifying unitary repre-
sentations with Dirac cohomology, focusing on the case when the group G is
a complex group viewed as a real group. They conjecture precise conditions
under which a unitary representation has nonzero Dirac cohomology and show
that it is sufficient.
2. P. Bressler, A. Gorokhovsky, R. Nest and B. Tsygan “Algebraic index the-

orem for symplectic deformations of gerbes.”
Gerbes and twisted differential operators play an increasingly important
role in global analysis. In this paper the authors continue their study of the
algebraic analogues of the algebra of twisted symbols, namely of the formal
vii
viii INTRODUCTION
deformations of gerbes. They extend the algebraic index theorem of Nest and
Tsygan to the deformations of gerbes.
3. J. Brüning, F. W. Kamber and K. Richardson “Index theory for basic Dirac
operators”
In this paper the authors consider the transverse Dirac operator on the
Riemannian foliation. It has been known for a long time that restricted to the
space of holonomy-invariant sections this operator is Fredholm, however there
was no formula for the index. Here such a formula is derived, expressing the
index in terms of the integrals of characteristic forms and η-invariants of certain
operators on the strata of the leaf closure space, provided by Molino’s theory.
4. A. Connes “The Witt construction in characteristic one and quantiza-
tion.”
This paper develops the analogue of the Witt construction in characteristic
one. The author constructs a functor from pairs (R, ρ) of a perfect semi-ring
R of characteristic one and an element ρ > 1 of R to real Banach algebras.
Remarkably the entropy function occurs uniquely as the analogue of the Te-
ichmüller polynomials in characteristic one. This construction is then applied
to the semi-field which plays a central role in idempotent analysis and tropical
geometry. It gives the inverse process of the “dequantization” and provides a
first hint towards an extension of the field of real numbers relevant both in
number theory and quantum physics.
5. M. Dubois-Violette and G. Landi “Lie prealgebras.”
This paper studies a generalization of Lie algebras based on the theory of
non-homogeneous quadratic algebras. It combines a review of existing theory of
quadratic homogeneous algebras with a proposal for a similar theory for the non-
homogeneous case. It describes in detail examples of the universal enveloping
algebras and the Lie-type algebra associated to the 3D calculus on a twisted or
quantum SU (2) group.
6. N. Higson “On the analogy between complex semisimple groups and
their Cartan motion groups.”
This paper provides a detailed exposition of how to make Mackey’s analogy
between the representations of a complex semisimple Lie group and its Cartan
motion group more precise, using representations of Hecke algebras. The author
takes the algebraic approach and obtains a Mackey-type bijection between the
admissible dual of a complex semisimple group and that of its motion group.
7. A. Kaygun “A survey on Hopf-cyclic cohomology and Connes-Moscovici
characteristic map.”
This paper reviews developments of Hopf cyclic cohomology and Connes-
Moscovici characteristic map. The author covers almost all topics related to the
Hopf cyclic cohomology. He starts with the origins of Hopf cyclic cohomology
and covers the further developments including variants of the theory and cup
products.
8. M. Khalkhali and A. Pourkia “A super version of Connes-Moscovici Hopf
algebra.”
In this paper the authors construct an analogue of Connes-Moscovici Hopf
algebra associated with the supergroup of diffeomorphisms of a superline R1,1 .
INTRODUCTION ix
They show that, similarly to the Connes-Moscovici Hopf algebra, this super
Hopf algebra can be realized as a bicrossed product.
9. V. Mathai and S. Wu “Analytic torsion of Z2 -graded elliptic complexes.”
This paper provides a construction of the analytic torsion for arbitrary Z2 -
graded elliptic complexes. It extends the construction of the analytic torsion for
twisted de Rham complex described in the previous work of the authors. As an
illustration an analytic torsion of twisted Dolbeault complexes is defined. The
authors also discuss applications of their results to topological field theories.
10. B. Monthubert and V. Nistor “The K-groups and the index theory of
certain comparison C ∗ -algebras.”
In this paper the authors consider a comparison algebra for a complete
Riemannian manifold which is the interior of a manifold with corners. This
is an algebra generated by the 0-order operators which are compositions of
differential operators with the inverse powers of the Laplacian. Relating this
algebra to the algebra of pseudodifferential operators on a suitable groupoid,
they perform calculations of the K-theory of this algebra and apply it to the
index problems.
11. H. Moriyoshi and P. Piazza “Relative pairings and the Atiyah-Patodi-
Singer index formula for the Godbillon-Vey cocycle.”
This paper outlines the authors’ approach to the extension of the Moriyoshi-
Natsume explicit formula for the pairing between the Godbillon-Vey cyclic co-
homology class and the K-theory index class of the longitudinal Dirac operator
to foliated bundles on the manifolds with boundary.
12. A. Némethi “Two exact sequences for lattice cohomology.”
In the earlier work the author introduced lattice cohomology with the goal
of providing a combinatorial description of the Heegaard–Floer homology of
Ozsváth and Szabó for the links of normal surface singularities. This has been
accomplished for several classes of examples but remains a conjecture in gen-
eral. In the meantime, the lattice cohomology has become an important tool
in studying links of singularities in its own right. This paper develops further
properties of the lattice cohomology.
13. B. Rangipour “Cup products in Hopf cyclic cohomology with coeffi-
cients in contramodules.”
This paper gives a refinement of the cup product construction in Hopf
cyclic cohomology, defined in the previous work of M. Khalkhali and the author.
This construction is particularly useful for study of the dependance of the cup
product on the coefficients.
14. M. Wodzicki “Algebras of p-symbols, noncommutative p-residue, and
the Brauer group”
This paper introduces the algebra of p-symbols, a characteristic p ana-
logue of the algebra of pseudodifferential symbols. The author shows that these
algebras have some remarkable properties and gives a construction of the non-
commutative residue for these algebras. Using elementary but subtle means the
author obtains deep and interesting results.
x INTRODUCTION
15. G. Yu “Large scale geometry and its applications.”

The ideas of large scale geometry played an important role in geometry,
topology and group theory beginning with the works of Mostow, Margulis and
Gromov. Recently there have been a number of important developments in
this area. This article surveys these developments and their applications to
geometry and topology of manifolds.
Acknowledgments. First, we would like to thank all people and institutions who
helped us to organize this conference. We received generous financial and logistical
support from the Hausdorff Center for Mathematics in Bonn. Its staff, especially
Anke Thiedemann and Laura Siklossy, provided an extensive help with the organi-
zation, and Heike Bacher and Kerstin Strehl-Müller handled the registration. The
assistants Batu Güneysu, Benjamin Himpel, Carolina Neira Jiménez, and Boris
Vertman helped with the lecture room technology.
We would like to warmly thank Arthur Greenspoon for copy-editing the papers
for these proceedings, and Dr. John R. Copeland for providing the photograph of
H Moscovici. We also express our thanks to Christine Thivierge, Boris Vertman
and Sam White for their assistance in preparing this volume.
Alain Connes
Matthias Lesch
Markus J. Pflaum
Bahram Rangipour
Conference Talks
This section lists all the talks at the conference together with the speakers’
abstracts.
Dirac Cohomology and unipotent representations
Dan Barbasch
In this talk we study the problem of classifying unitary representations with
Dirac cohomology. We will focus on the case when the group G is a complex reduc-
tive group viewed as a real group. It will easily follow that a necessary condition
for having nonzero Dirac cohomology is that twice the infinitesimal character is
regular and integral. The main conjecture is the following.
Conjecture: Let G be a complex reductive Lie group viewed as a real group,
and π be an irreducible unitary representation such that twice the infinitesimal
character of π is regular and integral. Then π has nonzero Dirac cohomology if and
only if π is cohomologically induced from an essentially unipotent representation
with nonzero Dirac cohomology. Here by an essentially unipotent representation
we mean a unipotent representation tensored with a unitary character. This work
is joint with Pavle Pandzic.
Equivariant K-homology
Paul Frank Baum
K-homology is the dual theory to K-theory. There are two points of view on K-
homology: BD (Baum-Douglas) and Atiyah-Kasparov. The BD approach defines
K-homology via geometric cycles. The resulting theory in a certain sense is simpler
and more direct than classical homology. For example, K-homology and K-theory
are made into equivariant theories in an utterly immediate and canonical way. For
classical (co)homology, there is an ambiguity about what is the “correct” definition
of equivariant (co)homology. In the case of twisted K-homology, the cycles of the
BD theory are the D-branes of string theory. This talk will give the definition of
equivariant BD theory and its extension to a bivariant theory. An application to
the BC (Baum-Connes) conjecture will be explained. The above is joint work with
N. Higson, H. Oyono-Oyono and T. Schick.
Moduli spaces of vector bundles on non-Kähler Calabi-Yau type 3-folds
Vasile Brinzanescu
We compute the relative Jacobian of a principal elliptic bundle as a coarse
moduli space and find out that it is the product of the fiber with the basis. Using
the relative Jacobian we adapt the construction of Caldararu to our case obtaining
a twisted Fourier-Mukai transform. Using this transform and the spectral cover
xi
xii CONFERENCE TALKS
we prove that the moduli space of rank n, relatively semi-stable vector bundles is
corepresented by the relative Douady space of length n and relative dimension 0
subspaces of the relative Jacobian.
The signature operator on Riemannian pseudomanifolds

Jochen Brüning
We consider an oriented Riemannian manifold which can be compactified by
adjoining a smooth compact oriented Riemannian manifold, B, of codimension at
least two, such that a neighbourhood of the singular stratum is given by a family
of metric cones. We show that there is a natural self-adjoint extension for the
Dirac operator on smooth compactly supported differential forms with discrete
spectrum, and we determine the condition of essential self-adjointness. We describe
the boundary conditions analytically and construct a good parametrix which leads
to the asymptotic expansion of the associated heat trace. We also give a new proof
of the local formula for the L2 -signature.
The lost Riemann-Roch index problem

Alain Connes
I will describe recent results of joint work with C. Consani. We determined the
real counting function N (q), (q ∈ (1, ∞)) for the hypothetical ”curve” C = Spec Z
over F1 , whose corresponding zeta function is the complete Riemann zeta function.
Then, we develop a theory of functorial F1 -schemes which reconciles the previous
attempts by C. Soulé and A. Deitmar. Our construction fits with the geometry
of monoids of K. Kato, is no longer limited to toric varieties and covers the case
of Chevalley groups. Finally we show, using the monoid of adèle classes over an
arbitrary global field, how to apply our functorial theory of Mo-schemes to interpret
conceptually the spectral realization of zeros of L-functions. I will end the lecture
by a speculation concerning a Riemann-Roch index problem which is so far lost in
a translation.
C ∗ -algebras associated with integral domains

Joachim Cuntz
We associate canonically a C ∗ -algebra with every (countable) integral domain.
We describe different realizations of this algebra which are used to analyze its
structure and K-theory.
The Baum-Connes conjecture and parametrization of group
representations
Nigel Higson
Associated to any connected Lie group G is its so-called contraction of G to a
maximal compact subgroup. This is a smooth family of Lie groups, and a conse-
quence of the Baum-Connes conjecture is that the reduced duals of all the groups
in the family are the same, at least at the level of K-theory. A rather surprising
development from the last several years is that in key cases the duals are actually
the same at the level of sets. I shall report on recent efforts, both algebraic and
geometric, to understand this phenomenon better.
CONFERENCE TALKS xiii
Hopf-cyclic cohomology and Connes-Moscovici characteristic map

Atabey Kaygun
In 1998 Alain Connes and Henri Moscovici invented a cohomology theory for
Hopf algebras and a characteristic map associated with the cohomology theory
in order to solve a specific technical problem in transverse index theory. In the
following decade, the cohomology theory they invented developed on its own under
the name Hopf-cyclic cohomology. But the history of Hopf-cyclic cohomology and
the characteristic map they invented remained intricately linked. In this survey
talk, I will give an account of the development of the characteristic map and Hopf-
cyclic cohomology.
Holomorphic structures on the quantum projective line
Masoud Khalkhali
In this talk I report on our joint ongoing work with Giovanni Landi and Walter
van Suijlekom. We define a notion of holomorphic structure in terms of a bigrading
of a suitable differential calculus over the quantum sphere. Realizing the quantum
sphere as a principal homogeneous space of the quantum group SUq (2) plays an
important role in our approach. We define a notion of holomorphic vector bundle
and endow the canonical line bundles over the quantum sphere with a holomor-
phic structure. We also define the quantum homogeneous coordinate ring of the
projective line CPq1 and identify it with the coordinate ring of the quantum plane.
Finally I shall formulate an analogue of Connes’ theorem, characterizing holomor-
phic structures on compact oriented surfaces in terms of positive currents, to our
noncommutative context. The notion of twisted positive Hochschild cocycles plays
an important role here.
Monopoles connections on the quantum projective plane
Giovanni Landi
We present several results on the geometry of the quantum projective plane.
They include: explicit generators for the K-theory and the K-homology; a real
calculus with a Hodge star operator; anti-selfdual connections on line bundles with
explicit computation of the corresponding invariants; quantum invariants via equi-
variant K-theory and q-indices; and more.
An index theorem in differential K-theory
John Lott
Differential K-theory is a refinement of the usual K-theory of a manifold. Its
objects consist of a vector bundle with a Hermitian inner product, a compatible
connection and an auxiliary differential form. Given a fiber bundle with a Riemann-
ian structure on its fibers, and a differential K-theory class on the total space, I
will define two differential K-theory classes on the base. These can be considered
to be topological and analytic indices. The main result is that they are the same.
This is joint work with Dan Freed.
xiv CONFERENCE TALKS
Noncommutative geometry and cosmology: a progress report

Matilde Marcolli
I will describe some ongoing work in collaboration with Elena Pierpaoli (Astron-
omy, USC/Caltech) on cosmological implications of the noncommutative geometry
approach to the standard model with neutrino mixing coupled to gravity previously
developed in joint work with Chamseddine and Connes. In particular, I will show
how applying the renormalization group flow for the standard model with Majorana
mass terms for right handed neutrinos to the asymptotic formula for the spectral
action one recovers naturally a wide range of theoretical cosmological models of the
inflationary epoch in the early universe.
Connes-Chern character and higher eta cocycles
Henri Moscovici
An intriguing avatar of Connes’ Chern character in K-homology assumes the
form of a higher eta cocycle. After recounting some previous occurrences of these
cocycles, in work of Connes and myself, and also of F. Wu and Getzler, we shall
explain how such higher eta cochains and their b-trace analogues can be assembled
together to produce representations in relative cyclic cohomology for the Connes-
Chern character of a Dirac operator on a manifold with boundary. The latter result
is joint work with M. Lesch and M. Pflaum.
Dynamical zeta functions and analytic torsion of hyperbolic manifolds
Werner Müller
In this talk we discuss the relation between the Ruelle zeta function and the
analytic torsion of a hyperbolic manifold. In particular, we derive an asymptotic
formula for the analytic torsion of a hyperbolic 3-manifold with respect to a special
sequence of representations of the fundamental group. We apply this formula to
study the torsion of the cohomology of arithmetic hyperbolic 3-manifolds.
Lattice (co)homology associated with plumbed 3-manifolds
András Némethi
For any negative definite plumbed 3-manifold M we construct from its plumbed
graph a graded Z[U ]-module. This, for rational homology spheres, conjecturally
equals the Heegaard-Floer homology of Ozsváth and Szabó, but it has even more
structure. In particular, it shares several properties of the Heegaard-Floer homol-
ogy, for example its normalized Euler-characteristic is the Seiberg-Witten invariant.
If M is a complex surface singularity link then the new invariant can be compared
with those coming from the analytic geometry. The talk will emphasize some of
the intriguing connections of the analytic and topological invariants in the light of
this new object. For example, we get new characterizations of rational and elliptic
singularities, a new guiding principle for classification, or description of geometric
genus (and other sheaf cohomologies) in terms of Seiberg-Witten invariant (subject
of the Seiberg-Witten Invariant Conjecture of Nicolaescu and the author). We list
some further open problems and conjectures as well.
CONFERENCE TALKS xv
A topological index theorem for manifolds with corners

Victor Nistor
We define a topological and an analytical index for manifolds with corners
M . They both live in the K-theory groups K0 (Cb∗ (M )) of the groupoid algebra
associated to our manifold with corners M by integrating the Lie algebra of vector
fields tangent to all faces of M . We prove that the topological and analytic index
coincide. For M smooth (no corners), this is the Atiyah-Singer index theorem. If
all the open faces of M are euclidean spaces, then the index maps are isomophisms,
which gives a way of computing the K-theory of the groups K∗ (Cb∗ (M )). The
proof uses a double-deformation groupoid obtained by integrating a suitable Lie
algebroid. This is a joint work with Bertrand Monthubert.
The signature package on Witt spaces
Paolo Piazza
Let X be an orientable closed compact riemannian manifold with fundamental
group G. Let X be a Galois G-covering and r: X → BG a classifying map for X .
The signature package for (X, r : X → BG) can be informally stated as follows:
• there is a signature index class in the K-theory of the reduced C ∗ -algebra
of G
• the signature index class is a bordism invariant
• the signature index class is equal to the C ∗ -algebraic Mishchenko signa-
ture, also a bordism invariant which is, in addition, a homotopy invariant
• there is a K-homology signature class in K ∗ (X) whose Chern character
is, rationally, the Poincare’ dual of the L-Class
• if the assembly map in K-theory is rationally injective one deduces from
the above results the homotopy invariance of Novikov higher signatures
The goal of my talk is to discuss the signature package on a class of stratified
psedomanifolds known as Witt spaces. The topological objects involve intersection
homology and Siegel’s Witt bordism groups. The analytic objects involve some
delicate elliptic theory on the regular part of the stratified pseudomanifold. Our
analytic results reestablish (with completely different techniques) and extend results
of Jeff Cheeger. This is joint work, some still in progress, with Pierre Albin, Eric
Leichtnam and Rafe Mazzeo.
Group measure space decomposition of factors and W ∗ -superrigidity
Sorin Popa
A free ergodic measure preserving action of a countable group on a probability
space, Γ X, gives rise to a II1 factor, L∞ (X) Γ, through the group measure
space construction of Murray and von Neumann. In general, much of the initial
data Γ X is “forgotten” by the isomorphism class of L∞ (X) Γ, for instance all
free ergodic probability measure preserving actions of amenable groups give rise to
isomorphic II1 factors (Connes 1975). But a rich and deep rigidity theory underlies
the non-amenable case. For instance, I have shown in 2005 that any isomorphism
of II1 factors associated with Bernoulli actions Γ X, Λ Y , of Kazhdan
groups Γ, Λ comes from a conjugacy of the actions (W ∗ -strong rigidity). I will
present a recent joint work with Stefaan Vaes, in which we succeeded to prove a
W ∗ -superrigidity result for Bernoulli (+ other) actions Γ X of amalgamated
xvi CONFERENCE TALKS
free product groups Γ = Γ1 ∗Σ Γ2 , where Γ1 is Kazhdan and Σ infinite amenable

satisfying a combination of singularity/normality conditions in Γ1 , Γ2 . It shows
that any isomorphism between L∞ (X) Γ and a factor L∞ (Y ) Λ arising from an
arbitrary free ergodic action Λ Y , comes from a conjugacy of Γ X, Λ Y .
Noncommutative residues and projections associated to boundary

value problems
Elmar Schrohe
On a compact manifold X with boundary we consider the realization B =
PT of an elliptic boundary problem, consisting of a differential operator P and
a differential boundary condition T . We assume that B is parameter-elliptic in
small sectors around two rays in the complex plane, say arg λ = φ and arg λ = θ.
Associated to the cuts along the rays one can then define two zeta functions ζφ and
ζθ for B. Both extend to meromorphic functions on the plane; the origin is a regular
point. We relate the difference of the values at the origin to a noncommutative
residue for the associated spectral projection Πφ,θ (B) defined by

i
Πφ,θ u = λ−1 B(B − λ)−1 u dλ, u ∈ dom(B)
2π Γθ,φ
where Γθ,φ is the contour which runs on the first ray from infinity to r0 eiφ for some
r0 > 0, then clockwise about the origin on the circle of radius r0 to r0 eiθ and back
to infinity along the second ray.
Fine structure of special symplectic spaces

Robert J. Stanton
Riemannian symmetric spaces can be classified according to Z3 and Z5 grad-
ings of Lie algebras. The Z5 gradings give rise to special symplectic spaces. Using
symplectic methods we give a rather complete description of the orbit structure of
special symplectic spaces. Applications to realizations of special geometric struc-
tures, to Lie theory, and to new composition structures on orbits will be presented.
This is joint with M. Slupinski.
Operations on Hochschild complexes

Boris Tsygan
The subject of operations on Hochschild and cyclic complexes of algebras is
surprisingly rich and difficult. It is far from being settled after at least twenty years
of study by many authors. I will try to outline its current state, including results of
Kontsevich and Soibelman, of myself and Bressler, Nest, Dolgushev and Tamarkin,
of Costello, and of Lurie.
The index of projective families of elliptic operators

Mathai Varghese
I will talk about ongoing research with Melrose and Singer, where we recently
established an index theorem for projective families of elliptic operators. In this
context, the index takes values in a smooth version of the twisted K-theory of the
parametrizing space.
CONFERENCE TALKS xvii
Aspects of free analysis

Dan Voiculescu
Motivated by free probability questions, the lecture will deal with duality for the
bialgebra of the free difference quotient derivation and the highly noncommutative
generalization of the Riemann sphere which appears in this context.
Algebras of p-symbols, noncommutative p-residue, and the Brauer
group
Mariusz Wodzicki
Importance of the pseudodifferential symbol calculus extends far beyond the
fundamental role it is known to play in Global and Microlocal Analysis. In this
article, we demonstrate that algebras of symbols contribute to subtle phenomena
in characteristic p > 0.
Geometric complexity and topological rigidity
Guoilang Yu
In this talk, I will introduce a notion of geometric complexity and discuss its
applications to geometric group theory and rigidity of manifolds. In particular, I
will show how to prove various geometric versions of the Borel conjecture under
certain a finiteness condition on the geometric complexity. This is joint work with
Erik Guentner and Romain Tessera.
This page intentionally left blank
List of Participants
Aldana Dominguez, Clara Lucia Himpel, Benjamin

Max-Planck Institute for Mathematics Universität Bonn
at Bonn
Iochum, Bruno
Anghel, Nicolae Université de Provence
University of North Texas at Denton
Kaygun, Atabey
Azzali, Sara Universidad de Buenos Aires
Universität Göttingen Khalkhali, Masoud
Barbasch, Dan University of Western Ontario at
Cornell University London
Landi, Giovanni
Baum, Paul
Università di Trieste
Penn State University
Lesch, Matthias
Brinzanescu, Vasile Universität Bonn
Institute of Mathematics of the
Romanian Academy at Bucharest Lott, John
University of California at Berkeley
Brüning, Jochen
Humboldt-Universität zu Berlin Marcolli, Matilde
Caltech
Connes, Alain
College de France at Paris Moscovici, Henri
Ohio State University at Columbus
Cuntz, Joachim
Universität Münster Müller, Werner
Universität Bonn
Farsi, Carla
Némethi, András
University of Colorado at Boulder
University of Budapest
Fermi, Alessandro Nistor, Victor
Universität Göttingen Penn State University
Gorokhovsky, Alexander Pflaum, Markus Josef
University of Colorado at Boulder University of Colorado at Boulder
Hajac, Piotr Piazza, Paolo
Polish Academy of Sciences at Warsaw Università di Roma “La Sapienza”
Higson, Nigel Ponge, Raphael
Pennsylvania State University University of Tokyo
xix
xx PARTICIPANTS
Popa, Sorin
University of California at Los Angeles
Rangipour, Bahram
University of New Brunswick
Savin, Anton
Peoples Friendship University of Russia
at Moscow
Schrohe, Elmar
Leibniz Universität Hannover
Sharygin, Georgy
Institute for Theoretical and
Experimental Physics at Moscow
Sitarz, Andrzeij Woiciech
Jagiellonian University at Krakow
Stanton, Robert
Ohio State University at Columbus
Sternin, Boris
Peoples Friendship University of Russia
at Moscow
Tang, Xiang
Washington University
Tsygan, Boris
Northwestern University
Varghese, Mathai
University of Adelaide
Voiculescu, Dan-Virgil
Wagner, Stefan
TU Darmstadt
Wodzicki, Mariusz
Wu, Siye
University of Colorado at Boulder
Xie, Zhizhang
The Ohio State University at Columbus
Yu, Gouliang
Vanderbilt University
Contemporary Mathematics
Volume 546, 2011
Dirac cohomology and unipotent representations of complex

groups
Dan Barbasch and Pavle Pandžić
This paper is dedicated to Henri Moscovici.
Abstract. This paper studies unitary representations with Dirac cohomology

for complex groups, in particular relations to unipotent representations.
1. Introduction
In this paper we will study the problem of classifying unitary representations
with Dirac cohomology. We will focus on the case when the group G is a complex
reductive group viewed as a real group. It will easily follow that a necessary condi-
tion for having nonzero Dirac cohomology is that twice the infinitesimal character
is regular and integral. The main conjecture is the following.
Conjecture 1.1. Let G be a complex reductive Lie group viewed as a real

group, and π be an irreducible unitary representation such that twice the infinites-
imal character of π is regular and integral. Then π has nonzero Dirac cohomology
if and only if π is cohomologically induced from an essentially unipotent represen-
tation with nonzero Dirac cohomology. Here by an essentially unipotent represen-
tation we mean a unipotent representation tensored with a unitary character.
We start with some background and motivation.
Let G be the real points of a linear connected reductive group. Its Lie algebra
will be denoted by g0 . Fix a Cartan involution θ and write g0 = k0 + s0 for the
Cartan decomposition. Denote by K the maximal compact subgroup of G with Lie
algebra k0 . The complexification g := (g0 )C decomposes as g = k + s.
A representation (π, H) on a Hilbert space is called unitary if H admits a
G-invariant positive definite inner product. One of the major problems of represen-
tation theory is to classify the irreducible unitarizable modules of G. As motivation
for why this problem is important, we present an example from automorphic forms.
2010 Mathematics Subject Classification. Primary 22E47 ; Secondary 22E46.

Key words and phrases. Dirac cohomology, unipotent representations.
The first author was supported in part by NSF Grants #0967386 and #0901104.
The second author was supported in part by a grant from the Ministry of Science, Education
and Sports of the Republic of Croatia.
1 2011
c American Mathematical Society
1
2 DAN BARBASCH AND PAVLE PANDŽIĆ
Let Γ ⊂ G be a discrete cocompact subgroup. A question of interest is the compu-

tation of H ∗ (Γ). Let X := Γ\G/K. Then H i (Γ) := Htop
i
(X, C), where Htop
i
(X, C)
denotes the usual cohomology of the topological space X: The theory of automor-
i
phic forms provides insight into Htop (X, C). A fundamental result of Gelfand and
Piatetski-Shapiro is that
L2 (Γ\G) = mπ Hπ
where π are irreducible unitary representations of G, and mπ < ∞. It implies that

H i (Γ) = Htop
i
(X, C) = i
mπ Hct (G, Hπ ) = mπ H i (g, K; Hπ ).
Here Hcti
(G, Hπ ) denotes the continuous cohomology groups (see [BW]), and the
groups H i (g, K; Hπ ) are the relative Lie algebra cohomology groups defined in
[BW], Chapter II, Section 6, or [VZ]. Here the unitary representation Hπ is
replaced by the corresponding (g, K)-module, denoted again by Hπ .
Thus to obtain information about H i (Γ) one needs to have information about
mπ and H i (g, K, π). It is very difficult to obtain information about the multiplic-
ities mπ . On the other hand, knowledge about the vanishing of H i (g, K) for all
unitary representations translates into vanishing of H i (Γ). This approach leads one
to consider the following problem.
Problem. Classify all irreducible admissible unitary modules with nonzero
(g, K) cohomology.
A more general problem where the trivial representation C of Γ is replaced by

an arbitrary finite dimensional representation was solved by Enright [E] for complex
groups. Introducing more general coefficients has the effect that Hπ is replaced by
Hπ ⊗ F ∗ for some finite-dimensional representation F . The results were generalized
later by Vogan-Zuckermann [VZ] to real groups as follows. The λ appearing below
is such that the infinitesimal character of Rsq (Cλ ) equals the infinitesimal character
of F . The answer is that π = Rsq (Cλ ), where
- q = l + u ⊂ g is a θ stable parabolic subalgebra,
- Cλ is a unitary character of l,
- Rsq is cohomological induction, and s = dim(u ∩ k).
i
The starting point for the proof is the fact H i (g, K; π) ∼ = HomK [ s, π]. The
reference [BW] gives consequences of these results. For a survey of related more
recent results, the reader may consult [LS].
A major role in providing an answer to the above problem is played by the

Dirac Inequality of Parthasarathy [P2]. The adjoint representation of K on s lifts
−→ Spin(s), where K
to Ad : K is the spin double cover of K. The Dirac operator
D : Hπ ⊗ Spin −→ Hπ ⊗ Spin
is defined as
D= bi ⊗ di ∈ U (g) ⊗ C(s),
i
where C(s) denotes the Clifford algebra of s with respect to the Killing form, bi is a
basis of s and di is the dual basis with respect to the Killing form, and Spin is a spin
module for C(s). D is independent of the choice of the basis bi and K-invariant. It
satisfies
D2 = −(Casg ⊗1 + ρg 2 ) + (Δ(Cask ) + ρk 2 ).
DIRAC COHOMOLOGY FOR COMPLEX GROUPS 3
In this formula, due to Parthasarathy [P1],

- Casg and Cask are the Casimir operators for g and k respectively,
- h = t + a is a fundamental θ-stable Cartan subalgebra with compatible
systems of positive roots for (g, h) and (k, t),
- ρg and ρk are the corresponding half sums of positive roots,
- Δ : k → U (g) ⊗ C(s) is given by Δ(X) = X ⊗ 1 + 1 ⊗ α(X), where α
is the action map k → so(s) followed by the usual identifications so(s) ∼
=
2
(s) → C(s).
If π is unitary, then Hπ ⊗ Spin admits a K-invariant inner product
, such that
D is self-adjoint with respect to this inner product. It follows that D2 ≥ 0 on
Hπ ⊗ Spin. Using the above formula for D2 , we find that
Casg +||ρg ||2 ≤ CasΔ(k) +||ρk ||2
on any K-type τ occurring in Hπ ⊗ Spin. Another way of putting this is
(1.1) ||χ||2 ≤ ||τ + ρk ||2 ,
for any τ occurring in Hπ ⊗ Spin, where χ is the infinitesimal character of π. This
is the Dirac inequality mentioned above.
These ideas are generalized by Vogan [V2] and Huang-Pandžić [HP1] as fol-
lows. For an arbitrary admissible (g, K)-module π, we define the Dirac cohomology
of π as
HD (π) = ker D/(ker D ∩ im D).
Then HD (π) is a module for K. If π is unitary, HD (π) = ker D = ker D2 .
The main result about HD is the following theorem conjectured by Vogan.
Theorem 1.2. [HP1] Assume that HD (π) is nonzero, and contains an ir-

reducible K-module with highest weight τ . Let χ ∈ h∗ denote the infinitesimal
character of π. Then wχ = τ + ρk for some w in the Weyl group W = W (g, h).
More precisely, there is w ∈ W such that wχ |a = 0 and wχ |t = τ + ρk .

Conversely, if π is unitary and τ = wχ − ρk is the highest weight of a K-type

occurring in π ⊗ Spin, then this K-type is contained in HD (π).
This result might suggest that difficulties should arise in passing between K-

types of π and K-types of π ⊗ Spin. For unitary π, the situation is however greatly
simplified by the Dirac inequality. Namely, together with (1.1), Theorem 1.2 shows

that the infinitesimal characters τ +ρk of K-types in Dirac cohomology have minimal
possible norm. This means that whenever such E(τ ) appears in the tensor product

of a K-type E(μ) of π and a K-type E(σ) of Spin, it necessarily appears as the
PRV component [PRV], i.e.,
(1.2) τ = μ + σ− up to Wk ,
where σ − denotes the lowest weight of E(σ).
For unitary representations, the relation of Dirac cohomology to (g, K) coho-
mology is as
follows. (For more details, see [HP1] and [HKP].) One can write the
K-module (s) as Spin ⊗ Spin if dim s is even, or the direct sum of two copies of
the same space if dim s is odd. It follows that

HomK ( (s), π ⊗ F ∗ ) = HomK (F ⊗ Spin, π ⊗ Spin),
or the direct sum of two copies of the same space if dim s is odd. Since D2 ≥ 0 on
π ⊗ Spin and D2 ≤ 0 on F ⊗ Spin, it follows that
H(g, K; π ⊗ F ∗ ) = HomK (HD (F ), HD (π)),
or the direct sum of two copies of the same space if dim s is odd. In particular, if
π is unitary and it has nontrivial (g, K) cohomology, then HD (π) = 0.
For a representation to have nonzero (g, K) cohomology with coefficients in a
finite dimensional representation, the infinitesimal character must be regular in-
tegral. Conversely, assume that π is unitary with regular integral infinitesimal
character. Then the main result of [SR] implies that π is an Aq (λ)-module, and
therefore it has nonzero (g, K) cohomology by the results of [VZ]. (Hence it also
has nonzero Dirac cohomology, as explained above.)
The hope is that unitary representations with Dirac cohomology will have sim-
ilarly nice properties. For HD (π) to be nonzero, Theorem 1.2 provides a restriction
on the infinitesimal character χπ which is weaker than regular integral. Namely,
because χπ |t must be conjugate to τ + ρc , it must be regular integral for the roots
in k. Thus one expects to have representations with nonzero Dirac cohomology with
infinitesimal character that is not regular integral. Indeed, we will describe many
such examples in this paper. On the other hand, the conditions of regularity and
integrality with respect to k are still quite restrictive and we cannot expect such
representations to capture the entire unitary dual. The relatively few unitary rep-
resentations that have nonzero Dirac cohomology are however the borderline cases
for unitarity in the sense of the Dirac inequality.
The paper is organized as follows. In Section 2 we first recall some well known
facts about complex groups and their representations. Then we prove one of the
main results of the paper, which says that if a representation π is unitarily induced
from a representation with nonzero Dirac cohomology, then π must have nonzero
Dirac cohomology, provided twice the infinitesimal character of π is regular and
integral. In Section 3 we strengthen this result by actually calculating HD for rep-
resentations induced from unitary characters whose infinitesimal character is ρ/2.
In Section 4 we generalize this result to GL(n, C) and more general infinitesimal
characters (we do not prove the full conjecture). Finally in Section 5 we discuss
unipotent representations with non-vanishing Dirac cohomology. In summary, the
main general results are Theorem 2.5 and Theorem 3.3. The other results of the
paper provide evidence for conjectures 1.1, 3.4, and 4.1. We plan to investigate the
validity of these conjectures in future papers.
We dedicate this paper to Henri Moscovici. Henri introduced the first author
to the beautiful theory of the heat kernel and index theory on semisimple groups.
2. Complex groups
2.1. General setting. Let G be a complex reductive group viewed as a real
group. Let K be a maximal compact subgroup of G. Let Θ be the corresponding
Cartan involution, and let g0 = k0 + s0 be the corresponding Cartan decomposition
of the Lie algebra g0 of G. Let H = T A be a θ-stable Cartan subgroup of G, with
Lie algebra h0 = t0 +a0 , a θ-stable Cartan subalgebra of g0 . We assume that t0 ⊆ k0
and a0 ⊆ s0 .
Let B = HN be a Borel subgroup of G. We identify the complexification

h∼= h0 × h0 of h0 with the complexifications
(2.1) t∼= {(x, −x) : x ∈ h0 }, a∼
= {(x, x) : x ∈ h0 }
of t0 , respectively a0 .
Admissible irreducible representations of G are parametrized by conjugacy
classes of pairs (λL , λR ) ∈ h0 × h0 under the diagonal Δ(W ) ⊂ W × W . More
precisely, the following theorem holds.
Theorem 2.1 ([Zh], [BV]). Let (λL , λR ) ∈ h0 ×h0 be such that μ := λL −λR is
integral. Write ν := λL + λR . We can view μ as a weight of T and ν as a character
of A. Let
B [Cμ ⊗ Cν ⊗ 11]K−finite .
X(λL , λR ) := IndG
Then the K-type with extremal weight μ occurs in X(λL , λR ) with multiplicity 1.
Let L(λL , λR ) be the unique irreducible subquotient containing this K-type. Then:
(1) Every irreducible admissible (g, K) module is of the form L(λL , λR ).
(2) Two such modules L(λL , λR ) and L(λL , λR ) are equivalent if and only if
the parameters are conjugate by Δ(W ) ⊂ Wc ∼ = W × W. In other words,
there is w ∈ W such that wμ = μ and wν = ν .
(3) L(λL , λR ) admits a nondegenerate Hermitian form if and only if there is
w ∈ W such that wμ = μ, wν = −ν.
This result is a special case of the more general Langlands classification, which
can be found for example in [Kn], Theorem 8.54.
We next describe the spin representation of the group K. We denote by ρ the
half-sum of roots in Δ(b, h). Let r be the rank of g.

Lemma 2.2. The spinor representation Spin viewed as a K-module is a direct
r
sum of [ 2 ] copies of the irreducible representation E(ρ) of K with highest weight
ρ.
Proof. The general description of the spin module is given for example in
[BW], Lemma 6.9. It says that the irreducible components of Spin correspond to
choices of positive roots for g compatible with a fixed choice of positive roots for k.
The multiplicity for each component is [ dim a
2 ]. In the complex case, there is only
one such choice of positive roots, and dim a is r.
Lemma 2.2 implies that in calculating HD (π) for unitary π, one can replace
Spin by E(ρ) and then in the end simply multiply the result by multiplicity [ r2 ].
By Theorem 1.2 and the above remark, a unitary representation L(λL , λR ) has
nonzero Dirac cohomology if and only if there is (w1 , w2 ) ∈ Wc such that
(2.2) w1 λL + w2 λR = 0, w1 λL − w2 λR = τ + ρ

where τ is the highest weight of a K-type which occurs in L(λL , λR ) ⊗ E(ρ). More
precisely,
r
(2.3) [HD (π) : E(τ )] = [ ] [π : E(μ)] [E(μ) ⊗ E(ρ) : E(τ )],
μ
2
where the sum is over all K-types E(μ) of π.

Write λ := λL . The first equation in (2.2) implies that λR = −w2−1 w1 λ. The
second one says that 2w1 λ = τ + ρ, so that w1 λ must be regular, and 2w1 λ regular
integral. Replace w1 λ by λ. Thus we can write the parameter of π as (λ, −sλ) with
λ dominant, and s ∈ W. Since L(λ, −sλ) is assumed unitary, it is Hermitian. So
there is w ∈ W such that
(2.4) w(λ + sλ) = λ + sλ, w(λ − sλ) = −λ + sλ.
This implies that wλ = sλ, so w = s since λ is regular, and wsλ = s2 λ = λ. So s
must be an involution.
Thus to compute HD (π) for π that are unitary, we need to know
(1) L(λ, −sλ) that are unitary with
(2.5) 2λ = τ + ρ,
in particular 2λ is regular integral,
(2) the multiplicity

(2.6) L(λ, −sλ) ⊗ E(ρ) : E(τ ) .
2.2. Unitarily induced representations. We consider the Dirac cohomol-
ogy of a representation π which is unitarily induced from a unitary representation
of the Levi component M of a parabolic subgroup P = M N .
We write π := IndG P [Cξ ⊗ πm ], where ξ is a unitary character of M, and πm
is a unitary representation of M such that the center of M acts trivially. It is
straightforward that πm has Dirac cohomology if and only if Cξ ⊗ πm has Dirac
cohomology.
The representation πm = Lm (λm , −sλm ) satisfies
(2.7) λm + sλm = μm , 2λm = μm + νm ,
(2.8) λm − sλm = νm , 2sλm = μm − νm ,
with s ∈ Wm . Assume that πm has Dirac cohomology. So
(2.9) 2λm = μm + νm = τm + ρm
is regular integral and dominant for a positive system Δm . Here τm is dominant
with respect to Δm , and ρm is the half sum of the roots in Δm . Also,

(2.10) πm ⊗ F (ρm ) : F (τm ) = 0.
Notation 2.3. For a dominant m-weight χ, we denote by F (χ) the finite-
dimensional m-module with highest weight χ. For a dominant g-weight η, we denote
by E(η) the finite-dimensional g-module with highest weight η. We are also going
to use analogous notation when χ and η are not necessarily dominant, but any
extremal weights of the corresponding modules.
The lowest K-type subquotient of π is L(λ, −sλ). It has parameters
λ = ξ/2 + λm , μ = ξ + μm ,
(2.11)
sλ = ξ/2 + sλm , ν = νm .
We assume that ξ is dominant for Δ(n) the roots of N. This is justified in view of
the results in [V1] and [B] which say that any unitary representation is unitarily
induced irreducible from a representation πm on a Levi component with these prop-
erties. In order to have Dirac cohomology, 2λ must be regular integral; so assume
this is the case. Let Δ be the positive system such that λ is dominant. Then
(2.12) 2λ = ξ + μm + νm = τ + ρ .
Here ρ is the half sum of the roots in Δ , and τ is dominant with respect to Δ .
In order to see that π has nonzero Dirac cohomology, we need the following lemma.
∗ 2.4. The restriction of the g-module E(ρ) to m is isomorphic to F (ρm )⊗

Lemma
C−ρn ⊗ n, where F (ρm ) denotes the irreducible m-module with highest weight
ρm and ρn denotes the half sum of roots in Δ(n).
Proof. Since g and m have the same rank, we can use Lemma 2.2 to replace
E(ρ) and F (ρm ) by the corresponding
spin modules. Recall that the spin module
Spinm can be constructed as ∗ m+ , where m+ is a maximal isotropic subspace of
m. We can choose m+ so that it contains all the positive root subspaces for m, as
well as a maximal isotropic subspace h+ of the Cartan subalgebra h. To construct
∗ isotropic subspace g = m ⊕ n of g. It follows that
+ +
Sping , we can use the maximal
Sping = Spinm ⊗ C−ρn ⊗ n. The ρ-shift comes from the fact that the highest
weight of Spin
m is ρ m and the highest weight of Sping is ρ, while the highest weight
of Spinm ⊗ ∗ n is ρm + 2ρn = ρ + ρn .
Since π is unitary, the computation for its Dirac cohomology is

π ⊗ E(ρ) : E(τ ) = πm ⊗ Cξ ⊗ E(−τ ) |m : E(ρ) |m =
∗
(2.13) πm ⊗ Cξ ⊗ E(−τ ) |m : F (ρm ) ⊗ C−ρn ⊗ n =
∗
Cξ+ρn ⊗ πm ⊗ F (ρm ) ⊗ E(−τ ) |m : n .
Here the first equality used Frobenius reciprocity, while the second equality used
Lemma 2.4. Note that the dual of E(τ ) is the module E(−τ ) which has lowest
weight −τ with respect to Δ .
Using (2.12) and (2.9), we can write
(2.14) −τ = −2λ + ρ = −ξ − μm − νm + ρ = −ξ − τm − ρm + ρ .
We have assumed ξ to be dominant for Δ(n), and 2λm is dominant for Δm . Thus
Δm ⊂ Δ, Δ . Because of (2.10), the LHS of the last line of (2.13) contains the
representation
Cξ+ρn ⊗ F (τm ) ⊗ E(−τ ) |m ⊇ Cξ+ρn ⊗ F (τm − τ ).
Namely, F (τm −τ ) is the PRV component of F (τm )⊗F (−τ ) ⊆ F (τm )⊗E(−τ ) |m .
By (2.12) and (2.9), τm − τ = −ξ − ρm + ρ , so
Cξ+ρn ⊗ F (τm − τ ) ⊇ F (ρn − ρm + ρ ) = F (wm ρ + ρ ),
where wm is the longest element of the Weyl group of m. Namely, wm sends all
roots in Δm to negative roots for m, while permuting the roots in Δ(n), so wm ρ =
−ρm + ρn .
So we see that the LHS of the last line of (2.13) contains the m-module F (wm ρ+
ρ ) = F (wm ρ +ρ). Namely, both wm ρ+ρ and wm ρ +ρ = wm (wm ρ+ρ ) are extremal
weights for the same module.
We will show that
∗
(2.15) F (wm ρ + ρ) : n = 0.
This will prove that (2.13) is nonzero, and consequently that π has nonzero Dirac
cohomology.
Note that wm ρ +ρ is a sum of roots in Δ(n), and antidominant for Δm , because

for any simple γ ∈ Δm ,
ρ , γ̌ ∈ N+ and
ρ, γ̌ = 1. Moreover,

(2.16) w m ρ + ρ = α.
α,wm ρ >0, α,ρ>0
To show that (2.15) holds, it is enough to show that

∗
(2.17) v := eα ∈ n
α,ρ>0, α,wm ρ >0
is a lowest weight vector for Δm . Here eα denotes a root vector for the root α.
Let γ ∈ Δm . Then, up to constant factors,

0 if α − γ is not a root,
(2.18) ad e−γ eα =
e−γ+α if α − γ is a root.
But
−γ, wm ρ > 0, and
α, wm ρ > 0 by assumption, so
(2.19)
−γ + α, wm ρ > 0 + 0 = 0.
Also, if −γ + α is a root, then it is in Δ(n), since α ∈ Δ(n) and n is an m-module.
So
−γ + α, ρ > 0. Thus every e−γ+α appearing in (2.18) is one of the factors in
(2.17).
The claim now follows from the formula

(2.20) ad e−γ eα = eα1 ∧ · · · ∧ ad e−γ eαi ∧ · · · .
In each summand either ad e−γ eαi equals 0, or is a multiple of one of the root
vectors already occurring in the same summand. So ad e−γ v = 0. We have proved
the following theorem.
Theorem 2.5. Let P = M N be a parabolic subalgebra of G and let Δ =
Δm ∪ Δ(n) be the corresponding system of positive roots. Let πm := Lm (λ, −sλ) be
an irreducible unitary representation of M with nonzero Dirac cohomology such that
its parameter is zero on the center of m. Let ξ be a unitary character of M which
is dominant with respect to Δ. Suppose that twice the infinitesimal character of
π = IndGP [πm ⊗ξ] is regular and integral. Then π has nonzero Dirac cohomology.
This theorem proves one direction of Conjecture 1.1, i.e., that the condition
of the conjecture is sufficient for the non-vanishing of Dirac cohomology. We do
not know how to prove the other direction, but we believe it to be true because of
examples.
Example 2.6. Let g := sp(10), and take infinitesimal character ρ/2, which is
conjugate to
(2, 1, 5/2, 3/2, 1/2).
According to [B], the spherical representation is not unitary. But the parameter
(2.21) (2, 1; 1/2, 5/2, 3/2) × (−1, −2; −1/2, 5/2, 3/2),
which has μ = (3, 3, 1, 0, 0) and ν = (1, −1; 0, 5, 3), is unitary because it is unitarily
induced from a representation on GL(2) × Sp(6) which is the trivial representation
on GL(2) and the nonspherical component of the metaplectic representation on
Sp(6) (see below). It is not unitarily induced from a unitary character (the coor-
dinates of ν corresponding to the 0 s in the coordinates of μ are 5, 3; they would
have had to have been 4, 2). Write
m = m1 × m2 = gl(2) × sp(6).
The parameter of the metaplectic representation is
(2.22) λm2 = (1/2, 5/2, 3/2), −sλm2 = (−1/2, 5/2, 3/2)
with μm2 = (1, 0, 0) and νm2 = (0, 5, 3). Its K-structure is (1 + 2k, 0, 0). The
Spin
representation
is a multiple of E(ρ) = E(5, 4, 3, 2, 1). So the multiplicity
L(λ, −sλ) : E(ρ) has a chance to be nonzero since the sums of coordinates in
μ and ρ have the same parity. Then

τm 2
= 2λm2 − ρm2 = (1, 5, 3) − (1, 3, 2) = (0, 2, 1).
The character of m1 = gl(2) is (3, 3), and we can view it as the character
ξ = (3, 3, 0, 0, 0) of m.
Let us change the parameter in (2.22) to
(2.23) λm2 = (5/2, 3, 2, 1/2), −sλm2 = (5/2, 3/2, −1/2)
so that Δm2 becomes the usual positive system. This changes ρ = (4, 2, 1, 5, 3)
into (4, 2, 5, 3, 1). One easily checksthat ρn = (9/2, 9/2, 0, 0, 0). Thus the last line
of (2.13), that is, the multiplicity π ⊗ E(ρ) : E(τ ) that determines the Dirac
cohomology, equals the multiplicity
∗
(2.24) C(15/2,15/2,0,0,0) ⊗ F (0, 0, 1 + 2k, 0, 0) ⊗ F (1/2, −1/2, 3, 2, 1) : n .
The LHS contains the representation with lowest weight conjugate to
wm ρ + ρ = (2, 4, −5, −3, −1) + (5, 4, 3, 2, 1) = (7, 8, −2, −1, 0).
This is the sum of the following roots in Δ(n) :

21 , 1 + 2 , 1 − 3 , 1 − 4 , 1 ± 5 ,
(2.25)
22 , 2 − 3 , 2 ± 4 , 2 ± 5 .
This set of roots is stable under the operation of adding negative

simple m-roots,
i.e. , −1 + 2 , −3 + 4 , −4 + 5 , −25 . Thus the vector eα is a lowest weight
vector for Δm . So (2.24) is not 0.
3. Infinitesimal character ρ/2

This case is the smallest possible in view of the necessary condition (2.5), and
thus it warrants special attention. In this case equation (2.5) becomes
(3.1) 2(ρ/2) = τ + ρ,
so τ = 0. Then the multiplicity in (2.6) becomes
(3.2) [L(ρ/2, −sρ/2) : E(ρ)].

3.1. Induced from a unitary character, infinitesimal character ρ/2.

We look at the special case when π has infinitesimal character ρ/2, and is unitarily
induced from a unitary character ξ on a Levi component m. In this case we will be
able to improve over the result of 2.2.
Choose a positive system Δ so that ξ is dominant, and let p = m + n be the
parabolic subalgebra determined by ξ. The representation π = L(λ, −sλ) satisfies
(3.3) λ + sλ = ξ, 2λ = ξ + 2ρm ,
(3.4) λ − sλ = 2ρm , 2sλ = ξ − 2ρm .
It can be shown that this implies s = wm , the long Weyl group element in W (m).
This fact is however not needed in the following.
Let Δ be a positive root system so that 2λ is dominant. Then 2λ = ρ and

τ = 0. Thus
ξ = ρ − 2ρm , sρ = ρ − 4ρm .
Next, the formula (2.13) for the case of general λ simplifies to

π : E(ρ) = Cξ : E(ρ)|m = Cξ : F (ρm ) ⊗ C−ρn ⊗ ∗ n =
(3.5) ∗
Cξ ⊗ F (ρm ) ⊗ Cρn : n .
The LHS of the last line of (3.5) has highest weight
(3.6) ξ + ρm + ρn = ρ − 2ρm + ρm + ρn = ρ + wm ρ,
and lowest weight
(3.7) wm (ρ + wm ρ) = wm ρ + ρ.
We have already shown in Subsection 2.2 that (3.5) is nonzero; we now show that
it is equal to 1, i.e., that the multiplicity of F (wm ρ + ρ) in ∗ n is equal to 1. We
are going to use some classical results of Kostant [K1], [K2] which we describe in
the following.
If B ⊂ Δ, denote by 2ρ(B) the sum of roots in B. In this notation,
(3.8) ρ + wm ρ = 2ρ(B),

where B := α ∈ Δ(n) :
ρ , α > 0 .
Lemma 3.1 (Kostant). Let B ⊂ Δ be arbitrary, and denote by B c the comple-
ment of B in Δ. Then

2ρ(B), 2ρ(B c ) ≥ 0,
with equality if and only if there is w ∈ W such that 2ρ(B) = ρ + wρ. In that case,
B is uniquely determined by w as B = Δ ∩ wΔ.
Proof. Since 2ρ(B) + 2ρ(B c ) = 2ρ, we have
(3.9)
2ρ(B), 2ρ(B c ) =
2ρ(B), 2ρ − 2ρ(B) =
ρ, ρ −
ρ − 2ρ(B), ρ − 2ρ(B).
But ρ − 2ρ(B) is a weight of E(ρ), so the expression in (3.9) is indeed ≥ 0. It is
equal to 0 precisely when ρ − 2ρ(B) is an extremal weight of E(ρ). In that case it is
conjugate to the lowest weight −ρ, i.e., there is w ∈ W such that ρ − 2ρ(B) = −wρ.
For the last statement, notice that Δ = (Δ ∩ wΔ) ∪ (Δ ∩ −wΔ), and that
consequently ρ + wρ = 2ρ(Δ ∩ wΔ), since the elements of Δ ∩ −wΔ cancel out in
the sum ρ + wρ.

∗
Corollary 3.2. The weight ρ + wm ρ occurs with multiplicity 1 in n.
Proof. We can write wm ρ = xρ for a unique x ∈ W. By the last statement of

Lemma 3.1, it follows that the set B from (3.8) is uniquely determined, and hence
the corresponding multiplicity is one.
We have proved
Theorem 3.3. Let P = M N ba a parabolic subalgebra of G, and let Δ be the
set of positive roots corresponding to P . Assume that π is a representation of G
with infinitesimal character ρ/2, which is unitarily induced from a character ξ of
M , such that ξ is dominant with respect to Δ. Then the Dirac cohomology of π
consists of the trivial K-module with multiplicity [Spin : E(ρ)].
The following subsections illustrate applications of this theorem, and also pro-
vide evidence for the following conjecture.
Conjecture 3.4. A unitary representation with infinitesimal character ρ/2
has Dirac cohomology either consisting of the trivial K-type with multiplicity
[Spin : E(ρ)] or equal to 0.
This conjecture sharpens the main conjecture in the introduction for the special
case of infinitesimal character ρ/2, in the sense that it predicts the size of HD (π)
precisely in case when it is nonzero.
3.2. Type A. In this case,
n−1 −n + 1
(3.10) λ=( ,..., ).
4 4
By the classification of unitary representations from [V1], the irreducible unitary
representations with infinitesimal character ρ/2 are all unitarily induced irreducible
from unitary characters on Levi components. Therefore, this case is covered by The-
orem 3.3. It follows that any irreducible unitary representation π with infinitesimal
character ρ/2 has Dirac cohomology consisting of the trivial K-type occurring with
multiplicity [Spin : E(ρ)].
3.3. Type B. In this case
2n − 1 1
(3.11) λ=( , . . . , ).
4 4
The coroots integral on λ form a subsystem of type A(n). In the notation of Section
5, ∨g(λ) ∼
= A(n). According to [B], the unitary representations are all unitarily
induced irreducible from unitary characters on Levi components. Thus Theorem 3.3
applies. It follows that any irreducible unitary representation π with infinitesimal
character ρ/2 has Dirac cohomology consisting of the trivial K-type occurring with
multiplicity [Spin : E(ρ)].
3.4. Type C. In this case
n 1
(3.12) λ = ( , . . . , ),
2 2

and in the notation of Section 5, ∨g(λ) ∼
= B( n2 ) × D( n+12 ) or B( n+1
2 ) × D( n2 )
depending on the parity of n.
According to [B], any irreducible unitary representation must be unitarily in-
duced irreducible from unitary characters on factors of type A of a Levi component,
and the trivial or metaplectic representation on the factor of type C. The latter case
is not covered by Theorem 3.3. This situation was illustrated in Example 2.6.
The metaplectic representations will be analyzed in Section 5. In particular,

we will see that depending on the parity of n, exactly one of the two metaplectic
representations has nonzero Dirac cohomology.
Our conjectures predict that the Dirac cohomology of an irreducible unitary
representation π with infinitesimal character ρ/2 is the trivial K-type occurring with
multiplicity [Spin : E(ρ)], in case the representation that π is induced from has the
metaplectic representation of appropriate parity, or the trivial representation on
the factor of type C of the Levi component. Otherwise, HD has to be zero.
3.5. Type D. In this case
n−1 1
(3.13) λ=( , . . . , , 0),
2 2

and in the notation of Section 5, ∨g(λ) ∼
= D( n+1 2 ) × D( n2 ). According to [B], the
unitary representations are all unitarily induced irreducible from unitary characters
on the factors of a Levi component of type A, and a unipotent representation of
the factor of the Levi component of type D.
Our conjectures predict that such a representation π has nonzero Dirac co-
homology precisely when the corresponding unipotent representation has nonzero
Dirac cohomology and that in this case HD (π) consists of the trivial K-type oc-
curring with multiplicity [Spin : E(ρ)]. We describe the unipotent representations
with nonzero Dirac cohomology in Section 5.
3.6. Type F. We investigate Conjecture 3.4. The calculations were performed
using LiE. We list the Hermitian parameters in the case of infinitesimal character
ρ/2 in the case of F4 . The simple roots, coroots and weights are
(3.14)
(0, 1, −1, 0) − (0, 0, 1, −1) =>= (0, 0, 0, 1) − (1/2, −1/2, −1/2, −1/2).
(3.15)
(0, 1, −1, 0) − (0, 0, 1, −1) =<= (0, 0, 0, 2) − (1, −1, −1, −1),
(3.16)
(1, 1, 0, 0) − (2, 1, 1, 0) =>= (3/2, 1/2, 1/2, 1/2) − (1, 0, 0, 0).
In these coordinates, ρ/2 = (5/2, 3/2, 1, 1/2). The Hermitian parameters are in
the list below. In all cases, λL is (5/2, 3/2, 1, 1/2). The corresponding λR are
in the first column. The K-type μ indicates a K-type which has signature op-
posite to that of the lowest K-type μ. For the parameters where the coordinates
are (. . . , 1, . . . ) × (. . . , −1, . . . ), the Langlands quotients are unitarily induced ir-
reducible from the remainder of the parameter on a B3 ; so the representation is
unitary if and only if the one with remainder on B3 is unitary. So we did not list
a μ which detects the nonunitarity. It is visible from the table that all irreducible
unitary representations with infinitesimal character ρ/2 have Dirac cohomology
consisting of the trivial K-type occurring with multiplicity [Spin : E(ρ)]. Namely,
each of these representations has K-type E(ρ) with multiplicity one. All unitary
representations are unitarily induced from unipotent representations tensored with
unitary characters. The results conform to Conjecture 3.4.
Table 1. Hermitian parameters at ρ/2 for F4 .
λR μ ν Unitary E(ρ) μ
(5/2, 3/2, 1, 1/2) (0, 0, 0, 0) (5, 3, 2, 1) NO (1, 0, 0, 0)
(5/2, 3/2, 1, −1/2) (1, 0, 0, 0) (0, 5, 3, 2) NO (1, 1, 0, 0)
(5/2, −3/2, 1, 1/2) (3, 0, 0, 0) (0, 5, 2, 1) NO (3, 1, 0, 0)
(−5/2, 3/2, 1, 1/2) (5, 0, 0, 0) (0, 3, 2, 1) YES 1
(5/2, −3/2, 1, −1/2) (3, 1, 0, 0) (0, 0, 5, 2) NO (3, 1, 1, 1)
(−5/2, 3/2, 1, −1/2) (5, 1, 0, 0) (0, 0, 3, 2) NO (5, 1, 1, 1)
(−5/2, −3/2, 1, 1/2) (5, 3, 0, 0) (0, 0, 2, 1) YES 1
(−5/2, −3/2, 1, −1/2) (5, 3, 1, 0) (0, 0, 0, 2) NO (5, 3, 1, 1)
(5/2, 1/2, 1, 3/2) (1, 1, 0, 0) (2, −2, 2, 5) NO (2, 0, 0, 0)
(−5/2, 1/2, 1, 3/2) (5, 1, 1, 0) (0, 2, −2, 2) NO (5, 2, 0, 0)
(3/2, 5/2, 1, 1/2) (1, 1, 0, 0) (4, −4, 2, 1) NO (1, 0, 0, 0)
(3/2, 5/2, 1, −1/2) (1, 1, 1, 0) (4, −4, 0, 2) NO (2, 0, 0, 0)
(1/2, 3/2, 1, 5/2) (2, 2, 0, 0) (3, −3, 2, 3) NO (3, 1, 0, 0)
(1/2, −3/2, 1, 5/2) (3, 2, 2, 0) (0, 3, −3, 2) NO (7/2, 5/2, 1/2, 1/2)
(5/2, −1/2, 1, −3/2) (2, 2, 0, 0) (1, −1, 5, 2) NO (3, 1, 0, 0)
(−5/2, −1/2, 1, −3/2) (5, 2, 2, 0) (0, 1, −1, 2) NO (5, 2, 2, 1)
(−3/2, −5/2, 1, 1/2) (4, 4, 0, 0) (1, −1, 2, 1) NO (9/2, 7/2, 1/2, 1/2)
(−3/2, −5/2, 1, −1/2) (4, 4, 1, 0) (1, −1, 1, 2) NO (4, 4, 1, 1)
(−1/2, −3/2, 1, 5/2) (3, 3, 0, 0) (2, −2, 3, 2) NO (4, 2, 0, 0)
(−1/2, −3/2, 1, −5/2) (3, 3, 3, 0) (2, 0, −2, 2) NO (9/2, 5/2, 3/2, 1/2)
(5/2, 3/2, −1, 1/2) (2, 0, 0, 0) (0, 5, 3, 1) YES 1
(5/2, 3/2, −1, −1/2) (2, 1, 0, 0) (0, 0, 5, 3) NO
(5/2, −3/2, −1, 1/2) (3, 2, 0, 0) (0, 0, 5, 1) NO
(−5/2, 3/2, −1, 1/2) (5, 2, 0, 0) (0, 0, 3, 1) YES 1
(5/2, −3/2, −1, 1/2) (3, 2, 0, 0) (0, 0, 5, 1) NO
(−5/2, 3/2, −1, −1/2) (5, 2, 1, 0) (0, 0, 0, 3) NO
(−5/2, −3/2, −1, 1/2) (5, 3, 2, 0) (0, 0, 0, 1) YES 1
(−5/2, −3/2, −1, −1/2) (5, 3, 2, 1) (0, 0, 0, 0) YES 1
(5/2, 1/2, −1, 3/2) (2, 1, 1, 0) (0, 2, −2, 5) NO
(−5/2, 1/2, −1, 3/2) (5, 2, 1, 1) (0, 0, 2, −2) NO
(3/2, 5/2, −1, 1/2) (2, 1, 1, 0) (0, 4, −4, 1) NO
(3/2, 5/2, −1, −1/2) (2, 1, 1, 1) (0, 4, 0, −4) NO
(1/2, 3/2, −1, 5/2) (2, 2, 2, 0) (0, 3, −3, 3) NO
(1/2, −3/2, −1, 5/2) (3, 2, 2, 2) (0, 3, −3, 0) NO
(5/2, −1/2, −1, −3/2) (2, 2, 2, 0) (1, 0, −1, 5) NO
(−5/2, −1/2, −1, −3/2) (5, 2, 2, 2) (0, 1, 0, −1) YES 1
(−3/2, −5/2, −1, 1/2) (4, 4, 2, 0) (1, −1, 0, 1) YES 1
(−3/2, −5/2, −1, −1/2) (4, 4, 1, 0) (1, −1, 1, 2) YES 1
(−1/2, −3/2, −1, 5/2) (3, 3, 2, 0) (2, −2, 0, 3) NO
(−1/2, −3/2, −1, −5/2) (3, 3, 3, 2) (2, 0, −2, 0) YES 1
4. The case of GL(n, C)

In this section we present evidence for the following conjecture. As mentioned
earlier, this conjecture is known to hold when λ is itself integral regular.
Conjecture 4.1. Let π = L(λ, −sλ) be an irreducible unitary representation
of GL(n, C), such that 2λ is regular and integral. Let Δ be the positive root system
such that λ is dominant, and let ρ be the corresponding half sum of the positive
roots. Then the Dirac cohomology of π is the K-type E(2λ − ρ), with multiplicity
[Spin : E(ρ)].
We use the usual coordinates. The unitary dual of GL(n, C) is known by
the results of Vogan [V1]. A representation is unitary if and only if it is a Stein
complementary series from a representation induced from a unitary character on
a Levi component. In order for π to have Dirac cohomology, 2λ must be integral.
Therefore π must be unitarily induced from a unitary character.
We note that a unitary character Cξ always has Dirac cohomology equal to
Cξ ⊗ Spin. Namely, D = 0 on Cξ ⊗ Spin.
In the next subsection we prove Conjecture 4.1 in case π is induced from a
unitary character of a maximal parabolic subgroup. We have also verified the
conjecture in some other cases but we do not present them because the notation
and arguments get increasingly complicated.
4.1. Maximal parabolic case. Let G = GL(n, C), and let P = M N be a
parabolic subgroup such that M = GL(a) × GL(b). We consider the case when
π is induced from a unitary character of M. Then λ is formed of integers or half-
integers. Conjugating 2λ to be dominant with respect to the usual positive form,
we can write it as

(4.1) 2λ = α + 2k, . . . , α + 2, α, α − 1, . . . , β + 1, β, β − 2, . . . , β − 2l ,
where α and β are integers of opposite parity, and k ≥ 0. The module is induced
from a unitary character on a Levi component GL(a) × GL(b) ⊂ GL(n = a + b).
Then a, b and the unitary character ξ = (ξ1 , ξ2 ) are
α−β+1 α−β+1
a= + k, b= + l,
(4.2) 2 2
α+β+1 α+β−1
ξ1 = + k, ξ2 = − l.
2 2
The case l < 0 is similar to l ≥ 0, so for simplicity of exposition we treat l ≥ 0 only.
The condition for ξ to be dominant for the standard positive system (λ is
dominant for it) is that k + l + 1 ≥ 0. By changing λ to −λ and conjugating to
make it dominant, we assume this to be the case. Assume that a ≥ b i.e. k ≥ l; the
other case is similar.
Proposition 4.2. The K-structure of π := IndG
P [ξ] is formed of
(ξ1 + x1 , . . . , ξ1 + xb , ξ1 , . . . , ξ1 , ξ2 − xb , . . . , ξ2 − x1 ), xj ∈ N, xi ≥ xi+1 .
occurring with multiplicity 1.
Proof. Change the notation so that G = U (a, b) with maximal compact sub-
group U (a)×U (b) for this proof only. The problem of computing the aforementioned
multiplicities is equivalent to computing the multiplicity of the K-type μ = ξ1 ⊗ ξ2
in any finite dimensional representation. A finite dimensional representation has
Langlands parameter given by a minimal principal series of G. The Levi compo-

nent is M0 = U (1)b × U (a − b). This principal series has to contain μ. But μ is
1-dimensional, and its restriction to M0 is clear. The result follows from computing
the parameter of a principal series whose Langlands subquotient is finite dimen-
sional and contains μ. The multiplicity follows from the fact that μ occurs with
multiplicity 1.
Recall that we need to consider the K-type with highest weight τ = 2λ − ρ and
try to realize it in the tensor product π ⊗ Spin, or equivalently in π ⊗ E(ρ), as a
PRV component. Since the number of coordinates is a + b,

α+β k+l α+β k+l
(4.3) ρ= + ,...,− − .
2 2 2 2
It follows that τ equals

β+α k−l β+α k−l
τ= + + k, . . . , + + 1,
2 2 2 2
(4.4) + ,..., + ,
2 2 2 2
+ − 1, . . . , + −l
2 2 2 2
On the other hand, since the K-types of π have highest weight equal to the sum of
ξ and roots in Δ(n), μ − ρ has coordinates

2β + 1 k − l β+α k−l
μ−ρ= + + x1 , . . . , + + l + xb ,
2 2 2 2
(4.5) + + l + 1 + xb+1 , . . . , + + k + xa ,
2 2 2 2
β+α k−l 2α − 1 k − l
+ − l − yb , . . . , + − y1
2 2 2 2
with
· · · ≥ xi ≥ xi+1 · · · ≥ 0
(4.6)
· · · ≥ yj ≥ yj+1 ≥ · · · ≥ 0.
The coordinates in the middle of μ−ρ are term by term bigger than the coordinates
appear at the beginning of τ. This forces xa−k+1 = · · · = xa = 0. By the same
argument yb = · · · = yb−l+1 = 0. Note from formula 4.2 that a ≥ k and b ≥ l.
The coordinates that are left over from τ are all equal, so the remaining yj , xi are
uniquely determined.
5. Unipotent representations with Dirac cohomology

In this section we give an exposition of unipotent representations, and compute
Dirac cohomology for many examples.
5.1. Langlands Homomorphisms. In order to explain the parameters of
unipotent representations we recast the classification of (g, K)-modules in terms of
Langlands homomorphisms.
First some notation: For the field of reals the Weil group is
WR := C× · {1, j}, j 2 = −1 ∈ C× , jzj −1 = z,
where Γ := Gal(C/R) is the Galois group. There is a canonical map WR −→ Γ that

maps C× to 1, and j to the generator γ of Γ.
A linear connected reductive algebraic group G is given by its root datum
(X ∗ , R, X∗ , Ř). (G ⊃ B = HN where B is a Borel subgroup, H a Cartan sub-
group. X∗ are the rational characters of H, X ∗ the 1-parameter subgroups, R the
roots and Ř the coroots).
A real form G(R) = G(R) is the fixed points of an antiholomorphic automor-
phism σ : G(C) −→ G(C). Then σ induces an automorphism a of the root datum,
and therefore an automorphism ∨ a of the dual root datum (X∗ , Ř, X ∗ , R). The
Langlands dual is L G := ∨ G Γ, where ∨ G is the complex group attached to
the dual root datum. The nontrivial element of Γ acts on ∨ G by an automorphism
induced by ∨ a.
A Langlands homomorphism is a continuous group homomorphism Φ from WR
into LG, satisfying the commutative diagram
Φ
WR −→ L
G

Γ
×
and such that Φ(C ) is formed of semisimple elements. The main result of the
Langlands classification is that ∨ G conjugacy classes of Langlands homomorphisms
parametrize equivalence classes of irreducible (g, K) modules (more precisely, char-
acters of Φ(WR )/Φ(WR )0 ). In the case of a complex group viewed as a real group,
this specializes to the following.
Example 5.1. G(R) is a complex group G0 viewed as a real group. Then
∨
G =∨ G0 ×∨ G0 , ∨
a(x, y) = (y, x).
∗ ∗
Φ ←→ (λL , λR ) ∈ h × h , λ L − λR ∈ X ∗
The irreducible module L(λL , λR ) is obtained as follows. Let B = HN be a
Borel subgroup with H = T · A a Cartan subgroup such that T = K ∩ H, and A is
split. Then μ := λL − λR determines a character of T, ν := λL + λR a character of
A. The standard module and irreducible module attached to Φ are as before,
B [Cμ ⊗ Cν ⊗ 11]K−finite ,
X(λL , λR ) := IndG
L(λL , λR ) unique irreducible quotient containing Vμ .
5.2. Unipotent Representations. An Arthur parameter is a homomorphism
Ψ : WR × SL(2) −→ L
G
such that Ψ(W
R ) 1/2
is bounded. The Langlands homomorphism attached to Ψ is
z 0
ΦΨ (z) := Ψ z,
0 z −1/2
A special unipotent parameter is an Arthur parameter satisfying Ψ |C× = Triv .
Then {Ψ}/∨G corresponds to ∨G conjugacy classes {∨θ,∨e,∨h,∨f } satisfying Ad∨ θ ∨e =
−∨e, Ad∨ θ ∨h =∨h, Ad∨ θ ∨f = −∨f, and Ad∨ θ 2 = Id . If we decompose ∨g =∨k +∨s ac-
cording to the eigenvalues of ∨θ, then the Ψ are in 1-1 correspondence with ∨K-orbits
of nilpotent elements in ∨s.
Fix an infinitesimal character χǑ =∨h/2. The unipotent packet attached to the
∨
G-nilpotent orbit Ǒ corresponding to Ψ is the set of irreducible representations
with annihilator in U (g) maximal containing the ideal in Z := U (g)G corresponding

to χǑ . It is described in [BV].
For general unipotent representations, the same definitions apply, but each Ǒ
has a finite number of infinitesimal characters attached to it. For complex classical
groups they are listed in [B].
Example 5.2. The metaplectic representation of Sp(2n, C) is unipotent. The
orbit Ǒ ⊂ so(2n + 1, C) corresponds to the partition (2n − 1, 1, 1). Then in standard
coordinates
χǑ = (n − 1, . . . , 2, 1, 1, 0)
but the infinitesimal character we want is
(n − 1/2, . . . , 1/2).
This is a special case of the procedure to attach finitely many infinitesimal charac-
ters to Ǒ.
Remark 5.3. In the following we will describe explicitly the unipotent repre-
sentations with nonzero Dirac cohomology for each of the types A,B,C,D, and E.
After identifying the representations, we will need a description of their K-types.
For type A, this follows from Proposition 4.2. In other cases, the information can
be obtained from realizing the unipotent representations in question via dual pair
correspondences. We state the precise result in each case, but only sketch the
arguments in type A and D.
5.3. Type A. Let G = GL(n, C). As we have seen, unipotent π correspond
to partitions of n. The partition into just one part corresponds to the trivial
representation, so we know the Dirac cohomology is equal to the spin module. In
the rest of this section we skip this obvious case.
Since λ must be regular, we see from the way λ is constructed from a partition
that we should only consider partitions of n into two parts a, b of opposite parity.
In particular, n must be odd. So we take
(5.1) 2λ = (a − 1, a − 3, . . . , b, b − 1, . . . , −b + 1, −b, . . . , −a + 3, −a + 1),
where we assume a > b. The corresponding unipotent representation is spherical,
GL(a+b)
(5.2) π = IndGL(a)×GL(b) [Triv ⊗ Triv].
Since GL(a)×GL(b) ⊂ GL(a+b) comes from a symmetric pair, Helgason’s theorem
applies (see [W], Section 3.3), and the K-structure is
(5.3) μ = (α1 , . . . , αb , 0, . . . , 0, −αb , . . . , −α1 ), αj ∈ N
occurring with multiplicity 1. The WF-set is the nilpotent orbit with two columns
of length a and b.
This case is covered by the results of Subsection 4.1. The Dirac cohomology
consists of a single K-type with highest weight
a−b−1 a−b−1
(5.4) τ =( , . . . 1, 0, . . . , 0, −1, . . . , − )
2 2
2b+1
and multiplicity [Spin : E(ρ)]. The K-type E(μ) of π such that E(τ ) appears in
E(μ) ⊗ E(ρ) is given by
a+b−1 a−b−1 a−b−1 a+b−1
(5.5) μ=( ,..., , 0, . . . , 0, − ,...,− ).
2 2 2 2
5.4. Type B. Let G = SO(2n + 1, C). We use the standard coordinates.

To ensure that 2λ is regular integral, there is only one possible Arthur param-
eter, namely the case of Ǒ equal to the principal nilpotent orbit. The results in
[B] give more unipotent representations, all spherical. They are associated to the
orbits Ǒ which have partitions formed of exactly two elements. The WF-set of a
nontrivial unipotent representation π must be a nilpotent orbit with two columns
of opposite parity, 2b + 1 and 2a. Then 2λ is W -conjugate to
(5.6) (2a, 2a − 3, . . . , 2; 2b − 1, 2b − 3, . . . , 1), a, b ∈ N.
Only the cases 2b + 1 > 2a, i.e. b ≥ a, are unitary. For b > a we get
(5.7) 2λ = (2b − 1, 2b − 3, . . . , 2a + 3, 2a + 1, 2a, 2a − 1, . . . , 2, 1).
Since
(5.8) ρ = (a + b − 1/2, a + b − 3/2, . . . , 1/2),
we see that
(5.9) τ = 2λ − ρ = (b − a − 1/2, b − a − 3/2, . . . , 3/2, 1/2, 1/2, . . . , 1/2),
with the last 2a + 1 coordinates equal to 1/2.
The K-structure of π is
(5.10) (α1 , α1 , α2 , α2 , . . . , αa , αa , 0, . . . , 0), αj ∈ N

b−a
occurring with multiplicity 1. We sketch a proof. By the results of [AB], π occurs

in the oscillator representation correspondence for the dual pair
G × G := O(2a + 2b + 1, C) × Sp(2b, C),
paired with the trivial representation of Sp(2b). Consider the see-saw pair
O ∗ (4a + 4b + 2) Sp(4a, R)
∪ ∪
O(2a + 2b + 1, C) Sp(2a, C)
∪ ∪
O(2a + 2b + 1, 0) Sp(2a, 0)

where G × G occurs in the middle row, and the bottom row is formed of the max-
imal compact subgroups of G and G . Let Ω be the (Fock model of the) oscillator
representation defined in [H]. Since π ⊗ Triv is the quotient of Ω, π is a subquotient
of Ω/g Ω. As a module for g, this latter quotient is admissible, has the infinitesimal
character of π and WF-set equal to W F (π). Thus this quotient is formed of unipo-
tent representations only. Since π is the only unipotent representation with this
WF-set and infinitesimal character, Ω/g Ω is a multiple of π. Since the spherical
vector occurs with multiplicity 1, it equals π. On the other hand we can view Ω as
a representation for the pair O(2a + 2b + 1) × Sp(4a, R). The correspondence is well
understood when one of the groups is compact; the oscillator representation de-
composes into a direct sum ⊕V (μ) ⊗ L(μ), where L(μ) is the lowest weight module
corresponding to μ, and μ occurs only when μ2a+1 = · · · = μa+b = 0. Taking the
quotient by g Ω, we find that a V (μ) only occurs if the corresponding representation
W (μ) of U (2a) contains the trivial representation of Sp(2a, 0). Since U (2a), Sp(2a)
is a symmetric pair, Helgason’s theorem implies that the K-types are of the form
(5.10), and occur with multiplicity 1.
Now we have to identify K-types E(μ) of π such that μ − ρ is conjugate to τ
under W . We calculate
(5.11) μ − ρ = (α1 − a − b + 1/2, α1 − a − b + 3/2, . . . ,
αa + a − b − 3/2, αa + a − b − 1/2,
a − b + 1/2, a − b + 3/2, . . . , −3/2, −1/2).

b−a
To be conjugate to τ , this expression must have 2a + 1 components equal to ±1/2.

Since there is only one such component among the last b − a components, the first
2a components must all be equal to ±1/2. Since the first component is smaller
than the second by one, the third component is smaller than the fourth by one,
etc., we see that the first, third etc. components must be −1/2 while the second,
fourth, etc. components must be 1/2. This completely determines μ:
(5.12) α1 = a + b − 1, α2 = a + b − 3, . . . , αa = b − a + 1.
It is now clear that for this μ we indeed get a contribution to HD (π), and moreover
we can see exactly which w conjugates μ − ρ to τ .
It remains to consider the case b = a. The calculation and the final result are
completely analogous. We get
(5.13) τ = (1/2, 1/2, . . . , 1/2),
corresponding to
(5.14) μ = (2a − 1, 2a − 1, 2a − 3, 2a − 3, . . . , 1, 1).
5.5. Type C. Let G = Sp(2n, C). We use the usual coordinates.
As in the other cases, the only Arthur parameters with 2λ regular integral
correspond to the principal nilpotent. In this case λ itself is integral. The only
other case when λ can be regular corresponds to the subregular Ǒ, corresponding
to the partition 1, 1, 2n − 1. In this case, the unipotent representations are the two
metaplectic representations, πeven and πodd . The corresponding λ is given by
(5.15) 2λ = (2n − 1, 2n − 3, . . . , 3, 1).
The other cases analogous to type B are not unitary. The K-structures of πeven
and πodd are
(2α, 0, . . . , 0),
(5.16)
(2α + 1, 0, . . . 0), α ∈ N.
Here α = 0 is allowed. The WF-set is the nilpotent with columns 2n − 1, 1.
Since in this case
(5.17) ρ = (n, n − 1, . . . , 2, 1),
we see that
(5.18) τ = 2λ − ρ = (n − 1, n − 2, . . . , 1, 0).
For each of the two metaplectic representations the K-types are given by μ =
(k, 0, 0, . . . , 0), and therefore
(5.19) μ − ρ = (k − n, −(n − 1), −(n − 2), . . . , −2, −1).
This should be equal to τ up to W , and this happens precisely when k = n. (Recall

that W consists of permutations and arbitrary sign changes.)
So we see that for even n, HD (πeven ) consists of E(τ ), for τ as in (5.18), without
multiplicity other than the global multiplicity [Spin : E(ρ)], while HD (πodd ) = 0.
For odd n, the situation is reversed: HD (πeven ) = 0, while HD (πodd ) consists
of E(τ ), with multiplicity [Spin : E(ρ)].
5.6. Type D. Let G = SO(2n, C). We use the usual coordinates.

Since 2λ must be regular integral, in this case the WF-sets of the nontrivial
unipotent representations can only be nilpotents with columns 2b, 2a − 1, 1, where
a + b = n.
By [B], there are two unipotent (so also unitary) representations with 2λ W -
conjugate to (2a − 1, 2a − 3, . . . , 1; 2b − 2, . . . , 0); the spherical one, and the one with
lowest K-type (1, 0, . . . , 0) and parameter
(a − 1/2, . . . , 3/2, −1/2, b − 1, . . . , 1, 0) × (a − 1/2, . . . , 3/2, 1/2, a − 1, . . . , 1, 0).
Made dominant for the standard positive system,
(5.20) 2λ = (2b − 2, 2b, . . . , 2a + 2, 2a, 2a − 1, 2a − 2, . . . , 1, 0).
(When b = a, the parameter is (2a − 1, 2a − 2, . . . , 1, 0).) Since
(5.21) ρ = (a + b − 1, a + b − 2, . . . , 1, 0),
we see that
(5.22) τ = 2λ − ρ = (b − a − 1, . . . , 1, 0, 0, . . . , 0)

2a
The K-structure of our unipotent representations is given by

μ = (α1 , . . . , α2a , 0, . . . , 0), αj ∈ N, αj ∈ 2N,
(5.23)
μ = (α1 , . . . , α2a , 0, . . . , 0), αj ∈ N, αj ∈ 2N + 1.
The argument is similar to type B, but more involved because π is not trivial.
The first case is for the spherical representation, the second for the other one.
Therefore,
(5.24) μ − ρ = (α1 − (a + b − 1), . . . , α2a − (b − a), −(b − a − 1), . . . , −1, 0).
Since τ has 2a + 1 zeros, the only way μ − ρ can be conjugate to τ is to have
(5.25) α1 = a + b − 1, α2 = a + b − 2, . . . , α2a = b − a.
Using (5.23), we conclude that for even a, the spherical unipotent representation
has HD equal to E(τ ), with multiplicity [Spin : E(ρ)], while the nonspherical
representation has HD = 0. For odd a the situation is reversed: the spherical
representation has HD = 0, while the nonspherical one has HD equal to E(τ ),
with multiplicity [Spin : E(ρ)]. (Recall that for type D the Weyl group consists of
permutations combined with an even number of sign changes, but we can use all
sign changes because we are wroking with the full orthogonal group.)
5.7. Type E6 . We use the Bourbaki realization. There are two integral sys-
tems, A5 A1 which gives the nilpotent 3A1 , and D5 T1 which gives 2A1 . The param-
eters are
λ = (−5/2, −3/2, −1/2, 1/2, 3/4, −3/4, −3/4, 3/4) ←→ 3A1
(5.26)
λ = (−9/4, −5/4, −1/4, 3/4, 7/4, −7/4, −7/4, 7/4) ←→ 2A1 .
The representations are factors in IndE
A5 [Cν ]. The parameter is
6
(−11/4, −7/4, −3/4, 1/4, 5/4, −5/4, −5/4, 5/4)+

(5.27)
+ν(1/2, 1/2, 1/2, 1/2, 1/2, −1/2, −1/2, 1/2).
The two points above are ν = 1/2 and ν = 1. The representations are unitary
because the induced module has multiplicity 1 K-structure.
5.8. Type E7 . We use the Bourbaki realization. There are three integral
systems, D6 A1 which gives the nilpotent (3A1 ) , E6 T1 which gives 2A1 , and A7
which gives 4A1 . The parameters for the first two are
λ = (0, 1, 2, 3, 4, 5, −1, 1) ←→ (3A1 )
(5.28)
λ = (0, 1, 2, 3, 4, −7/2, −17/4, 17/4) ←→ 2A1 .
The first representation is a factor in IndE
D6 [Cν ]. The parameter is
6
(5.29) (0, 1, 2, 3, 4, 5, 0, 0) + ν(0, 0, 0, 0, 0, 0, −1, 1).

The point above is ν = 1, an end point of a complementary series. In any case
the representation is multiplicity-free, so the representation is unitary. The second
representation is a factor in IndE 7
E6 [Cν ]. The parameter is
(5.30) (0, 1, 2, 3, 4 − 4, −4, 4) + ν(0, 0, 0, 0, 0, 1, −1/2, 1/2)

with ν = 1/2. The representation is unitary because it is at an end point comple-
mentary series; also the induced module is multiplicity-free.
The third representation has parameter
(5.31) (−9/4, −5/4, −1/4, 3/4, 7/4, 11/4, −4, 4).
This is the minimal length parameter which gives the integral system A7 . By the
work of Adams-Huang-Vogan [AHV], the K-structure is multiplicity free and a
full lattice in E7 . It does not occur in a multiplicity free induced module, and is
not an end point of a complementary series.
5.9. Type E8 . We use the Bourbaki realization. There are two integral sys-
tems, D8 which gives the nilpotent 4A1 , and E7 A1 which gives 3A1 . The parameters
are
λ = (0, 1, 2, 3, 4, 5, 6, 8) ←→ 4A1
(5.32)
λ = (0, 1, 2, 3, 4, 5, −8, 9) ←→ 3A1 .
These are the minimal length parameters which gives the integral systems D8 and
E7 A1 . By the work of Adams-Huang-Vogan [AHV], the K-structure of the first one
is multiplicity free and a full lattice in E8 . It does not occur in a multiplicity-free
induced module, and is not an end point of a complementary series. The second
one is also multiplicity-free, and occurs at an endpoint of a complementary series.
It is conjectured that these representations are unitary.
References
[AB] J. Adams, D. Barbasch, Reductive dual pair correspondence for complex groups, J. Funct.
Anal. 132 (1995), no. 1, 1-42.
[AHV] J. Adams, J.-S. Huang, D.A. Vogan, Jr., Functions on the model orbit in E8 , Represent.
Theory 2 (1998), 224–263.
[B] D. Barbasch, The unitary dual for complex classical Lie groups, Invent. Math. 96 (1989),
no. 1, 103–176.
[BV] D. Barbasch, D. Vogan, Unipotent representations of complex semisimple groups, Ann. of
Math. 121 (1985), 41–110.
[BW] A. Borel, N.R. Wallach, Continuous cohomology, discrete subgroups, and representations
of reductive groups, second edition, Mathematical Surveys and Monographs 67, American
Mathematical Society, Providence, RI, 2000.
[E] T. Enright, Relative Lie algebra cohomology and unitary representations of complex Lie
groups, Duke Math. J. 46 (1979), no. 3, 513–525.
[H] R. Howe, Transcending classical invariant theory, J. Amer. Math. Soc. 2 (1989), no. 3,
535–552.
[HKP] J.-S. Huang, Y.-F. Kang, P. Pandžić, Dirac cohomology of some Harish-Chandra modules,
Transform. Groups 14 (2009), no. 1, 163–173.
[HP1] J.-S. Huang, P. Pandžić, Dirac cohomology, unitary representations and a proof of a con-
jecture of Vogan, J. Amer. Math. Soc. 15 (2002), 185–202.
[HP2] J.-S. Huang, P. Pandžić, Dirac Operators in Representation Theory, Mathematics: Theory
and Applications, Birkhauser, 2006.
[Kn] A.W. Knapp, Representation Theory of Semisimple Groups: An Overview Based on Ex-
amples, Princeton University Press, Princeton, 1986. Reprinted: 2001.
[K1] B. Kostant, A formula for the multiplicity of a weight, Trans. Amer. Math. Soc. 93 (1959),
53–73.
[K2] B. Kostant, Lie algebra cohomology and the generalized Borel-Weil theorem, Ann. of Math.
74 (1961), 329–387.
[LS] J.-S. Li, J. Schwermer, Automorphic representations and cohomology of arithmetic groups,
Challenges for the 21st century (Singapore, 2000), 102–137, World Sci. Publ., River Edge,
NJ, 2001.
[P1] R. Parthasarathy, Dirac operator and the discrete series, Ann. of Math. 96 (1972), 1–30.
[P2] R. Parthasarathy, Criteria for the unitarizability of some highest weight modules, Proc.
Indian Acad. Sci. 89 (1980), 1–24.
[PRV] K.R. Parthasarathy, R. Ranga Rao, V.S. Varadarajan, Representations of complex
semisimple Lie groups and Lie algebras, Ann. of Math. 85 (1967), 383–429.
[SR] S.A. Salamanca-Riba, On the unitary dual of real reductive Lie groups and the Aq (λ)
modules: the strongly regular case, Duke Math. J. 96 (1998), 521–546.
[V1] D.A. Vogan, Jr., The unitary dual of GL(n) over an Archimedean field, Invent. Math. 83
(1986), no. 3, 449–505.
[V2] D.A. Vogan, Jr., Dirac operators and unitary representations, 3 talks at MIT Lie groups
seminar, Fall 1997.
[VZ] D.A. Vogan, Jr., and G.J. Zuckerman, Unitary representations with non-zero cohomology,
Comp. Math. 53 (1984), 51–90.
[W] G. Warner, Harmonic analysis on semisimple Lie groups I, Springer-Verlag, Berlin, Hei-
delberg, New York, 1972.
[Zh] D.P. Zhelobenko, Harmonic analysis on complex semisimple Lie groups, Mir, Moscow,
1974.
Department of Mathematics, Cornell University, Ithaca NY 14850, USA
Dept. of Mathematics University of Zagreb, Bijenička 30, 10000 Zagreb, Croatia

E-mail address: pandzic@math.hr
Volume 546, 2011
Algebraic index theorem for symplectic deformations of

gerbes.
Paul Bressler, Alexander Gorokhovsky, Ryszard Nest, and Boris Tsygan
Dedicated to Henri Moscovici on his 65th birthday.
Abstract. We extend the algebraic index theorem from [7] to symplectic

deformations of gerbes in the class of algebroid stacks.
Contents
1. Introduction
2. Deformations of gerbes
3. Statement of the result
4. Local constructions
5. Some Lie algebra cohomology classes
6. Proof of Theorem 3.1
References
1. Introduction
In this paper, we continue the program of studying algebroid stack deformations
of gerbes and modules over them. This program is being carried out from different
perspectives in [3], [4], [6], [5], [16], [17], [18], [27], [26], [9], and in other works.
An algebroid stack is a natural generalization of a sheaf of rings. It gives rise to a
sheaf of categories that, in the case of a sheaf of rings, is the sheaf of categories of
modules. The role of algebroid stacks in deformation theory was first emphasized in
[14] and in [19]. Deformations of a sheaf of rings as such are more difficult to classify
than their deformations as a stack. This is closely related to the fact that some of
the most natural deformations appearing in complex analysis happen to be stacks
and not sheaves. As an example, the sheaf of differential operators on a manifold
gives rise to a deformation of the sheaf of functions on the cotangent bundle. If
one replaces the cotangent bundle by an arbitrary holomorphic symplectic manifold,
this deformation has a natural generalization which in general is an algebroid stack.
2010 Mathematics Subject Classification. Primary 53D55; Secondary 58J22, 58J42 .

A. Gorokhovsky was partially supported by NSF grant DMS-0900968. B. Tsygan was
partially supported by NSF grant DMS-0906391.
1
23
24
2 P.P.BRESSLER,
BRESSLER, A. GOROKHOVSKY, R.
A.GOROKHOVSKY, NEST,AND
R.NEST, ANDB.TSYGAN
B. TSYGAN
The first obstruction for this stack to arise from a sheaf of algebras is the first
Rozansky-Witten class in the second Dolbeault cohomology [3].
Analytic constructions of algebras on a manifold twisted by a gerbe appeared
in [22]. In this paper the authors also prove a related index theorem. On the more
algebraic side, in [17], Kashiwara and Schapira defined the Hochschild homology
of an algebroid stack deformation of the sheaf of functions on a manifold, and the
characteristic class hh (M) in this homology for a coherent sheaf M. For symplectic
deformations, they constructed the trace density morphism from the their version
of Hochschild homology to the de Rham cohomology. On the other hand, in [5]
we defined the Hochschild and cyclic homologies of an algebroid stack, as well as
the Chern character ch (M) of a perfect complex of modules in the negative cyclic
homology. Presumably, the two definitions of the Hochschild homology coincide,
and hh (M) is the image of ch (M) under the map from the negative cyclic to the
Hochschild homology.
In this paper, we define the trace density for gerbes. It is a morphism from
our versions of the Hochschild and the negative cyclic homology of a symplectic
deformation of a gerbe to the de Rham cohomology. We expect that our map
from the Hochschild cohomology coincides with the one defined in [17]. There
is another map from the negative cyclic homology to the de Rham cohomology,
namely reduction modulo followed by the gerbe version of the Hochschild-Kostant-
Rosenberg map.
The main result of this paper (Theorem 3.1) is the computation of the trace
density map for gerbes in terms of the Hochschild-Kostant-Rosenberg map. Specif-
ically, we establish in Theorem 3.1 that the trace density map is the Hochschild-
Kostant-Rosenberg map, multiplied by the cohomology class A (TM ) eθ where
θ is the characteristic class of the deformation defined in [3] (see also Section 2).
From this we deduce the Riemann-Roch formula for the Chern character of a per-
fect complex (Theorem 3.2). These results were proven for the sheaf deformations
in [24] for the smooth case and in [7] for the analytic case. The index theorem for
elliptic pairs conjectured in [28] follows from the partial case when the manifold is
the cotangent bundle with the standard symplectic structure.
The proof goes essentially through the same steps as the one given in [7]. One
needs however to replace all the constructions by the appropriate twisted versions.
The development of these is the focus of the present paper. Using the results from
[3] and [5] we develop the analogue of Fedosov’s construction [10] for gerbes. We
also construct an appropriate analogue of the Gelfand-Fuks map which allows us
to apply the arguments of formal differential geometry developed by Gelfand and
Kazhdan [12].
This paper is organized as follows. In Section 2 we recall the definitions of
gerbes and classification of their deformations. We also define the gerbe versions of
the Hochschild and cyclic complexes. We are following here [3, 5]. In Section 3.1
we state the main result of the paper and show that it implies the Riemann-Roch
formula for the Chern character of a perfect complex. In Section 4 we introduce
some of the main new ingredients of our proof: the Fedosov-type construction of
the gerbe deformations and the corresponding twisted version of the Gelfand-Fuks
map. In Section 5 we define the Lie algebra cohomology which is needed for the
applications to gerbes. Finally, in the Section 6 we give the proof of the main
theorem, using the results and constructions of the preceding sections.
ALGEBRAIC INDEX THEOREM FOR SYMPLECTIC DEFORMATIONS OF GERBES. 25
3
2. Deformations of gerbes
Here we briefly recall the definitions and results from [3].
Definition 2.1. Let M be a topological space. A stack on M is an equivalence
class of the following data:
(1) an open cover {Uα }α∈I of M ;
(2) for each α ∈ I
a sheaf of rings Aα on Uα ;
(3) for every α, β ∈ I
an isomorphism of sheaves of rings Gαβ : Aβ | (Uα ∩ Uβ ) ∼
= Aα | (Uα ∩ Uβ );
(4) for every α, β, γ ∈ I
an invertible element cαβγ ∈ Aα (Uα ∩ Uβ ∩ Uγ ) satisfying
(2.1) Gαβ Gβγ = Ad (cαβγ ) Gαγ
and such that, for every α, β, γ, δ ∈ I,
(2.2) cαβγ cαγδ = Gαβ (cβγδ ) cαβδ
We refer to [3] for the definitions of the morphisms and 2-morphisms of stacks.
With these operations stacks form a two-groupoid.
A gerbe on a manifold M is a stack for which Aα = OUα (the structure sheaf of
Uα ) and Gαβ = 1. Gerbes are classified up to isomorphism by cohomology classes
∗
in H 2 (M, OM ).
(0)
Definition 2.2. Consider a gerbe given by a two-cocycle cαβγ . A deformation
quantization of this gerbe is a stack such that:
(1) Aα = OUα [[]] as a sheaf of vector spaces, with an associative C [[]]-linear
product structure ∗ of the form
∞

f ∗ g = fg + (i)m Pm (f, g) .
m=1
where Pm are bidifferential

∞ operators and 1 ∗ f = f ∗ 1 = f .
m
(2) Gαβ (f ) = f + m=1 (i) Tm (f ) where Tm are differential operators;
∞ m (m)
(3) cαβγ = m=0 (i) cαβγ .
Let M be a manifold (real smooth or complex analytic) with a symplectic form

ω (smooth or holomorphic). Let A0 be a gerbe on M and let A be a deformation of
A0 as defined above. The commutator {f, g} = i 1
[f, g] |=0 is a well defined Poisson
bracket on OM , and we call A a symplectic deformation of A0 , or a deformation of
A0 along ω, if {f, g} is the Poisson bracket associated to the symplectic structure ω.
The following theorem was proven in [3]:
∗
Theorem 2.3. Let c ∈ H 2 (M, OM ) be the class of the gerbe A0 . and let log c
denote its image in H 2 (M, OM /CM ) under the map induced by the morphism
× log
∗
OM → OM /C×
M −
−→ OM /CM .
(i) If log c is not zero, the set of isomorphism classes of deformations of A0 along
ω is empty,
26
4 P.P.BRESSLER,
R.NEST, ANDB.TSYGAN
B. TSYGAN
(ii) If log c = 0, the set of isomorphism classes of deformations of A0 along ω is

in a bijective correspondence with the affine space of series
∞

1 k
(2.3) θ= [ω] + (i) θk
i
k=0
where [ω] is the class of ω in H (M, C) and θk ∈ H 2 (M, C) .

2
The correspondence alluded to in the second statement above, as well as other

constructions in this paper, depend on the choice of a flat connection on the gerbe
[8], which exists precisely when log c = 0. From now on assume that such a choice
is made. Let us note however that the components θk , k = 0, are independent of
the choice of the connection.
2.1. Twisted simplicial matrix algebras. Let A be an algebroid stack, as
in definition 2.1.
Let us fix an open cover U = {Uα } of M satisfying the conditions of the
definition. For any p-simplex
σ = (α0 , . . . , αn ) of the nerve N (U) of U, set Iσ =
{α0 , . . . , αp } and Uσ = α∈Iσ Uα .
Definition 2.4. The algebra Matrσtw (A) has as elements finite matrices

aαβ Eαβ
α,β∈Iσ
where Eαβ are matrix units and aαβ ∈ Aα (Uσ ) . The product is defined by
aαβ Eαβ · aγ Eγ = δβ aαβ Gαβ (aβγ ) cαβγ Eαγ
For σ ⊂ τ, we define the inclusion
iστ : Matrσtw (A) → Matrτtw (A) ,

aαβ Eαβ → (aαβ |Uτ ) Eαβ ; iστ is a nonunital morphism of algebras and, clearly,
iρτ iτ σ = iρσ .
If V = {Vβ } is a refinement of U with f the refinement assignment, (i.e.,Vβ ⊂
Uf (β) ), we define the pull-back
f ∗ : Matrtw (A) → Matrσtw (A)
f (σ)
as the algebra homomorphism defined by

∗

f aαβ Eαβ = af (γ)f (δ) |Vσ Eγδ .

α,β∈I(f (σ)) γ,δ∈I(σ)
2.2. Homological complexes associated to the stack. Let us start by

fixing some notation. Let A be an associative unital algebra over a unital ring k.
Set
Cp (A, A) = Cp (A) = A⊗(p+1) .
We denote by b : Cp (A) → Cp−1 (A) and B : Cp (A) → Cp+1 (A) the standard differ-
entials from the Hochschild and cyclic homology theory (cf. [21]). The Hochschild
chain complex is by definition (C• (A) , b). Let u be a formal variable of degree −2.
Define
CC− • (A) = (C• (A) [[u]] , b + uB) ;

CC• (A) = C• (A) u, u−1 , b + uB ;

per

CC• (A) = C• (A) u, u−1 / (uC• (A) [[u]]) , b + uB .

5
These are, respectively, the negative cyclic, the periodic cyclic, and the cyclic com-
plexes of A over k. These definitions can be naturally extended to sheaves of
algebras.
In [5], we defined the Hochschild complex C• (A) and the negative cyclic com-
plex CC−• (A) for any algebroid stack. We briefly recall the definitions.
Definition 2.5. (cf. [5] for details) Given an algebroid stack A, set
⎛ ⎞

CC−• (A) =
⎝lim CC−
σp
•−p Matrtw (A) , b + uB + ∂
ˇ⎠
−→
U σ0 ⊂σ1 ⊂···⊂σp
where σi run through simplices of N (U) and

p−1
k p
ˇ σ ···σ =
∂s σk ···σp + (−1) iσp−1 σp sσ0 ···σp−1 .
(−1) sσ0 ···
0 p
k=0
Note that CC−

• (A) is a complex of sheaves on M . The definition of the Hochschild,
cyclic, and periodic cyclic complexes are similar.
Suppose that A is a deformation of a gerbe. The stack of DGLAs C• (A) [1]
is given by setting C• (Aα ) [1] to be the sheaf C• (Aα , Aα ) [1] of polydifferential
Hochschild cochains on Aα with its standard DGLA structure.
3. Statement of the result

3.1. Analytic trace density. In this paper we define the trace density mor-
phism in the derived category of sheaves (see Definition 5.2)
(3.1) τ : CC−
• (A) → CM [[u]] (()) [2d]
where 2d is the dimension of M . Localizing by u and passing to the periodic cyclic

homology, we get the analytic trace density
(3.2) τ : CCper
• (A) → CM ((u)) (()) .
3.2. Topological index. It follows from Theorem 7.1.2 of [3] that CC− • (A0 )
is quasi-isomorphic to CC−
• (O M ) (the theorem in [3] is proven for cochains, but
the argument works for chains without any change). The topological index map is
the composition
σ τ HKR
• (A) −→ CC• (A0 ) −→ CM ((u))
τ top : CCper per
(3.3)
Here σ is the morphism of reduction modulo and τ HKR is the gerbe version of
the Hochschild-Kostant-Rosenberg map; see Section 5.2.
3.3. Formal index theorem for deformation of a gerbe.

Theorem 3.1. At the level of cohomology,

τ =τ top
A (TM ) eθ
Note the difference from the formulation in [7]. Here we use the usual Chern
class instead of its square root
classes of the tangent bundle. There we had the A
because we were using another definition of characteristic classes.
28
6 P.P.BRESSLER,
R.NEST, ANDB.TSYGAN
B. TSYGAN
3.4. Riemann-Roch formula for the Chern character. Recall that in [5]
we defined the Chern character

ch (M) ∈ H0Λ M, CC−

−• (A)
for an algebroid stack A and a perfect complex M of twisted A-modules supported

on a closed subset Λ. Let A be a symplectic deformation of a gerbe A0 that
corresponds to a cohomology class θ. Then σM is a perfect complex of twisted
A0 -modules. Applying the Hochschild-Kostant-Rosenberg map to it, we get
ch (σM) ∈ HΛ• (M, C) .
Theorem 3.1 implies the following
Theorem 3.2.

τ (ch (M)) = ch (σM) (TM ) eθ
A
Proof. It is enough to observe that
ch (σM) = τ top (ch (M)) .
4. Local constructions
4.1. Fedosov construction for deformations of gerbes. Define the Weyl
∗
bunde WM as Sym [[TM ]] [[]] with the Sp (TM )-equivariant Moyal-Weyl product.
Recall that this is the bundle
W / WM

M
with the fiber at the point m ∈ M given by the Weyl algebra of the symplectic
∗ t
vector space (Tm (M ) , ωm ):

Tm ∗
(M )⊗n / ξ ⊗ η − η ⊗ ξ = iωm
t
(ξ, η) .
n
t ∗
with the symplectic structure ωm on Tm (M ) induced by the symplectic structure
ωm on Tm (M ).
Recall the following from [3].
Proposition 4.1. If log c = 0 (cf. Theorem 2.3), then deformations of A0
along ω are classified by equivalence classes of pairs (∇, R) where ∇ is a Der WM -
valued connection on WM preserving the Moyal-Weyl product and R ∈ Ω2 (M, WM )
is a WM -valued two-form such that ∇2 = ad (R) and ∇ (R) = 0.
We refer the reader to [3] for the definition of equivalence; it is not used in the
rest of this paper.
Remark 4.2. We note that the class θ of the deformation from Theorem 2.3
can be obtained from the pair (∇, R) as follows. There exists a lift of ∇ to a i
1
WM -
valued connection (see Section 4.2 for the detailed discussion of this Lie algebra)
2
Then ∇
which we denote by ∇. − R is a closed 1 Ω2 (M )[[]]-valued form. The

cohomology class of this form is the class θ from equation (2.3).
7
Proof. Let us sketch the main points of the proof.

DGLA controlling deformations of the gerbe
Let us consider the complex of differential forms with coefficients in the bundle
∗
of DGLAs of Hochschild cohomological complexes of Sym [[TM ]] with the Gersten-
haber bracket, (cf. [13])
Ω• (M, C• (Sym [[TM
∗
]]) [1]) [[]] .
We give it the structure of a differential graded Lie algebra as follows. Let ∇0 be
the flat connection that is the image of the canonical connection ∇can on the jet
bundle JM under an identification
∼ ∗
(4.1) JM → gr JM = Sym [[TM ]] .
∗
The connection ∇0 induces a flat connection on the bundle of algebras Sym [[TM ]]
• ∗
and therefore on the bundle of their Hochschild complexes C ((Sym [[TM ]]) [1]) [[]],
which will still be denoted by ∇0 , and
(4.2) (Ω• (M, C• (Sym [[TM
∗
]]) [1]) [[]] , ∇0 + δ)
is our DGLA.
It is proven in [3] that for the Hochschild cochain complex C• (A0 ) [1] of any
gerbe one has an L∞ quasi-isomorphism
∼
(4.3) C• (A0 [1]) → (Ω• (M, C• (Sym [[TM
∗
]]) [1]) , ∇0 + δ + [κ, ·])
Here the cochain κ represents the image of log c under the map from H 2 (M, OM /CM )
to the cohomology of the complex (Ω• (M, Sym [[TM ∗
]] /OM ) , ∇0 ) induced by the
composition of morphisms of (complexes of) sheaves
O/C → (Ω•M ⊗ (JM /OM ) , ∇can ) → (Ω•M ⊗ (Sym [[TM
∗
]] /OM ) , ∇0 ) .
Here the first map is induced by taking the infinite jets and the second is the
identification (4.1). We note that the homotopy class of this quasi-isomorphism is
not canonical; it can be fixed however by the choice of a flat connection on the
gerbe A0 . In our case the choices can be made so that κ = 0 and, in particular, the
isomorphism classes of deformations of A0 are in a bijective correspondence with
equivalence classes of Maurer-Cartan elements

∗
(R, A, Π) ∈ Ωi M, Cj (Sym [[TM ]] [[]]) [1]
i+j=2
of (4.2).
One also has (cf. [3]) a quasi-isomorphism of cyclic complexes
∼ •

(4.4) CC−
−• (A0 ) → Ω M, CC− ∗
−• (Sym [[TM ]]) , ∇0 + b + uB .
Deformations along ω
∗
We fix a linear isomorphism of the bundles Sym [[TM ]] [[]] WM which is the
∗
identity on TM and which preserves the natural filtrations on both bundles.
The symplectic deformations that are described in the

proposition above cor-
respond to Maurer-Cartan elements whose Ω0 M, C2 -component is given by
Π (a, b) = a ∗ b − ab, where a ∗ b is the Moyal-Weyl product.
The passage from such a Maurer-Cartan element
to a pair (∇, R)
is given by
∗
∇ = ∇0 + A, where A is the component in Ω1 M, C1 (Sym [[TM ]]) [1] .
30
8 P.P.BRESSLER,
R.NEST, ANDB.TSYGAN
B. TSYGAN
We need to express the negative cyclic complex of the deformed stack in terms
of the above proposition.
Recall (see for example [25]) that, for any associative algebra A, the Lie algebra
of derivations of A and the Abelian graded Lie algebra A [1] act on CC− −• (A) by
operators LD , D ∈ Der (A) , and La , a ∈ A satisfying the relations
[b + uB, La ] = Lad a ; [LD , La ] = LD(a) .
This action extends to an action of the DGLA C• (A) [1] with the Gerstenhaber
bracket.
Proposition 4.3. If log c = 0 and A is a deformation corresponding to the
pair (∇, R) as in Proposition 4.1, then there is a quasi-isomorphism of complexes
∼ •

CC− −• (A) → Ω M, CC− −• (WM ) , ∇ + b + uB + LR
In particular there is an analogous quasi-isomorphism with the negative com-

plexes replaced by periodic.
Proof. Repeating the proof of Theorem 7.1.2 of [3] verbatim, we get a quasi-
isomorphism of negative cyclic complexes as L∞ -modules:
∼ •

CC−−• (A0 [[]]) → Ω M, CC− ∗

−• (Sym [[TM ]] [[]]) [1] , ∇0 + b + uB
Here the left hand side is a module over C• (A0 [[]]) [1], the right hand side is a
module over (Ω• (M, C• (Sym [[TM ∗
]] [[]]) [1]) , ∇0 + δ), and the quasi-isomorphism
is compatible with the L∞ quasi-isomorphism (4.3).
Given a deformation A corresponding to a Maurer-Cartan element of the DGLA
γ of C• (A0 ) [[]] [1] , the negative cyclic complex of A is quasi-isomorphic to the
negative cyclic complex of A0 [[]] twisted by γ, i.e.,to the complex
•

Ω M, CC− ∗
−• (Sym [[TM ]] [[]]) [1] , ∇0 + b + uB + Lγ .
If A is a symplectic deformation, we can choose γ, up to equivalence, of the form

(R, A , Π) as above. As explained in the proof of the previous proposition, the
negative cyclic complex twisted by such a Maurer-Cartan element is precisely the
right hand side in Proposition 4.3.
4.2. The Gelfand-Fuks construction. Put

W=C x d , ξ1 , . . . , ξd ,
1 , . . . , x
with the Moyal-Weyl product. This gives rise to two Lie algebras:
1 1 1
(4.5) g = Dercont (W) = W/ C; g = W,
i i i
where the bracket in
g is given by
2
1 1 1
f, g = (f ∗ g − g ∗ f )
i i i
Put H = Sp (2d) ; h = sp (2d) ; introduce the grading | xi | = |ξi | = 1; || = 2. We
can identify g0 with h.
Consider the DGLA g g [1] with the differential (X, a) → (ad (a) , 0) . This
DGLA has two sub DGLAs: g g [1] and g 2
g [1]. A (g g[1], H)-module is
a DG module over g g[1] such that the action of h integrates to a representation
of H which is a direct product of finite-dimensional representations. Similarly for
9
g [1]-modules and g 2

g g [1]-modules. We denote the action of an element
(X, a) ∈ g g[1] by LX + La .
Let M be a symplectic manifold with the data (∇, R) as in Proposition 4.1.
Let (L• , ∂L ) be a g 2
g [1]-module. Let
L• = Psymp ×H L•
be the associated graded vector bundle on M (here Psymp is the bundle of symplectic
frames). Define the differential on Ω• (M, L• ) as follows. In local coordinates, if
∇ = dDR + A, the differential is defined as
dDR + LA + LR + ∂L .

Denote by C•Lie g 2 g [1] , h; L• the complex of relative Lie cochains, with the
differential given by the sum of Lie cohomology coboundary and ∂L .
Then we have a morphism of complexes

(4.6) GF : C•Lie g 2 g [1] , h; L• → Ω• (M, L• )
given by
GF (ϕ) = ϕ (A + R, A + R, . . . , A + R)
where we use the notations X = (X, 0) and a = (0, a) for elements of g
g[1].
Example 4.4. Let L• = C with the trivial action. Then Ω• (M, L• ) is the
de Rham complex of M .
−1
Example 4.5. Let L• = CC− −• W with the action of g g[1] as described
after Proposition 4.1. Without localizing in , CC− −• (W) is a (g g [1] , H)-
module. Then Ω• (M, L• ) becomes the right hand side in Proposition 4.3.

Example 4.6. Let O =C x d , ξ1 , . . . , ξd be the algebra of functions
1 , . . . , x
on a formal neighborhood of the origin; let
• = O d
Ω xd , dξ1 , . . . , dξd
x1 , . . . , d
with the differential dDR be the DGA of forms on this neighborhood. Define also the
Lie algebra of Hamiltonian vector fields on the formal neighborhood to be g0 = O/C
with the Poisson bracket defined by {
j } = 0, {ξi , ξj } = 0, {ξ
xi , x j } = δij . An
i, x
element f of O defines a derivation Xa = {a, ?} of O. Let L = Ω• , dDR with
•
the action of g0 g [1] defined by La + Lb = LXa + ιXb . The DGLA g [] acts on
L• via the morphism g g[1] → g0 g0 [1] induced by the map i
1
a → a.
To identify the complex Ω• (M, L• ) in this example introduce the following
notations. Let OM = Sym [[T ∗ ]] and Ω • = Sym [[T ∗ ]] ⊗ • T ∗ . The latter is a
M M M M
DGA with the differential dDR . The complex Ω• (M, L• ) can then be written as the

de Rham complex Ω• M, Ω • , ∇0 + dDR .
M
5. Some Lie algebra cohomology classes

5.1. The trace density. Let us start by recalling the following result (see [7]).
32
10 P.P.BRESSLER,
R.NEST, ANDB.TSYGAN
B. TSYGAN
Theorem 5.1. Let W denote,

as above, the Weyl algebra on 2d generators. The
Hochschild homology of W h−1 over C (()) is one-dimensional, with a generator
in degree 2d given by
1
c= 1 ⊗ x̂1 ∧ ξˆ1 ∧ · · · ∧ x̂d ∧ ξˆd .
(2d)!d
We denote by μ0 the associated quasi-isomorphism of complexes:

μ0 : C−• W(h) → C (()) [2d] ;
In particular,
the
spectral

sequence
associated to the filtration of the negative cyclic
complex CC− −• W (h) , b + uB associated to the filtration of

CC−−• = un C •
n≥0
by powers of u collapses, hence μ0 extends to a C [[u]]-linear quasi-isomorphism of

complexes
−

CC−• W(h) , b + uB → (C (()) [[u]]) [2d]

still denoted by μ0 .

We give the complexes CC− −• W(h) , b + uB and (C (()) [[u]]) [2d] the struc-
ture of (g g[1], H)-module as described in Examples 4.5 and 4.4. This induces a
(g g[1], H)-module structure on

(5.1) Hom•C[[u]](()) CC− −• W() , C (()) [[u]] [2d]
and we will denote the hypercohomology of the complex

(5.2) g[1], H; Hom•C[[u]](()) CC−
C•Lie g −• W() , C (()) [[u]] [2d]
by

H• g g[1], H; Hom•C[[u]](()) CC− −• W () , C (()) [[u]] [2d] .

Since the homologies of both CC− −• W(h) , b + uB and (C (()) [[u]]) [2d] are con-
centrated in a single dimension and (g g[1], H)-invariant, the cohomology of the
complex 5.1 is also one-dimensional and (g g[1], H)-invariant. As a result, using
the spectral sequence of the complex (5.2) induced by the filtration in the degrees
of the coefficient complex, it is immediate to see that μ0 extends to a generator

(5.3) μ ∈ H 0 g g[1], H; Hom•C[[u]](()) CC− −• W() , C (()) [[u]] [2d]
Definition 5.2. The class μ constructed above is called the trace density in
the Lie algebra cohomology. The trace density τ for a deformation of a gerbe is
defined as the composition
•

GF(μ ) •
τ : CC−
−• (A) → Ω M, CC−
−• (WM ) , ∇ + b + uB + LR −→ Ω (M ) [[u]] .
Here the first map is the quasi-isomorphism from Proposition 4.3.

11

5.2. The topological index map. Recall that O = C x̂1 , ξˆ1 , . . . , x̂d , ξˆd .

We give CC−
−• O the structure of (g g[1], H)-module as follows. An element

(X, a) ∈ g g[1] mod has a form X̃, a0 with X̃ ∈ Der O and a0 = a mod .

Its action on CC−
−• O is given by LX̃ + La0 where the second summand is the
shuffle product with 1 ⊗ a0 ,. The action of h integrates to the action of H (induced
by the linear transformations in the variables).
Let

σ : CC−−• (W) → CC −
−• O

be reduction modulo . This is a map of (g g[1], H)-modules (where the module

structure on the left hand side is defined in Example 4.5).
Let

• [[u]] , ud = O d
Ω xd , dξ1 , . . . , dξd [[u]] , ud
x1 , . . . , d
The Hochschild-Kostant-Rosenberg
denote the periodized de Rham complex of O.
morphism

μ : CC− •
−• O → Ω [[u]]
is a C [[u]]-linear morphism of complexes given by
1
a0 ⊗ a1 ⊗ · · · ⊗ an → a0 da1 · · · dan .
n!
Both σ and μ are (g g[1], H)-equivariant, hence the composition μtop = μ ◦ σ
gives a cohomology class

μtop ∈ H0 g g[1], H; Hom• CC− •
−• (W) , Ω [[u]]
Definition 5.3. The topological index map is the composition of morphisms

of complexes:
•

GF(μtop ) •
τ top : CC−
−• (A) → Ω M, CC−
−• (WM ) , ∇ + b + uB + LR −→ Ω (M ) [[u]] .
Here the first map is the quasi-isomorphism from Proposition 4.3.
Note that GF(μ) induces a quasi-isomorphism

GF(μ) : Ω• M, CC− ∗ •
−• (Sym [[TM ]]) , ∇0 + b + uB → Ω (M ) .
Composing this morphism with the quasi-isomorphism (4.4) gives a gerbe version
of the Hochschild-Kostant-Rosenberg quasi-isomorphism (cf. [23]):
τ HKR : CC− •
−• (A0 ) → Ω (M ) .
With this definition it is easy to see that the definition of the topological index map
given in this section agrees with the one given in equation (3.3).
34
12 P.P.BRESSLER,
R.NEST, ANDB.TSYGAN
B. TSYGAN
5.3. The Chern classes. Every H-invariant polynomial of degree p on h

induces a cohomology class in H 2p (g
g[1], H; C) as follows. Let A : g → h be an
h-invariant splitting of the embedding h → g. Extend A by zero to g g[1]. Then
the class corresponding to an invariant polynomial F is equal to
cF = F (Θ, . . . , Θ)
where
1
Θ = ∂A + [A, A]
2
(cf. [11]). Here ∂ is the full differential in the cochain complex of the DGLA
C•Lie (g
g[1], H; C), i.e.,the Chevalley-Eilenberg differential plus the one induced
by the differential on g g[1].
In particular, since the ad-action of h integrates to the action of H, we can apply
this construction to the Chern classes cp ∈ H 2p (g, h; C) for even p, which corre-
spond to the restrictions to sp (2d) of elementary symmetric polynomials, viewed
as invariant polynomials on gl (2d). Their images under the map GF are the Chern
classes of the tangent bundle.
5.4. The characteristic class. The Lie algebra extension (cf. (4.5))
1
0 → C → g̃ → g → 0
i

defines a cohomology class θ ∈ HLie 2 1

g, H; i C [[]] .

We extend it to a class θ̃ ∈ HLie 2

g g[1], H; i 1
C [[]] as follows. Since the
action of h on the coefficient algebra is trivial, one notes that

• 1 • 1
HLie g g [1] , H; C [[]] = HLie g g [1] , h; C [[]]
i i

• 1 1
= HLie
g g [1] , h ⊕ C [[]] ; C [[]]
i i
Then apply the construction from Section 5.3 above to the invariant polynomial of
degree one on h ⊕ i 1
C [[]] which is the projection to i 1
C [[]] .
Proposition 5.4. (see [3]) The image of θ̃ under the map GF is the charac-
teristic class θ from Theorem 2.3.
5.5. Algebraic index theorem in Lie algebra cohomology. Now we want
to compare the class μ from Section 5.1 to μtop from Section 5.2. For that, we
have to put them into the same cohomology group, to wit
−1
•
(5.4) H0 g 2 g[1], H; Hom•C((u,)) CCper
−• W
((u, ))
,Ω
• The map

g[1], H; Hom•C[[u]](()) CC−
H0 g −• W() , C (()) [[u]] [2d]
−1
•
H 0
g g[1], H; Hom•C((u,))
2
CCper
−• W ,
Ω ((u, ))
• (()) followed by localizing in u,

is given by embedding of C (()) into Ω
−d
multiplying by u and restricting to the subalgebra g 2 g[1];
13
• The map

g[1], H; Hom• CC−
H0 g •
−• (W) , Ω [[u]]
−1
•
H 0
g g[1], H; Hom•C((u,))
2
CCper
−• W
((u, ))
,Ω
is given by extending scalars from C to C (()) and again restricting to

the subalgebra g 2
g[1].
First an observation
Lemma 5.5. The restriction map
• •
HLie (g
g[1], H; C ((u, ))) → HLie (g, H; C ((u, )))
is injective, and H• (g
g[1] [1] , H; C ((u, ))) is the polynomial ring generated by
the Chern classes and θ̃, i.e.,

(5.5) H• (g
g[1], H; C ((u, ))) = C ((, u)) c2 , . . . , c2d , θ̃
Proof. The differential graded Lie algebra g̃ g̃ [1] is contractible, hence

•
HLie (g̃ g̃ [1] , H; C ((u, ))) = C ((, u)) [c2 , . . . , c2d ] .
The injectivity of the restriction map is immediate, since it has a left inverse induced
by the inclusion g → g g[1]. The Hochschild-Serre spectral sequence for the
1
inclusion i C [[]] → g̃ g̃ [1] degenerates at the E 2 -term to a long exact sequence
in cohomology associated to the short exact sequence of complexes
0 −→ C•Lie (g
g[1], H; C ((u, ))) −→ C•Lie (g̃ g̃ [1] , H; C ((u, )))
−→ C•−1
Lie (g
g[1], H; C ((u, ))) −→ 0.
The corresponding connecting homomorphism in cohomology
• •+2
∂ : HLie (g
g[1], H; C ((u, ))) → HLie (g
g[1], H; C ((u, )))
is given by multiplication by the class θ̃. The result of the Lemma follows.
6. Proof of Theorem 3.1

The main technical result is the following.
Theorem 6.1 (The algebraic index theorem in Lie algebra cohomol-
ogy). The equality

μ = μtop up θ̃
Ae
2p
p
holds in the cohomology group

−1
•
g[1], H; Hom•C((u,)) CCper
H0 g 2 −• W
((u, )) .
,Ω
Here A is the A-polynomial

of the Chern classes from Section 5.3,
θ̃ is the class from

Section 5.4, and the subscript 2p means the component in H 2p g 2 g[1], h; C .
36
14 P.P.BRESSLER,
R.NEST, ANDB.TSYGAN
B. TSYGAN
Proof. Since u is invertible, 1 → 1 defines an equivariant quasi-isomorphism

• ((u, )) ,
C ((u, )) Ω
and hence

g[1], H; Hom•C((u,)) CCper
H• g 2
−1 • ((u, ))
−• W [] ,Ω

H• g 2 g[1], H; C ((u, )) .

Moreover, since is invertible, it is also isomorphic to

(6.1) H• (g g[1], H; C ((u, ))) = uk H •+2k (g
g[1], H; C (())) .
k>−∞

The components of μ and μ in H (g g [1] , H; C ((u, ))) coincide and are equal
top 0
to the class of 1; hence, by above lemma, μ is a product of μtop and a class α in

H0 (g g [1] , H; C ((u, ))). By Theorem 3.3.1 in [7], the restriction of this class to
p θ
H (g, H; C ((u, ))) is equal to the restriction of the class
0
u
Ae . But
p
2p
since this restriction is injective, the
claimed equality follows.
Once again, note that we have A instead of A
as in [7] because our definition
of the Chern classes is different. Here, cp is the invariant polynomial of degree p on
sp (2d) which is the restriction of the elementary symmetric polynomial on gl (2d) .
In [7], cp was the invariant polynomial on sp (2d) whose restriction to gl (d) is the
elementary symmetric polynomial of degree p.
6.1. Proof of the main result. We just need to collect the pieces.
• τ , the trace map on the cyclic periodic complex (see Section 5.1) is
the composition of the (periodic version of
the) quasi-isomorphism from
Proposition 4.3 with the morphism GF μ .
• Similarly, τ top (see Section 5.2) is the composition of the (periodic ver-
sion of the) quasi-isomorphism from Proposition 4.3 with the morphism
GF (μtop ).
Theorem 3.1 then follows immediately from Theorem 6.1 by applying the map GF.
References
[1] F. Bayen, M. Flato, C. Fronsdal, A. Lichnerowicz, and D. Sternheimer. Deformation theory
and quantization. I. Deformations of symplectic structures. Ann. Physics, 111(1):61–110,
1978.
[2] L. Breen. On the classification of 2-gerbes and 2-stacks. Astérisque, (225):160, 1994.
[3] P. Bressler, A. Gorokhovsky, R. Nest, and B. Tsygan. Deformation quantization of gerbes.
Adv. Math., 214(1):230–266, 2007.
[4] P. Bressler, A. Gorokhovsky, R. Nest, and B. Tsygan. Deformations of Azumaya algebras.
In Proceedings of the XVIth Latin American Algebra Colloquium (Spanish), Bibl. Rev. Mat.
Iberoamericana, pages 131–152. Rev. Mat. Iberoamericana, Madrid, 2007.
[5] P. Bressler, A. Gorokhovsky, R. Nest, and B. Tsygan. Chern character for twisted complexes.
In Geometry and dynamics of groups and spaces, volume 265 of Progr. Math., pages 309–324.
Birkhäuser, Basel, 2008.
[6] P. Bressler, A. Gorokhovsky, R. Nest, and B. Tsygan. Deformations of gerbes on smooth
manifolds. In K-theory and noncommutative geometry, EMS Ser. Congr. Rep., pages 349–
392. Eur. Math. Soc., Zürich, 2008.
[7] P. Bressler, R. Nest, and B. Tsygan. Riemann-Roch theorems via deformation quantization.
I, II. Adv. Math., 167(1):1–25, 26–73, 2002.
15
[8] J.-L. Brylinski. Loop spaces, characteristic classes and geometric quantization, volume 107
of Progress in Mathematics. Birkhäuser Boston Inc., Boston, MA, 1993.
[9] A. D’Agnolo and P. Polesello. Stacks of twisted modules and integral transforms. In Geometric
aspects of Dwork theory. Vol. I, II, pages 463–507. Walter de Gruyter GmbH & Co. KG,
Berlin, 2004.
[10] B. V. Fedosov. A simple geometrical construction of deformation quantization. J. Differential
Geom., 40(2):213–238, 1994.
[11] D. B. Fuks. Cohomology of infinite-dimensional Lie algebras. Contemporary Soviet Mathe-
matics. Consultants Bureau, New York, 1986. Translated from the Russian by A. B. Sosinskiı̆.
[12] I. M. Gelfand and D. A. Kazhdan. Certain questions of differential geometry and the com-
putation of the cohomologies of the Lie algebras of vector fields. Dokl. Akad. Nauk SSSR,
200:269–272, 1971.
[13] M. Gerstenhaber. On the deformation of rings and algebras. Ann. of Math. (2), 79:59–103,
1964.
[14] M. Kashiwara. Quantization of contact manifolds. Publ. Res. Inst. Math. Sci., 32(1):1–7,
1996.
[15] M. Kashiwara and P. Schapira. Categories and sheaves, volume 332 of Grundlehren der Math-
ematischen Wissenschaften [Fundamental Principles of Mathematical Sciences]. Springer-
Verlag, Berlin, 2006.
[16] M. Kashiwara and P. Schapira. Deformation quantization modules I:Finiteness and duality.
ArXiv e-prints, February 2008, 0802.1245.
[17] M. Kashiwara and P. Schapira. Deformation quantization modules II. Hochschild class. ArXiv
e-prints, September 2008, 0809.4309.
[18] M. Kashiwara and P. Schapira. Deformation quantization modules. ArXiv e-prints, March
2010, 1003.3304.
[19] M. Kontsevich. Deformation quantization of algebraic varieties. Lett. Math. Phys., 56(3):271–
294, 2001. EuroConférence Moshé Flato 2000, Part III (Dijon).
[20] M. Kontsevich. Deformation quantization of Poisson manifolds, I. Lett. Math. Phys., 66:157–
216, 2003, arXiv:q-alg/9709040.
[21] J.-L. Loday. Cyclic homology, volume 301 of Grundlehren der Mathematischen Wis-
senschaften [Fundamental Principles of Mathematical Sciences]. Springer-Verlag, Berlin, sec-
ond edition, 1998.
[22] V. Mathai, R. B. Melrose, and I. M. Singer. Fractional analytic index. J. Differential Geom.,
74(2):265–292, 2006.
[23] V. Mathai and D. Stevenson. On a generalised Connes-Hochschild-Kostant-Rosenberg theo-
rem. Adv. Math., 200(2):303–335, 2006.
[24] R. Nest and B. Tsygan. Algebraic index theorem. Comm. Math. Phys., 172(2):223–262, 1995.
[25] R. Nest and B. Tsygan. Product structures in (cyclic) homology and their applications.
In Operator algebras and quantum field theory (Rome, 1996), pages 416–439. Int. Press,
Cambridge, MA, 1997.
[26] P. Polesello. Classification of deformation quantization algebroids on complex symplectic
manifolds. Publ. Res. Inst. Math. Sci., 44(3):725–748, 2008.
[27] P. Polesello and P. Schapira. Stacks of quantization-deformation modules on complex sym-
plectic manifolds. Int. Math. Res. Not., (49):2637–2664, 2004.
[28] M. Schlessinger and J. Stasheff. The Lie algebra structure of tangent cohomology and defor-
mation theory. J. Pure Appl. Algebra, 38(2-3):313–322, 1985.
[29] J.-L. Tu, P. Xu, and C. Laurent-Gengoux. Twisted K-theory of differentiable stacks. Ann.
Sci. École Norm. Sup. (4), 37(6):841–910, 2004.
[30] A. Vistoli. Grothendieck topologies, fibered categories and descent theory. In Fundamental
algebraic geometry, volume 123 of Math. Surveys Monogr., pages 1–104. Amer. Math. Soc.,
Providence, RI, 2005.
38
16 P.P.BRESSLER,
R.NEST, ANDB.TSYGAN
B. TSYGAN
Max-Planck-Institut für Mathematik, Vivatsgasse 7, 53111 Bonn, Germany

E-mail address: paul.bressler@gmail.com
Department of Mathematics, UCB 395, University of Colorado, Boulder, CO 80309-

0395, USA
E-mail address: Alexander.Gorokhovsky@colorado.edu
Department of Mathematics, Copenhagen University, Universitetsparken 5, 2100

Copenhagen, Denmark
E-mail address: rnest@math.ku.dk
Department of Mathematics, Northwestern University, Evanston, IL 60208-2730,

USA
E-mail address: tsygan@math.northwestern.edu
Volume 546, 2011
Index theory for basic Dirac operators

on Riemannian foliations
Jochen Brüning, Franz W. Kamber, and Ken Richardson
This paper is dedicated to Henri Moscovici at the occasion of his 65th anniversary
Abstract. In this paper we prove a formula for the analytic index of a basic
Dirac-type operator on a Riemannian foliation, solving a problem that has
been open for many years. We also consider more general indices given by
twisting the basic Dirac operator by a representation of the orthogonal group.
The formula is a sum of integrals over blowups of the strata of the foliation and
also involves eta invariants of associated elliptic operators. As a special case, a
Gauss-Bonnet formula for the basic Euler characteristic is obtained using two
independent proofs.
Contents
1. Introduction
2. Riemannian foliations and basic Dirac operators
3. Fredholm properties and equivariant theory
4. Desingularization of the foliation
5. The equivariant index theorem
6. The basic index theorem
7. The representation-valued basic index theorem
8. The basic index theorem for foliations given by suspension
9. An example of transverse signature
10. The Basic Euler characteristic
References
2010 Mathematics Subject Classification. Primary: 53C12, 57R30, 58G10, secondary: 58C40,
58D19, 58J28.
Key words and phrases. foliation, basic, index, transversally elliptic.
This work was supported by Deutsche Forschungsgemeinschaft in the Collaborative Research
Centre 647.
2011
c 0000
c Mathematical
American (copyright Society
holder)
1
39
40
2 J. BRÜNING, F. W. KAMBER, AND K. RICHARDSON
1. Introduction
Let (M, F) be a smooth, closed manifold endowed with a Riemannian foliation.
Let DbE : Γb (M, E + ) → Γb (M, E − ) be a basic, transversally elliptic differential
operator
acting on the basic sections of a foliated vector bundle E. The basic index
indb DbE is known to be a well-defined integer, and it has been an open problem
since the 1980s to write this integer in terms of geometric
and topological invariants.
Our main theorem (Theorem 6.1) expresses indb DbE as a sum of integrals over
the different strata of the Riemannian foliation, and it involves the eta invariant
of associated equivariant elliptic operators on spheres normal to the strata. The
result is

+
r
indb DbE = A0,b (x) |dx| β (Mj ) ,

M 0 F j=1
1 1
β (Mj ) = −η D S+,τ
+ h D S+,τ .
Aτj,b (x) |dx|
j j
2 τ nτ rank W τ Mj F
The notation will be explained later; the integrands A0,b (x) and Aτj,b (x) are the
familar Atiyah-Singer integrands corresponding to local heat kernel supertraces of
induced elliptic operators over closed manifolds. Even in the case when the operator
D is elliptic, this result was not known previously. We emphasize that every part
of the formula is explicitly computable from local information provided by the
operator and foliation. Even the eta invariant of the operator DjS+,τ on a sphere
is calculated directly from the principal transverse symbol of the operator DbE at
one point of a singular stratum. The de Rham operator provides an important
example illustrating the computability of the formula, yielding the basic Gauss-
Bonnet Theorem (Theorem 10.1).
This new theorem is proved by first writing indb DbE as the invariant index of a
G-equivariant, transversally elliptic operator D on a G-manifold W associated to the
foliation, where G is a compact Lie group of isometries. Using our equivariant index
theorem in [14], we obtain an expression for this index in terms of the geometry
and topology of W and then rewrite this formula in terms of the original data on
the foliation.
We note that a recent paper of Gorokhovsky and Lott addresses this transverse
index question on Riemannian foliations in a very special case. Using a different
technique, they prove a formula for the index of a basic Dirac operator that is
distinct from our formula, in the case where all the infinitesimal holonomy groups
of the foliation are connected tori and if Molino’s commuting sheaf is abelian and
has trivial holonomy (see [27]). Our result requires at most mild topological as-
sumptions on the transverse structure of the strata of the Riemannian foliation. In
particular, the Gauss-Bonnet Theorem for Riemannian foliations (Theorem 10.1)
is a corollary and requires no assumptions on the structure of the Riemannian
foliation.
The paper is organized as follows. The definitions of the basic sections, holonomy-
equivariant vector bundles, basic Clifford bundles, and basic Dirac-type operators
are given in Section 2. In Section 3, we describe the Fredholm properties of the basic
index and show how to construct the G-manifold W and the G-equivariant operator
D, using a generalization of Molino theory [48]. We also use our construction to
obtain asymptotic expansions and eigenvalue asymptotics of transversally elliptic
INDEX THEORY ON RIEMANNIAN FOLIATIONS 41
3
operators on Riemannian foliations in Section 3.2, which is of independent inter-

est. In Section 3.4, we construct bundles associated to representions of the isotropy
subgroups of the G-action; these bundles are used in the main theorem. In Section
4, we describe a method of cutting out tubular neighborhoods of the singular strata
of the foliation and doubling the remainder to produce a Riemannian foliation with
fewer strata. We also deform the operator and metric and determine the effect
of this desingularization operation on the basic index. We recall the equivariant
index theorem in [14] in Section 5 and prove the basic index theorem in Section
6. Finally, we prove a generalization of this theorem to representation-valued basic
indices in Section 7.
We illustrate the theorem with a collection of examples. These include foliations
by suspension (Section 8), a transverse signature (Section 9), and the basic Gauss-
Bonnet Theorem (Section 10).
One known application of our theorem is Kawasaki’s Orbifold Index Theorem
([39], [40]). It is known that every orbifold is the leaf space of a Riemannian
foliation, where the leaves are orbits of an orthogonal group action such that all
isotropy subgroups have the same dimension. In particular, the contributions from
the eta invariants in our transverse signature example (Section 9) agree exactly
with the contributions from the singular orbifold strata when the orbifold is four-
dimensional.
We thank James Glazebrook, Efton Park and Igor Prokhorenkov for help-
ful discussions. The authors would like to thank variously the Mathematisches
Forschungsinstitut Oberwolfach, the Erwin Schrödinger International Institute for
Mathematical Physics (ESI), Vienna, the Department for Mathematical Sciences
(IMF) at Aarhus University, the Centre de Recerca Matemàtica (CRM), Barcelona,
and the Department of Mathematics at TCU for hospitality and support during the
preparation of this work.
2. Riemannian foliations and basic Dirac operators

2.1. Basic definitions. A foliation of codimension q on a smooth manifold
M of dimension n is a natural generalization of a submersion. Any submersion
f : M → N with fiber dimension p induces locally, on an open set U ⊂ M , a
diffeomorphism φ : U → Rq × Rp (y, x), where p + q = n. A foliation F is
a (maximal) atlas {φα : Uα → Rq × Rp } of M such that the transition functions
φα ◦ φ−1
β : R × R → R × R preserve the fibers, i.e. they have the form
q p q p
φα ◦ φ−1
β (y, x) = (ταβ (y) , ψαβ (x, y)) .
This local description has many equivalent formulations, as expressed in the famous
Frobenius Theorem. Geometrically speaking, M is partitioned into p-dimensional
immersed submanifolds called the leaves of the foliation; the tangent bundle T F
to the leaves forms an integrable subbundle of the tangent bundle T M .
In the case of a submersion, the normal bundle to T F is naturally identified with
the tangent bundle of the base, which then forms the space of leaves. In general,
such a description is not possible, since the space of leaves defined by the obvious
equivalence relation does not form a manifold. Nevertheless, reasonable transverse
geometry can be expressed in terms of the normal bundle Q := T M T F of the
foliation. We are particularly interested in the case of a Riemannian foliation, which
generalizes the concept of a Riemannian submersion. That is, the horizontal metric
42
gh on the total space of a Riemannian submersion is the pullback

of the metric
∂ ∂
on the base, such that in any chart φ as above, gh ∂yi , ∂yi depends on the base
coordinates y alone. Another way to express this is that LX gh = 0 for all vertical
vector fields X, where LX denotes

the Lie derivative. In the case of a foliation, the
q
∂
normal bundle Q is framed by ∂yj , and this foliation is called Riemannian if
j=1
it is equipped with a metric gQ on Q such that LX gQ = 0 for all X ∈ C ∞ (M, T F)
(see [48], [58]). For example, a Riemannian foliation with all leaves compact is a
(generalized) Seifert fibration; in this case the leaf space is an orbifold ([48, Section
3.6]). Or if a Lie group of isometries of a Riemannian manifold has orbits of constant
dimension, then the orbits form a Riemannian foliation. A large class of examples
of Riemannian foliations is produced by suspension (see Section 8).
Consider the exact sequence of vector bundles
π
0 → T F → T M → Q → 0.
The Bott connection ∇Q on the normal bundle Q is defined as follows. If

s ∈ C ∞ (Q) and if π (Y ) = s, then ∇Q X s = π ([X, Y ]). The basic sections of Q
are represented by basic vector fields, fields whose flows preserve the foliation.
Alternately, a section V of Q is called a basic vector field if for every X ∈ C ∞ (T F),
[X, V ] ∈ C ∞ (T F) (see [33] or [48]).
A differential form ω on M is basic if locally it is a pullback of a form on
the base. Equivalently, ω is basic if for every vector field X tangent to the leaves,
iX ω = 0 and iX (dω) = 0, where iX denotes interior product with X. If we extend
∗ ∗
the Bott connection to a connection ∇Λ Q on Λ∗ Q∗ , a section ω of Λ∗ Q∗ is basic
Λ∗ Q ∗
if and only if ∇X ω = 0 for all X tangent to F. The exterior derivative of a basic
form is again basic, so the basic forms are a subcomplex Ω∗ (M, F) of the de Rham
complex Ω∗ (M ). The cohomology of this subcomplex is the basic cohomology
H ∗ (M, F).
2.2. Foliated vector bundles. We now review some standard definitions

(see [33] and [48]). Let G be a compact Lie group. With notation as above, we
say that a principal G–bundle P → (M, F) is a foliated principal bundle if it
is equipped with a foliation FP (the lifted foliation) such that the distribution
T FP is invariant under the right action of G, is transversal to the tangent space to
the fiber, and projects to T F. A connection ω on P is called adapted to FP if
the associated horizontal distribution contains T FP . An adapted connection ω is
called a basic connection if it is basic as a g-valued form on (P, FP ). Note that in
[33] the authors showed that basic connections always exist on a foliated principal
bundle over a Riemannian foliation.
Similarly, a vector bundle E → (M, F) is foliated if E is associated to a foliated
principal bundle P → (M, F) via a representation ρ from G to O (k) or U (k). Let
Ω (M, E) denote the space of forms on M with coefficients in E. If a connection
form ω on P is adapted, then we say that an associated covariant derivative operator
∇E on Ω (M, E) is adapted to the foliated bundle. We say that ∇E is a basic
2
connection on E if in addition the associated curvature operator ∇E satisfies
2
iX ∇E = 0 for every X ∈ T F, where iX denotes the interior product with X.
Note that ∇E is basic if ω is basic.
5
Let C ∞ (E) denote the smooth sections of E, and let ∇E denote a basic con-
nection on E. We say that a section s : M → E is a basic section if and only if
∞
∇EX s = 0 for all X ∈ T F. Let Cb (E) denote the space of basic sections of E. We
will make use of the fact that we can give E a metric such that ∇E is a metric
basic connection.
The holonomy groupoid GF of (M, F) (see [59]) is the set of ordered triples
(x, y, [γ]), where x and y are points of a leaf L and [γ] is an equivalence class of
piecewise smooth paths in L starting at x and ending at y; two such paths α and
β are equivalent if and only if β −1 α has trivial holonomy. Multiplication is defined
by (y, z, [α]) · (x, y, [β]) = (x, z, [αβ]), where αβ refers to the curve starting at x and
ending at z that is the concatenation of β and α. Because (M, F) is Riemannian,
GF is endowed with the structure of a smooth (n + p)–dimensional manifold (see
[59]), where n is the dimension of M and p is the dimension of the foliation.
We say that a vector bundle E → M is GF –equivariant if there is an action
of the holonomy groupoid on the fibers. Explicitly, if the action of g = (x, y, [γ]) is
denoted by Tg , then Tg : Ex → Ey is a linear transformation. The transformations
{Tg } satisfy Tg Th = Tg·h for every g, h ∈ GF for which g · h is defined, and we
require that the map g −→ Tg is smooth. In addition, we require that for any unit
u = (x, x, [α]) (that is, such that the holonomy of α is trivial), Tu : Ex → Ex is the
identity.
We say that a section s : M → E is holonomy–invariant if for every g =
(x, y, [γ]) ∈ GF , Tg s (x) = s (y).
Remark 2.1. Every GF –equivariant vector bundle E → (M, F) is a foliated vector
bundle, because the action of the holonomy groupoid corresponds exactly to parallel
translation along the leaves. If the partial connection is extended to a basic con-
nection on E, we see that the notions of basic sections and holonomy–invariant
sections are the same.
On the other hand, suppose that E → (M, F) is a foliated vector bundle
that is equipped with a basic connection. It is not necessarily true that paral-
lel translation can be used to give E the structure of a GF –equivariant vector
bundle. For example, let α be an irrational multiple of 2π, and consider E =
[0, 2π] × [0, 2π] × C (0, θ, z) ∼ 2π, θ, eiα z , which is a Hermitian line bundle over
the torus S 1 × S 1 , using the obvious product metric. The natural flat connection for
E over the torus is a basic connection for the product foliation F = {Lθ }, where
Lθ = (φ, θ) | φ ∈ S 1 . However, one can check that parallel translation cannot be
used to make a well-defined action of GF on the fibers.
An example of a GF –equivariant vector bundle is the normal bundle Q, given
by the exact sequence of vector bundles
π
0 → T F → T M → Q → 0.
The Bott connection ∇Q on Q is a metric basic connection. (Recall that if s ∈
C ∞ (Q) and if π (Y ) = s, then ∇Q X s = π ([X, Y ]).) The basic sections of Q are
represented by basic vector fields, fields whose flows preserve the foliation. Al-
ternately, a section V of Q is called a basic vector field if for every X ∈ C ∞ (T F),
[X, V ] ∈ C ∞ (T F) (see [33] or [48]).
Lemma 2.2. Let E → (M, F) be a foliated vector bundle with a basic connection
∇E . Let V ∈ C ∞ (Q) be a basic vector field, and let s : M → E be a basic section.
Then ∇EV s is a basic section of E.
44
Proof. For any X ∈ C ∞ (T F), [X, V ] ∈ C ∞ (T F), so that ∇E

X s = ∇[X,V ] s =
E
0. Thus,

∇E ∇
X V
E
s = ∇ ∇
E E
X V − ∇ ∇
E E
V X − ∇ E
[X,V ] s
E 2
= ∇ (X, V ) s = 0,
since ∇E is basic.

Another example of a foliated vector bundle is the exterior bundle Q∗ ; the
induced connection from the Bott connection on Q is a metric basic connection.
The set of basic sections of this vector bundle is the set of basic forms Ω (M, F),
which is defined in the ordinary way in Section 2.1. It is routine to check that these
two definitions of basic forms are equivalent.
2.3. Basic Clifford bundles. Identifying Q with the normal bundle of the
Riemannian foliation (M, F), we form the bundle of Clifford algebras Cl (Q) =
Cl (Q) ⊗ C over M .
Definition 2.3. Let E be a bundle of Cl (Q) –modules over a Riemannian foliation
(M, F). Let ∇ denote the Levi–Civita connection on M, which restricts to a metric
basic connection on Q. Let h = (·, ·) be a Hermitian metric on E, and let ∇E be a
connection on E. Let the action of an element ξ ∈ Cl (Qx ) on v ∈ Ex be denoted
by c (ξ) v. We say that E, h, ∇E is a basic Clifford bundle if
(1) The bundle E → (M, F) is foliated.
(2) The connection ∇E is a metric basic connection.
(3) For every ξ ∈ Qx , c (ξ) is skew-adjoint on Ex .
(4) For every X ∈ C ∞ (T M ) , Y ∈ C ∞ (Q) , and s ∈ C ∞ (E) ,
∇E
X (c (Y ) s) = c (∇X Y ) s + c (Y ) ∇X (s) .
E

Lemma 2.4. Let E, h, ∇E be a basic Clifford module over (M, F). Let V ∈
C ∞ (Q) be a basic vector field, and let s : M → E be a basic section. Then c (V ) s
is a basic section of E.
∞
Proof. If ∇E
X s = 0 and ∇X V = 0 for every X ∈ C (T F), then
X (c (V ) s) = c (∇X V ) s + c (V ) ∇X (s) = 0.
∇E E

2.4. Basic Dirac operators.

Definition 2.5. Let E, (·, ·) , ∇E be a basic Clifford bundle. The transversal
E
Dirac operator Dtr is the composition of the maps
(∇E )tr ∼
C ∞ (E) → C ∞ (Q∗ ⊗ E) → C ∞ (Q ⊗ E) → C ∞ (E) ,
= c
tr
where the operator ∇E is the obvious projection of ∇E : C ∞ (E) → C ∞ (T ∗ M ⊗ E)
and the ismorphism ∼= is induced via the holonomy–invariant metric on Q.
If {e1 , ..., eq } is an orthonormal basis of Q, we have that

q
E
Dtr = c (ej ) ∇E
ej .
j=1
7
∗
pE : T ∗M → M be ∗the projection.
Let The restriction
E of the principal symbol
σ Dtr : T M → End (p E) to Q∗ is denoted σ tr Dtr , and it is given by
E
σ tr Dtr (ξ) = c ξ # .
Since this map is invertible for ξ ∈ Q∗ \ 0 , we say that Dtr

E
is transversally
elliptic.
E
Lemma 2.6. The operator Dtr restricts to a map on the subspace Cb∞ (E).
Proof. Suppose that s : M → E is a basic section, so that ∇E

X s = 0 for every
X ∈ C ∞ (T F). Near a point x of M, choose an orthonormal frame field (e1 , ..., eq )
of Q consisting of basic fields. Then
E
q
∇E
X Dtr (s) = ∇E
X c (ej ) ∇E
ej s
j=1

q
= X ∇ej s ,
c (ej ) ∇E E
j=1
since each ej is basic, and the result is zero by Lemma 2.2.
We now calculate the formal adjoint of Dtr E

on Cb∞ (E). Letting (s1 , s2 ) denote
the pointwise inner product of sections of E and choosing an orthonormal frame
field (e1 , ..., eq ) of Q consisting of basic fields, we have that
q

E
Dtr s1 , s2 − s1 , Dtr
E
s2 = ej s1 , s2 − s1 , c (ej ) ∇ej s2
c (ej ) ∇E E
j=1
q
= c (ej ) ∇E
ej s1 , s2 + c (ej ) s1 , ∇ej s2
E
j=1
q
⊥
= ∇E
ej (c (ej ) s 1 ) , s 2 − c ∇ej ej s 1 , s 2
j=1

+ c (ej ) s1 , ∇Eej s 2
⎛ ⎞ ⎛ ⎛ ⎞ ⎞
q q
= ⎝ ∇⊥ ej (c (ej ) s1 , s2 )
⎠ − ⎝c ⎝ ∇⊥
ej ej
⎠ s1 , s2 ⎠
j=1 j=1
⎛ ⎞

q q
= − ∇⊥
ej iej ω + ω
⎝ ∇⊥
ej ej
⎠,
j=1 j=1
46
where ω is the basic form defined by ω (X) = − (c (X) s1 , s2 ) for X ∈ C ∞ (Q).

Continuing,
⎛ ⎞
E
q q
E
s2 = − ∇⊥
ej iej ω + ω
⎝ ∇⊥ej ej
⎠
j=1 j=1
⎛ ⎞
q
q
= − iej ∇⊥
ej + i∇⊥
e ej
ω+ω⎝ ∇⊥
ej ej
⎠
j
j=1 j=1

q
= − iej ∇⊥
ej ω
j=1
Note we have been using the normal Levi-Civita connection ∇⊥ . If we (locally)

complete the normal frame field to an orthonormal frame field {e1 , ..., en } for T M
near x ∈ M . Letting ∇M = ∇⊥ + ∇tan be the Levi-Civita connection on Ω (M ),
the divergence of a general basic one-form β is
n
δβ = − iej ∇M
ej β
j=1

n
n
= − iej ∇⊥
ej β +− iej ∇tan
ej β
j=1 j=1

q
n
= − iej ∇⊥
ej β + − iej ∇tan
ej β
j=1 j>q
q ∗
Letting β = k=1 βk ek , then each βk is basic and
q
q
n
δβ = − iej ∇⊥
ej β − iej ∇tan
ej βk e∗k
j=1 j>q k=1

q
q
n
= − iej ∇⊥
ej β − βk iej ∇tan ∗
ej (ek )
j=1 k=1 j>q

q
q
= − iej ∇⊥
ej β − βk iej ∇M ∗ ∗
ej (ek ) , em em
∗
j=1 k=1 j>q m>q

q
q
= − iej ∇⊥
ej β + βk iej ∇M
ej (e∗
m ) , e∗
k em
∗
j=1 k=1 j>q m>q

q
q ∗ ∗
= − iej ∇⊥
ej β + βk ∇ej ej , ek
M
j=1 k=1 j>q m>q
q
= − iej ∇⊥
ej β + iH β,
j=1
where H is the mean curvature vector field of the foliation. Thus, for every basic
one-form β,
q
− iej ∇⊥
ej β = δβ − iH β.
j=1
9
Applying this result to the form ω defined above, we have

E
q
E
s2 = − iej ∇⊥
ej ω
j=1
= δω − iH ω
= δω + (c (H) s1 , s2 )
= δω − (s1 , c (H) s2 )
Next, letting P : L2 (Ω (M )) → L2 (Ωb (M, F)) denote the orthogonal projection
onto the closure of basic forms in L2 (Ω (M )), we observe that δb = P δ is the adjoint
of db , the restriction of the exterior derivative to basic forms. Using the results of
[49], P maps smooth forms to smooth basic forms, and the projection of the smooth

function (s1 , c (H) s2 ) is simply (s1 , c (Hb ) s2 ) , where Hb is the vector field P H ,
the basic projection of the mean curvature vector field. If we had originally chosen
our bundle-like metric to have basic mean curvature, which is always possible by
[18], then Hb = H. In any case, the right hand side of the formula above is a basic
function, so that
E
Dtr s1 , s2 − s1 , Dtr E
s2 = δb ω − (s1 , c (Hb ) s2 ) .
We conclude:
E ∗
Proposition 2.7. The formal adjoint of the transversal Dirac operator is Dtr =
E
Dtr − c (Hb ).
Definition
2.8.
The basic Dirac operator associated to a basic Clifford module
E, (·, ·) , ∇E over a Riemannian foliation (M, F) with bundle-like metric is
1
DbE = Dtr E
− c (Hb ) : Cb∞ (E) → Cb∞ (E) .
2
E ∗
Remark 2.9. Note that the formal adjoint of DbE is Dtr + 12 c (Hb ) = Dtr
E
−
1 E E
2 c (H b ) = Db . Thus, Db is formally sel-adjoint. In [28], the researchers showed
that the eigenvalues of DbE are independent of the choice of the bundle-like metric
that restricts to the given transverse metric of the Riemannian foliation.

Let E, (·, ·) , ∇E be a basic Clifford bundle over the Riemannian foliation
(M, F) , and let DbE : Cb∞ (E) → Cb∞ (E) be the associated basic Dirac operator.
∗
Assume that E = E + ⊕ E − , with DbE± : Cb∞ (E ± ) → Cb∞ (E ∓ ). Let DbE± :
L2b (E ∓ ) → L2b (E ± ) denote the adjoint of DbE± .
Definition 2.10. The analytic basic index of DbE is
∗
indb DbE = dim ker DbE+ L2 (C ∞ (E + )) − dim ker DbE− .
b L2 (Cb∞ (E − ))
Remark 2.11. At this point, it is not clear that these dimensions are finite. We
demonstrate this fact inside this section.
2.5. Examples. The standard examples of ordinary Dirac operators are the
spinc Dirac operator, the de Rham operator, the signature operator, and the Dol-
beault operator. Transversally elliptic analogues of these operators and their cor-
responding basic indices are typical examples of basic Dirac operators.
Suppose that the normal bundle Q = T M T F → M of the Riemannian foli-
ation (M, F) is spinc . Then there exists a foliated Hermitian basic Clifford bundle
48

S, (·, ·) , ∇S over M such that for all x ∈ M , Sx is isomorphic to the standard
spinor representation of the Clifford algebra Cl (Qx ) (see [42]). The associated
S∂b is called a basic spin Dirac operator. The meaning
S c
basic Dirac operator
of the integer indb ∂b is not clear, but it is an obstruction to some transverse
curvature and other geometric conditions (see [26], [30], [41], [28]).
Suppose F has codimension q. The basic Euler characteristic is defined as

q
χ (M, F) = (−1)k dim H k (M, F) ,
k=0
provided that all of the basic cohomology groups H k (M, F) are finite-dimensional.
Although H 0 (M, F) and H 1 (M, F) are always finite-dimensional, there are foli-
ations for which higher basic cohomology groups can be infinite-dimensional. For
example, in [25], the author gives an example of a flow on a 3-manifold for which
H 2 (M, F) is infinite-dimensional. There are various proofs that the basic coho-
mology of a Riemannian foliation on a closed manifold is finite-dimensional; see for
example [22] for the original proof using spectral sequence techniques or [37] and
[49] for proofs using a basic version of the Hodge theorem.
It is possible to express the basic Euler characteristic as the index of an oper-
ator. Let db denote the restriction of the exterior derivative d to basic forms over
the Riemannian foliation (M, F) with bundle-like metric, and let δb be the adjoint
of db . It can be shown that δb is the restriction of the operator P δ to basic forms,
where δ is the adjoint of d on all forms and P is the L2 -orthogonal projection
of the space of forms onto the space of basic forms. For general foliations, this is
not a smooth operator, but in the case of Riemannian foliations, P maps smooth
forms to smooth basic forms (see [49]), and P δ is a differential operator. In perfect
analogy to the fact that the index of the de Rham operator
d + δ : Ωeven (M ) → Ωodd (M )
is the ordinary Euler characteristic, it can be shown that the basic index of the
differential operator d + P δ, that is the index of
Db = db + δb : Ωeven
b (M, F) → Ωodd
b (M, F) ,
is the basic Euler characteristic. The same proof works; this time we must use
the basic version of the Hodge theorem (see [22], [37], and [49]). Note that the
equality of the basic index remains valid for nonRiemannian foliations; however, the
Fredholm property fails in many circumstances. It is interesting to note that the
operator db +δb fails to be transversally elliptic in some examples of nonRiemannian
foliations.
The principal symbol of Db is as follows. We define the Clifford multiplication
of Cl (Q) on the bundle ∧∗ Q∗ by the action

v· = v ∧ − (v)
for any vector v ∈ N F ∼= Q. With the standard connection and inner product
defined by the metric on Q, the bundle ∧∗ Q∗ is a basic Clifford bundle. The
corresponding basic Dirac operator, called the basic de Rham operator on basic
forms, satisfies
1 1
Db = d + δb − (κb ∧ +κb ) = Db − (κb ∧ +κb ) .
2 2
11
The kernel of this operator represents the twisted basic cohomology classes, the
cohomology of basic forms induced by the differential d defined as
1
d = d − κb ∧ .
2
See [29] for an extended discussion of twisted basic cohomology, the basic de Rham
operator, and its properties. We have indb (Db ) = indb (Db ) because they differ by
a zeroth order operator (see the Fredholm properties of the basic index in Section
3.1 below), and thus
indb (Db ) = indb (Db ) = χ (M, F) ,
the basic Euler characteristic of the complex of basic forms.
3. Fredholm properties and equivariant theory

3.1. Molino theory and properties of the basic index. Let M −→p
M
denote the principal bundle of ordered pairs of frames (φx , ψx ) over x ∈ M , where
φx : Rq → Nx F is an isometry and ψ : Ck → Ex is a complex isometry. This is a
principal G–bundle, where G ∼ = O (q) × U (k), and it comes equipped with a natural
metric connection ∇ associated to the Riemannian and Hermitian structures of
E → M . The foliated vector bundles Q → (M, F) and E → (M, F) naturally give
the structure of a foliated principal bundle with lifted foliation F.
M Transferring
the normalized, biinvariant metric on G to the fibers and using the connection ∇,
we define a natural metric (·, ·)M
on M that is locally a product. The connection
∗
∇ pulls back to a basic connection ∇p E on p∗ E; the horizontal subbundle Hp∗ E
E
of T p∗ E is the inverse image of the horizontal subbundle HE ⊂ T E under the

natural map T p∗ E → T E. It is clear that the metric is bundle-like for the lifted

foliation F.
Observe that the foliation F is transversally parallelizable, meaning that the
normal bundle of the lifted foliation is parallelizable by F-basic vector fields. To
see this, we use a modification of the standard construction of the parallelism of
the frame bundle of a manifold (see [48, p.82] for this construction in the case
where the principle bundle is the bundle of transverse orthonormal frames). Let
G = O (q) × U (k), and let θ denote the Rq –valued solder form of M → M . Given
the pair of frames z = (φ, ψ) where φ : R → Np(z) F and ψ : C → Ep(z) and given
q k

X z ∈ Tz M , we define θ (Xz ) = φ−1 π ⊥ p∗ Xz , where π ⊥ : Tp(z) M → Np(z) F is
the orthogonal projection. Let ω denote the o (q) ⊕ u (k)–valued connection one-
form. Let {e1 , ..., eq } be the standard orthonormal basis of Rq , and let {Ej }dim j=1
G
denote a fixed orthonormal basis of o (q) ⊕ u (k). We uniquely define the vector
fields V1 , ..., Vq , E1 , ..., Edim G on M by the conditions
(1) Vi ∈ Nz F, Ej∈ N
z F for every i, j.
(2) ω (Vi ) = 0, ω Ej = Ej for every i, j.

(3) θ (Vi ) = ei , θ Ej = 0 for every i, j.

Then the set of F–basic vector fields V1 , ..., Vq , E1 , ..., Edim G is a transverse

parallelism on M , F associated to the connection ∇. By the fact that M , F is
Riemannian
and the structure theorem of Molino [48, Chapter 4], the leaf closures
of M , F are the fibers of a Riemannian submersion π:M →W .
50
Next, we show that the bundle p∗ E → M is G –equivariant. An element of

F
the foliation groupoid GF is a triple of the form (y, z, [·]), where y and z are points
of a leaf of F and [·] is the set of all piecewise smooth curves starting at y and
ending at z, since all such curves are equivalent because the holonomy is trivial on
M . The basic connection on E induces a G –action on p∗ E, defined as follows.
F
Given a vector (y, v) ∈ (p∗ E)y so that y ∈ M , v ∈ Ep(y) , we define the action of
g = (y, z, [·]) by Sg (y, v) = (z, Pγ v), where γ is any piecewise smooth curve from
p (y) to p (z) in the leaf containing p (y) that lifts to a leafwise curve in M from
y to z and where Pγ denotes parallel translation in E along the curve γ. It is
easy to check that this action makes p∗ E into a GF –equivariant, foliated vector
bundle. The pullback p∗ maps basic
∗
sections of E to basic sections of p E. Also,
the O (q) × U (k)–action on M , F induces an action of O (q) × U (k) on p∗ E that
preserves the basic sections.
Observe that if s ∈ Cb∞ (E), then p∗ s is a basic section of p∗ E that is O (q) ×
U (k)–invariant. Conversely, if s ∈ C ∞ (p∗ E) is O (q) × U (k)–invariant, then s =
p∗ s for some s ∈ C ∞ (E). Next, suppose s = p∗ s is O (q) × U (k)–invariant and
basic. Given any vector X ∈ Tp(y) F and its horizontal lift X ∈ Ty F, we have
∗ ∗
0 = ∇p E s = ∇p E p∗ s = p∗ ∇E
X s,
X X
so that s is also basic. We have shown that Cb∞ (E) is, indeed, isomorphic to
O(q)×U (k)
Cb∞ M , p∗ E .
We now construct a Hermitian vector bundle E over W , similar to the con-
structions in [52] and [21]. Given w ∈ W and the corresponding leaf closure

π −1
(w) ∈ M , consider a basic section s ∈ C ∞ M , p∗ E restricted to π −1 (w).
b
Given any y ∈ M , the vector s (y) uniquely determines s on the entire leaf clo-
sure by parallel transport, because the section is smooth.
Similarly,
given a vector
∗ ∞ ∗
vy ∈ (p E)y , there exists a basic section s ∈ Cb M , p E such that s (y) = vy ,
because there is no obstruction to extending, by the following argument. Given a
basis {b1 , ..., bk } of Ck , we define the k linearly independent, basic sections sk of
p∗ E by sj ((φ, ψ)) = ψ (bj ) ∈ (p∗ E)(φ,ψ) = Ep((φ,ψ)) . Thus, given a local frame
{vj } for p∗ E on a F –transversal submanifold near y, there is a unique extension of
this frame to be a frame consisting of basic sections on a tubular neighborhood of
the leaf closure containing y; in particular a vector
may be extended to be a basic
∗ ∞ ∗
section of p E. We now define Ew = Cb M , p E ∼w , where two basic sections
→ p∗ E are equivalent(s ∼w s ) if s (y) = s (y) for every y ∈ π
s, s : M −1 (w). By
the reasoning above, Ew is a complex vector space whose dimension is equal to the
complex rank of p∗ E → M . Alternately, we could define Ew to be the vector space
∗
of F -basic sections of p E restricted to the leaf closure π −1 (w). The union ∪w∈W
Ew forms a smooth, complex vector bundle E over W ; local trivializations of E are
given by local, basic framings of the trivial bundle p∗ E → M . We remark that in
the constructions of [52] and [21], the vector bundle was lifted to the transverse
orthonormal frame bundle M , and in that case the corresponding bundle E in those
papers could have smaller rank than E.
13

We let the invertible Φ : C ∞ W , E → C ∞ M , p∗ E, F be the almost tauto-
b
is
logical map defined as follows. Given a section s of E, its value at each w ∈ W
−1
an equivalence class [s]w of basic sections. We define for each y ∈ π
(w),
s) (y) = s (y) ∈ (p∗ E)y .
Φ (
By the continuity of the basic section s, the above is independent of the choice
of this basic section in the equivalence class. By the definition of Φ and of the
trivializations of E, it is clear that Φ is a smooth map. Also, the G = O (q) × U (k)
action on basic sections of p∗ E pushes forward to a G action on sections of E.
We have the following commutative diagram, with W = M F = W G the leaf
closure space of (M, F).
p∗ E E
↓
, F
π
G → M −→ W
↓p
↓
E → (M, F) −→ W
Observe that we have the necessary data to construct the basic Dirac operator
corresponding to the pullback foliation p∗ F on M
on sections of p∗ E over M . The
p∗ E
connection ∇ is a basic connection with respect to this Riemannian foliation,
and the normal bundle N (p∗ F) projects to the normal bundle Q = N F, so that
the action of Cl (Q) on E lifts to an action of Cl (N (p∗ F)) on p∗ E. Using this
basic Clifford bundle structure, we construct the transversal
Dirac-type
operator
p∗ E p∗ E ∞ ∗ ∗
Dtr,p∗ and the basic Dirac-type operator Db,p∗ on Cb M , p E, p F ; we add the
subscript p∗ to emphasize that we are working with p∗ F rather than the lifted folia-
G
tion. Observe that C ∞ M , p∗ E, p∗ F = C ∞ M , p∗ E, F ⊂ C∞ M , p∗ E, F .
b b b
∗ ∞
It isclear from the construction that ∗p is an isomorphism from Cb (M, E) to
Cb∞ M , p∗ E, p∗ F and p∗ ◦ DE = Dp E∗ ◦ p∗ .
b
b,p
∞ , E by
We define the operator D : C W , E → C∞ W

p∗ E 1
D = Φ−1 ◦ Dtr,p ∗ − c Hb ◦ Φ,
2
where H b is the basic mean curvature of the pullback foliation, which is merely the
G
horizontal lift of Hb . Let D G denote the restriction of D to C ∞ W , E . Note
that
G G
Φ : C∞ W , E → Cb∞ M , p∗ E, F
is an isomorphism. Observe that the Hermitian metric on p∗ E induces a well-defined

Hermitian metric on E that is invariant under the action of G.
Assume that E = E + ⊕ E − with DbE± : Cb∞ (M, E ± ) → Cb∞ (M, E ∓ ). We
±
define Db,p ∗ to be the restrictions

p∗ E 1
, p∗ E ± , p∗ F → Cb∞ M

, p∗ E ∓ , p∗ F ,
Dtr,p∗ − c Hb : Cb∞ M
2
52
We define the bundles E ± and the operator

−1 p∗ E 1
, E + → C ∞ W , E −

(3.1) D = Φ ◦ Dtr,p∗ − c Hb
+
◦ Φ : C∞ W
2
in an analogous way. We now have the following result.
Proposition 3.1. Let DbE+ : Cb∞ (M, E + ) → Cb∞ (M, E − ) be a basic Dirac op-
erator for the rank k complex vector bundle E = E + ⊕ E − over the transversally
oriented Riemannian foliation (M, F) , and let G = O (q) × U (k). Then

indb DbE+ = ind DG ,

where ind DG refers to the index of the transversally elliptic operator D+ re-
stricted to G –invariant sections (equivalently, the supertrace of the invariant part
of the virtual representation–valued equivariant index of the operator
D).
It is not
− ∞ − ∞
necessarily the case that the adjoint D : C W,E →C W,E +
coincides
∗

with Φ−1 ◦ Dtr,pp E
∗ −
1
2 c Hb ◦ Φ , but the principal transverse symbols
,E − )
C ∞ (W
of D + and D − evaluated on a normal space to an orbit in W correspond with the

restriction of the principal transverse symbol of Db and DbE− restricted to the
E+
normal space to a leaf closure in M .

Proof. The kernels satisfy

ker(DbE+ ) ∼
= ker p∗ ◦ DbE+

∼ ∗
∗ ◦ p
+
= ker Db,p
Cb∞ (M,E + )

∼ p∗ E 1
∗
= ker Dtr,p∗ − c Hb ◦p
2 Cb∞ (M,E + )

∗ 1
∼ −1
= ker Φ ◦ Dtr,p∗ − c H
p E b ◦ Φ
2 C ∞ (W ,E + )G
+G
∼
= ker D ,
the kernel of the operator restricted to G –invariant sections. Next, while DbE− is
the adjoint of DbE+ with respect to the L2 -inner product on the closure of the space
of basic sections of E, it is not necessarily true that the adjoint of D+ is D− , if we
write
p∗ E 1
± −1
D = Φ ◦ Dtr,p∗ − c Hb ◦ Φ ,
2 ,E ± )
C ∞ (W
because although the operators have the same principal transverse symbol, the
volumes of the orbits on W need not coincide with the volumes of the leaf closures
on M , at least with the metric on W that we have chosen. However, it is possible
to choose a different metric, similar to that used in [52, Theorem 3.3], so by using
the induced L2 -metric on invariant sections of E over W and the L2 metric on basic

sections of E on M , Φ is an isometry. Specifically,
−1 let φ : W → R be the smooth
positive function defined by φ (w) = vol π (w) . Let dW be the dimension of

W . We determine a new metric g on W by conformally multiplying the original
metric g on W by φ2/dW ∈ C∞ , so that the volume form on W
W is multiplied
15

by φ. Note that φ (w) volg (Ow ) = vol L by the original construction, where
−1
L=p π (w) is the leaf closure corresponding to the orbit Ow = wG ⊂ W . By

using the new metric g on W , we see that Φ extends to an L -isometry and that
2
. Then
G still acts by isometries on W

E−
∼ −1 p∗ E 1

ker Db = ker Φ ◦ Dtr,p∗ − c Hb ◦ Φ
2 ,E − )G
C ∞ (W

∼
= ker DG,adj ,

where the superscript adj refers to the adjoint with respect to the L2 metrics
G
C∞ W , E ± induced by g . Therefore, the analytic basic index satisfies

indb DbE+ = ind DG ,

where ind DG is the analytic index of the transversally elliptic operator D re-
stricted to G-invariant sections, with adjoint calculated with respect to the choice
of metric g . Because the restriction of D to G-invariant sections is a Fredholm
operator (see [1]), ind DG is independent of the choice of metric.
The Fredholm properties of the equivariant index of transversally elliptic oper-
ators (see [1]) imply the following.

Corollary 3.2. In the notation of Proposition 3.1, the analytic basic index indb DbE+
is a well-defined integer. Further, it is invariant under smooth deformations of the
basic operator and metrics that preserve the invertibility of the principal
∗ symbol
∗
σ (ξx ) of DbE+ for every x ∈ M , but only for ξx ∈ Qx = Tx M Tx Lx , the dual
to the normal space to the leaf closure through x.
Note that if f is a smooth function on W such that dfw is an element of the
dual space to the normal bundle to the orbit space at w ∈ W , and if s is a smooth
section of E , then
+
+
D , f s = Φ−1 c d ( π ∗ f )# Φ (
s)

= Φ−1 c ( π ∗ df ) Φ (
#
s)
#
= c df s.
∗
This implies that D + is a Dirac operator on sections of E + , since (D+ ) D+ is a
generalized Laplacian. The analogous result is true for D− .
It is possible use the Atiyah–Segal Theorem ([4]) to compute indG (D + ), but
only in the case where D is a genuinely elliptic operator. Recall that if D is an
elliptic operator on a compact, connected manifold M that is equivariant with
respect to the action of a compact Lie group G , then G represents on both finite–
dimensional vector spaces ker D and ker D∗ in a natural way. For g ∈ G ,
indg (D ) := tr (g| ker D ) − tr (g| ker D∗ ) .
where dg is the normalized, bi-invariant measure on G . The Atiyah-Segal Theorem
computes this index in terms of an integral over the fixed point set of g. We will use
our equivariant index theorem in [14] to evaluate indG (D + ) in terms of geometric
invariants of the operator restricted to the strata of the foliation.
54
Remark 3.3. Under the additional assumption that E → (M, F) is GF –equivariant,

the pullback of E to the transverse orthonormal frame bundle is already transver-
sally parallelizable. Thus, it is unnecessary in this case to pull back again to the
unitary frame bundle. We may then replace G by O (q), and W is the so-called
basic manifold of the foliation, as in the standard construction in [48].
Remark 3.4. If (M, F) is in fact transversally orientable, we may replace O (q)

with SO (q) and work with the bundle of oriented orthonormal frames.
3.2. The asymptotic expansion of the trace of the basic heat kernel
. In this section, we will state some results concerning the spectrum of the square
of a basic Dirac-type operator and the heat kernel corresponding to this operator,
which are corollaries of the work in the previous section and are of independent
interest.
Proposition 3.5. Let DbE+ : Cb∞ (M, E + ) → Cb∞ (M, E − ) be a basic Dirac op-
erator for the rank k complex vector bundle E = E + ⊕ E − over the transversally
adj
oriented Riemannian foliation (M, F) , and let DbE+ be the adjoint operator.
Then the operators
adj E+
L+ = DbE+ Db ,
adj
L− = DbE+ DbE+
are essentially self–adjoint, and their spectrum consists of nonnegative real eigen-
values with finite multiplicities. Further, the operators L± have the same positive
spectrum, including multiplicities.
Proof. By (3.1) and the proof of Proposition 3.1, the operators L+ and L−
are conjugate to essentially self-adjoint, second order, G-equivariant, transversally
.
elliptic operators on W
The basic heat kernel Kb (t, x, y) for L is a continuous section of E E ∗ over

R>0 × M × M that is C 1 with respect to t, C 2 with respect to x and y, and satisfies,
for any vector ey ∈ Ey ,

∂
+ Lx Kb (t, x, y)ey = 0
∂t

lim Kb (t, x, y) s(y) dV (y) = s(x)
t→0+
M
for every continuous basic section s : M → E. The principal transverse symbol of L

satisfies σ (L) (ξx ) = |ξx |2 Ix for every ξx ∈ Nx F, where Ix : Ex → Ex is the identity
operator. The existence of the basic heat kernel has already been shown in [51].
Let q be the codimension of the leaf closures of (M, F) with maximal dimension.
The following theorems are consequences of [12], [13] and the conjugacy mentioned
in the proof of the proposition above.
Theorem 3.6. Under the assumptions

in Proposition 3.5, let 0 < λb0 ≤ λb1 ≤
∞
λ2 ≤ ... be the eigenvalues of L Cb (E) , counting multiplicities. Then the spectral
b
17
counting function Nb (λ) satisfies the asymptotic formula

Nb (λ) : = # λbm λbm < λ
rank (E) Vtr
∼ λq/2 .
(4π)q/2 Γ q2 + 1
Theorem 3.7. Under the assumptions in Proposition 3.5, the heat operators e−tL
+
−
and e−tL are trace class, and they satisfy the following asymptotic expansions.
Then, as t → 0,
⎛ ⎞
1 ⎜ ⎟
±
Tre−tL = Kb± (t) ∼ ⎜a± + a± j/2
(log t)k ⎟
⎝ 0 j,k t ⎠,
tq/2 j≥1
0≤k<K0
where K0 is less than or equal to the number of different dimensions of closures of

infinitesimal holonomy groups of the leaves of F.
Remark 3.8. (The basic zeta function and determinant of the generalized basic
Laplacian) We remark that due to the singular asymptotics lemma of [15], we have
that ajk = 0 for j ≤ q, k > 0. We conjecture that all the logarithmic terms vanish.
Note that the fact that aqk = 0 for k > 0 implies that the corresponding zeta function
ζL (z) is regular at z = 0, so that the regularized determinant of L may be defined.
3.3. Stratifications of G-manifolds and Riemannian foliations. In the
following, we will describe some standard results from the theory of Lie group
actions and Riemannian foliations (see [9], [38], [48]). Such G-manifolds and Rie-
mannian foliations are stratified spaces, and the stratification can be described
explicitly. In the following discussion, we often have in mind that W is the ba-
sic manifold corresponding to (M, F) described in the last section, but in fact the
ideas apply to any Lie group G acting on a smooth, closed, connected manifold W .
In the context of this paper, either G is O (q), SO (q), or the product of one of
these with U (k). We will state the results for general G and then specialize to the
case of Riemannian foliations (M, F) and the associated basic manifold. We also
emphasize that our stratification of the foliation may be finer than that described
in [48], because in addition we consider the action of the holonomy on the relevant
vector bundle when identifying isotropy types.
Given a G-manifold W and w ∈ W , an orbit Ow = {gw : g ∈ G} is naturally
diffeomorphic to G/Hw , where Hw = {g ∈ G |wg = w} is the (closed) isotropy
or stabilizer subgroup. In the foliation case, the group Hw is isomorphic to the
structure group corresponding
−1 to the principal bundle p : π−1 (w) → L, where L
is the leaf closure p π (w) in M . Given a subgroup H of G, let [H] denote
the conjugacy class of H. The isotropy type of the orbit Ox is defined to be the
conjugacy class [Hw ] , which is well–defined independent of w ∈ Ox . On any such
G-manifold, there are a finite number of orbit types, and there is a partial order
on the set of orbit types. Given subgroups H and K of G, we say that [H] ≤
[K] if H is conjugate to a subgroup of K, and we say [H] < [K] if [H] ≤ [K]
and [H] = [K]. We may enumerate the conjugacy classes of isotropy subgroups
as [G0 ] , ..., [Gr ] such that [Gi ] ≤ [Gj ] implies that i ≤ j. It is well-known that
the union of the principal orbits (those with type [G0 ]) form an open dense subset
0 = W
W ([G0 ]) of the manifold W , and the other orbits are called singular. As
56
a consequence, every isotropy subgroup H satisfies [G0 ] ≤ [H]. Let W j denote the
of orbit type [Gj ] for each j; the set W
set of points of W j is called the stratum
j contains
corresponding to [Gj ]. If [Gj ] ≤ [Gk ], it follows that the closure of W
k . A stratum W
the closure of W j is called a minimal stratum if there does not
exist a stratum W k W
k such that [Gj ] < [Gk ] (equivalently, such that W j ). It is

known that each stratum is a G-invariant submanifold of W , and in fact a minimal
stratum is a closed (but not necessarily connected) submanifold. Also, for each j,
≥j :=
the submanifold W W k is a closed, G-invariant submanifold.
[Gk ]≥[Gj ]
Now, given a proper, G-invariant submanifold S of W and ε > 0, let Tε (S)

denote the union of the images of the exponential map at s for s ∈ S restricted
to the open ball of radius ε in the normal bundle at S. It follows that Tε (S) is
also G-invariant. j is a stratum and ε is sufficiently small, then all orbits in
If W

Tε W j \ W j are of type [Gk ], where [Gk ] < [Gj ]. This implies that if j < k,
j ∩ W
W k = ∅, and W k W j , then W j and W k intersect at right angles, and
their intersection consists of more singular strata (with isotropy groups containing
conjugates of both Gk and Gj ).
Fix ε > 0. We now decompose W as a disjoint union of sets W
ε, . . . , W
rε . If
0
ε ε
there is only one isotropy type on W , then r = 0, and we let W0 = Σ0 = W0 = W .
Otherwise, for j = r, r − 1, ..., 0, let εj = 2 ε, and let
j

(3.2) j \
Σεj = W ε, W
W ε = Tε W j \ W ε,
k j j k
k>j k>j
Thus,

jε , Σεj ⊂ W
Tε Σεj ⊂ W j .
We now specialize to the foliation case. Let (M, F) be a Riemannian foliation,

and let E → M be a foliated Hermitian vector bundle over M (defined in Section
2.2). Let the G-bundle M → M be either the orthonormal transverse frame bundle
of (M, F) or the bundle of ordered pairs (α, β), with α a orthonormal transverse
frame and β an orthonormal frame of E with respect to the Hermitian inner product
on E, as in Section 3.1. In the former case, M −→ p
M is an O (q)-bundle, and in
is an O (q) × U (k)-bundle. We also note that in the case where
the latter case, M
(M, F) is transversally oriented, we may replace O (q) with SO (q) and choose
oriented transverse frames. In Section 3.1, we showed that the foliation F lifts to a
foliation F on M
, and the lifted foliation is transversally parallelizable. We chose a
natural metric on M , as explained in Section 3.1. By Molino theory ([48]), the leaf

closures of F are diffeomorphic and have no holonomy; they form a Riemannian
−→
π , on which the
fiber bundle M W over what is called the basic manifold W
group G acts by isometries.
We identify the spaces W ε, W ε G, and W ε G with the corresponding M ε ,
i i i i
Mi F , and Mi F on M via the correspondence
ε ε

p π −1 G–orbit on W = leaf closure of (M, F) .
19
The following result is contained in [48], which is a consequence of Riemannian fo-

liation theory and the decomposition theorems of G-manifolds (see [38]). However,
we note that the decomposition described below may be finer than that described
in Molino, as the bundle E → M is used in our construction to construct the basic
manifold, and the group acting may be larger than the orthogonal group. The
action of the holonomy on the bundle may participate in the decomposition of the
foliation.
Lemma 3.9. Let (M, F) be a Riemannian foliation with bundle-like metric. Let
F denote the(possibly)
singular
foliation by leaf
closures
of F. We let Mj =
−1 −1 ε , W ([Gj ]), W
ε de-

M ([Gj ]) = p π ε
W ([Gj ]) , M = p πi W with W
i i
fined as above on the basic manifold. Note that Mj is a stratum on M correspond-
ing to the union of all leaf closures whose structure group of the principal bundle
p:L → L is in [Gj ], where L is a leaf closure of M
that projects to L. It follows
that all the leaf closures in Mj have the same dimension. Then we have, for every
i ∈ {1, . . . , r} and sufficiently small ε > 0:
!r
(1) M = Miε (disjoint union).
i=0
(2) Miε is a union of leaf closures.
(3) The manifold Miε is diffeomorphic to the interior of a compact manifold
with corners; the leaf closure space space Miε F ∼ =W ε O (q) is a smooth
i
manifold that is isometric to the interior of a triangulable, compact man-
ifold with corners. The same is true for each Σεi , Mi F .
(4) If [Gj ] is the isotropy type of an orbit in Miε , then j ≤ i and [Gj ] ≤ [Gi ].
(5) The distance between the submanifold Mj and Miε for j > i is at least ε.
Remark 3.10. The lemma above remains true if at each stage Tε (Mj ) is replaced
by any sufficiently small open neighborhood of Mj that contains Tε (Mj ), that is a
F-saturated, and whose closure is a manifold with corners.
Remark 3.11. The additional frames of E have no effect on the stratification of
→ M is
M ; the corresponding Miε , Mi are identical whether or not the bundle M
chosen to be the O (q)-bundle or the O (q) × U (k)-bundle. However, the isotropy
subgroups and basic manifold W are different and depend on the structure of the
bundle E.
Definition 3.12. With notation as in this section, suppose that [H] is a maxi-
mal isotropy type with respect to the partial order ≤. Then the closed, saturated
submanifold M ([H]) is called a minimal stratum of the foliation (M, F).
3.4. Fine components and canonical isotropy bundles. First we review
some definitions from [14] and [32] concerning manifolds X on which a compact
H
Lie group G acts by isometries with
single Horbit type [H]. Let X be the fixed
point set of H, and for α ∈ π0 X , let Xα denote the corresponding connected
H
component of X H .
Definition 3.13. We denote Xα = GXαH , and Xα is called a component of X
relative to G.
Remark 3.14. The space Xα is not necessarily connected, but it is the inverse
image of a connected component of GX = N X H under the projection X →
58
GX. Also, note that Xα = Xβ if there exists n ∈ N such that nXαH = XβH . If X
is a closed manifold, then there are a finite number of components of X relative to
G.
We now introduce a decomposition of a G-bundle E → X over a G-space
with single orbit type [H] that is a priori finer than the normalized isotypical
decomposition. Let Eα be the restriction E|X H . For σ : H → U (Wσ ) an irreducible
α
unitary representation, let σ n : H → U (Wσ ) be the irreducible representation
defined by
σ n (h) = σ n−1 hn .
Let N[σ] = {n ∈ N : [σ n ] is equivalent to [σ] } . If the isotypical component
Ex[σ] := iσ (HomH (Wσ , Ex ) ⊗ Wσ )

is nontrivial, then it is invariant under the subgroup N α,[σ] ⊆ N
[σ] that leaves in
H
addition the connected component Xα invariant; again, this subgroup has finite
index in N . The isotypical components transform under n ∈ N as
∼
= [σ n ]
n : Eα[σ] −→ En(α) ,
where n denotes the residue class class of n ∈ N in N N α,[σ] . Then a decompo-

[σ]
sition of E is obtained by ‘inducing up’ the isotypical components Eα from N α,[σ]
to N . That is,
N
Eα,[σ] = N ×Nα,[σ] Eα[σ]

[σ]
is a bundle containing Eα H . This is an N -bundle over N XαH ⊆ X H , and a
Xα
similar bundle may be formed over each distinct N XβH , with β ∈ π0 X H . Further,
N
observe that since each bundle Eα,[σ] is an N -bundle over N XαH , it defines a unique
G
G bundle Eα,[σ] .
G
Definition 3.15. The G-bundle Eα,[σ] over the submanifold Xα is called a fine
component or the fine component of E → X associated to (α, [σ]).
If GX is not connected, one must construct the fine components separately
over each Xα . If E has finite rank, then E may be decomposed as a direct sum of
N
distinct fine components over each Xα . In any case, Eα,[σ] is a finite direct sum of
H
isotypical components over each Xα .
Definition 3.16. The direct sum decomposition of E|Xα into subbundles E b that
G
are fine components Eα,[σ] for some [σ], written
"
E|Xα = Eb ,
b
is called the refined isotypical decomposition (or fine decomposition) of
E|Xα .
In the case where GX is connected,
the group π0 (N H) acts transitively
on the connected components π0 X H , and thus Xα = X. We comment that if
[σ]
[σ, Wσ ] is an irreducible H-representation present in Ex with x ∈ XαH , then Ex
n
[σ ]
is a subspace of a distinct Exb for some b. The subspace Exb also contains Ex for
H H
every n such that nXα = Xα .
21
Remark 3.17. Observe that by construction, for x ∈ XαH the multiplicity and
[σ n ]
dimension of each [σ] present in a specific Exb is independent of [σ]. Thus, Ex
[σ]
and Ex have the same multiplicity and dimension if nXαH = XαH .
Remark 3.18. The advantage of this decomposition over the isotypical decom-
position is that each E b is a G-bundle defined over all of Xα , and the isotypical
decomposition may only be defined over XαH .
Definition 3.19. Now, let E be a G-equivariant vector bundle over X, and let E b be
a fine component as in Definition 3.15 corresponding to a specific component Xα =
GXαH of X relative to G. Suppose that another G-bundle W over Xα has finite
rank and has the property that the equivalence classes of Gy -representations present
in Eyb , y ∈ Xα exactly coincide with the equivalence classes of Gy -representations
present in Wy , and that W has a single component in the fine decomposition. Then
we say that W is adapted to E b .
Lemma 3.20. In the definition above, if another G-bundle W over Xα has finite
rank and has the property that the equivalence classes of Gy -representations present
in Eyb , y ∈ Xα exactly coincide with the equivalence classes of Gy -representations
present in Wy , then it follows that W has a single component in the fine decom-
position and hence is adapted to E b . Thus, the last phrase in the corresponding
sentence in the above definition is superfluous.
Proof. Suppose that we choose an equivalence class [σ] of H-representations
present in Wx , x ∈ XαH . Let [σ ] be any other equivalence class; then, by hypoth-
esis, there exists n ∈ N such that nXαH = XαH and [σ ] = [σ n ]. Then, observe
[σ] [σ n ] [σ n ]
that nWx = Wnx = Wx , with the last equality coming from the rigidity of
irreducible H-representations. Thus, W is contained in a single fine component,
and so it must have a single component in the fine decomposition.
In what follows, we show that there are naturally defined finite-dimensional
vector bundles that are adapted to any fine components.
Once and for all, we
enumerate the irreducible representations ρj , Vρj j=1,2,... of G. Let [σ, Wσ ] be
any irreducible H-representation. Let G ×H Wσ be the corresponding homoge-
neous vector bundle over the homogeneous space GH. Then the L2 -sections of
this
vector
bundle decompose into irreducible G-representations. In particular, let
ρj0 , Vρj0 be the equivalence class of irreducible representations that is present in
L2 (GH, G ×H Wσ ) and that has the lowest index j0 . Then Frobenius reciprocity
implies

0 = HomG Vρj0 , L2 (GH, G ×H Wσ ) ∼ = HomH VRes(ρj ) , Wσ ,
0
so that the restriction of ρj0 to H contains the H-representation [σ]. Now, for a
component XαH of X H , with Xα = GXαH its component in X relative to G, the
trivial bundle
X α × Vρj0
is a G-bundle (with diagonal action) that contains a nontrivial fine component
[σ]
Wα,[σ] containing XαH × Vρj0 .
Definition 3.21. We call Wα,[σ] → Xα the canonical isotropy G-bundle as-

Observe that Wα,[σ] depends only on the
sociated to (α, [σ]) ∈ π0 X H × H.
60
enumeration of irreducible representations of G, the irreducible H-representation

[σ] and the component XαH . We also denote the following positive integers associ-
ated to Wα,[σ] :

• mα,[σ] = dim HomH Wσ , Wα,[σ],x = dim HomH Wσ , Vρj0 (the asso-
ciated multiplicity), independent of the choice of [σ, Wσ ] present in
Wα,[σ],x , x ∈ XαH (see Remark 3.17).
• dα,[σ] = dim Wσ (the associated representation dimension), indepen-
dent of the choice of [σ, Wσ ] present in Wα,[σ],x , x ∈ XαH .
rank(Wα,[σ] )
• nα,[σ] = mα,[σ] dα,[σ] (the inequivalence number), the number of in-
equivalent representations present in Wα,[σ],x , x ∈ XαH .
Remark 3.22. Observe that Wα,[σ] = Wα ,[σ ] if [σ ] = [σ n ] for some n ∈ N such
that nXαH = XαH .
The lemma below follows immediately from Lemma 3.20.
Lemma 3.23. Given any G-bundle E → X and any fine component E b of E over
some Xα = GXαH , there exists a canonical isotropy G-bundle Wα,[σ] adapted to
E b → Xα .
An example of another foliated bundle over a component of a stratum Mj is
the bundle defined as follows.

Definition 3.24. Let E → M be any foliated vector bundle. Let Σαj = π p−1 (Mj )
be the corresponding component of the stratum relative to G on the basic manifold
(see Section 3.3), and let W τ → Σα be a canonical isotropy bundle (Defini-
W j
tion 3.21). Consider the bundle π ∗ W τ ⊗ p∗ E → p−1 (Mj ), which is foliated and
basic for the lifted foliation restricted to p−1 (Mj ). This defines a new foliated bun-
dle E τ → Mj by letting Exτ be the space of G-invariant sections of π ∗ W τ ⊗ p∗ E
−1
restricted to p (x). We call this bundle the W -twist of E → Mj .
τ
4. Desingularization of the foliation

4.1. Topological Desingularization. Assume that (M, F) is a Riemannian
foliation, with principal stratum M0 and singular strata M1 , ..., Mr corresponding
to isotropy types [G0 ], [G1 ], [G2 ], ..., [Gr ] on the basic manifold, as explained in
Section 3.3. We will construct a new Riemannian foliation (N, FN ) that has a
single stratum (of type [G0 ]) and that is a branched cover of M , branched over
the singular strata. A distinguished fundamental domain of M0 in N is called the
desingularization of M and is denoted M #. This process closely parallels the process
of desingularizing a G-manifold, which is described in [14].
Recall the setup from Section 3.3. We are given E → M , a foliated Hermitian
vector bundle over M , and the bundle M −→p
M is the bundle of ordered pairs (α, β)
with structure group G = O (q) × U (k), with α a orthonormal transverse frame and
β an orthonormal frame of E with respect to the Hermitian inner product on E,
as in Section 3.1; in many cases the principal bundle may be reduced to a bundle
with smaller structure group. The foliation F lifts to a foliation F on M , and the
lifted foliation is transversally parallelizable. We chose the natural metric on M as
in Section 3.1. By Molino theory ([48]), the leaf closures of F are diffeomorphic,
−→
π
have no holonomy, and form a Riemannian fiber bundle M W over the basic
23
manifold W , on which the group G acts by isometries. The G-orbits on W

and leaf
closures of (M, F) are identified via the correspondence

p π−1 G–orbit on W = leaf closure of (M, F) .
A sequence of modifications is used to construct N and M # ⊂ N . Let Mj

be a minimal stratum. Let Tε (Mj ) denote a tubular neighborhood of radius ε
around Mj , with ε chosen sufficiently small so that all leaf closures in Tε (Mj ) \ Mj
correspond to isotropy types [Gk ], where [Gk ] < [Gj ]. Let
N 1 = (M \ Tε (Mj )) ∪∂Tε (Mj ) (M \ Tε (Mj ))
be the manifold constructed by gluing two copies of (M \ Tε (Mj )) smoothly along
the boundary. Since the Tε (Mj ) is saturated (a union of leaves), the foliation lifts
to N 1 . Note that the strata of the foliation F 1 on N 1 correspond to strata in
M \ Tε (Mj ). If Mk ∩ (M \ Tε (Mj )) is nontrivial, then the stratum corresponding
to isotropy type [Gk ] on N 1 is
Nk1 = (Mk ∩ (M \ Tε (Mj ))) ∪(Mk ∩∂Tε (Mj )) (Mk ∩ (M \ Tε (Mj ))) .
1 1
Thus, N , F is a foliation with one fewer stratum than (M, F), and M \ Mj is
diffeomorphic to one copy of (M \ Tε (Mj )), denoted M #1 in N 1 . One may radially
modify metrics
so that
a bundle-like metric on (M, F) transforms to a bundle-like
metric on N 1 , F 1 . In fact, N
1
is a branched double cover of M , branched over
Mj . If the leaf closures of N , F 1 correspond to a single orbit type, then we set
1

N = N 1 and M #=M #1 . If not, we repeat the process with the foliation N 1 , F 1 to
2 2
produce a new Riemannian foliation N , F with two fewer strata than (M, F)
and that is a 4-fold branched cover of M . Again, M #2 is a fundamental domain of
#
M \ {a minimal stratum}, which is a fundamental domain of M with two strata
1
removed. We continue until (N, FN ) = (N r , F r ) is a Riemannian foliation with all

leaf closures corresponding to orbit type [G0 ] and is a 2r -fold branched cover of M ,
branched over M \ M0 . We set M #=M #r , which is a fundamental domain of M0 in
N.
Further, one may independently desingularize M≥j , since this submanifold is
itself a closed G-manifold. If M≥j has more than one connected component, we
may desingularize all components simultaneously. The isotropy type corresponding

to all leaf closures of M
≥j is [Gj ], and M≥j F is a smooth (open) manifold.
4.2. Modification of the metric and differential operator. We now

more precisely describe the desingularization. If (M, F) is equipped with a ba-
sic, transversally elliptic differential operator on sections of a foliated vector bundle
over M , then this data may be pulled back to the desingularization M #. Given the
j
bundle and operator over N , simply form the invertible double of the operator on
N j+1 , which is the double of the manifold with boundary N j \ Tε (Σ), where Σ is
a minimal stratum on N j .
Specifically, we modify the bundle-like metric radially so that there exists suffi-
ciently small ε > 0 such that the (saturated) tubular neighborhood B4ε Σ of Σ in N j
is isometric to a ball of radius 4ε in the normal bundle N Σ. In polar coordinates,
this metric is ds2 = dr 2 + dσ 2 + r 2 dθσ2 , with r ∈ (0, 4ε), dσ 2 is the metric on Σ, and
dθσ2 is the metric on S (Nσ Σ), the unit sphere in Nσ Σ; note that dθσ2 is isometric to
the Euclidean metric on the unit sphere. We simply choose the horizontal metric
62
on B4ε Σ to be the pullback of the metric on the base Σ, the fiber metric to be
Euclidean, and we require that horizontal and vertical vectors be orthogonal. We
do not assume that the horizontal distribution is integrable. We that the metric
constructed above is automatically bundle-like for the foliation.
Next, we replace r 2 with f (r) = [ψ (r)]2 in the expression for the metric,
where ψ (r) is increasing, is a positive constant for 0 ≤ r ≤ ε, and ψ (r) = r for
2ε ≤ r ≤ 3ε. Then the metric is cylindrical for r < ε.
In our description of the modification of the differential operator, we will need
the notation for the (external) product of differential operators. Suppose that F →
π
X→ B is a fiber bundle that is locally
a metric product. Given an operator A1,x :
Γ π −1 (x) , E1 → Γ π −1 (x) , F1 that is locally given as a differential operator
A1 : Γ (F, E1 ) → Γ (F, F1 ) and A2 : Γ (B, E2 ) → Γ (B, F2 ) on Hermitian bundles,
we define the product
A1,x ∗ A2 : Γ (X, (E1 E2 ) ⊕ (F1 F2 )) → Γ (X, (F1 E2 ) ⊕ (E1 F2 ))
as the unique linear operator that satisfies locally

A1 1 −1 A∗2
A1,x ∗ A2 =
1 A2 A∗1 1
on sections of
E1 E2
F1 F2

u1 u2
of the form , where u1 ∈ Γ (F, E1 ), u2 ∈ Γ (B, E2 ), v1 ∈ Γ (F, F1 ),
v1 v2
v2 ∈ Γ (B, E2 ). This coincides with the product in various versions of K-theory (see,
for example, [1], [42, pp. 384ff]), which is used to define the Thom Isomorphism
in vector bundles.
Let D = D+ : Γ N j , E + → Γ N j , E − be the given first order, transversally
elliptic, F j -basic differential operator. Let Σ be a minimal stratum of N j . We
assume for the moment that Σ has codimension at least two. We modify the
bundle radially so that the foliated bundle E over B4ε (Σ) is a pullback of the
bundle E|Σ → Σ. We assume that near Σ, after a foliated homotopy D+ can be
written on B4ε (Σ) locally as the product
+
(4.1) D+ = (DN ∗ DΣ ) ,
where DΣ is a transversally elliptic, basic, first order operator on the stratum
(Σ, F|Σ ), and DN is a basic, first order operator on B4ε (Σ) that is elliptic on the
fibers. If r is the distance from Σ, we write DN in polar coordinates as

1 S
DN = Z ∇∂r + D
E
r
where Z = −iσ (DN ) (∂r ) is a local bundle isomorphism and the map DS is a purely
first order operator that differentiates in the unit normal bundle directions tangent
to Sx Σ.
π
We modify the operator DN on each Euclidean fiber of N Σ → Σ by adjusting
the coordinate r and function 1r so that DN ∗ DΣ is converted to an operator on a
cylinder; see [14, Section 6.3.2] for the precise details. The result is a G-manifold
#j with boundary ∂ M
M #j , a G-vector bundle E j , and the induced operator D
j , all of
which locally agree with the original counterparts outside Bε (Σ). We may double
25
#j along the boundary ∂ M

M #j and reverse the chirality of E j as described in [8, Ch.
9]. Doubling produces a closed manifold N with foliation F j , a foliated bundle
j
E j , and a first-order transversally elliptic differential operator Dj . This process

may be iterated until all leaf closures are principal. The case where some strata
have codimension 1 is addressed in the following paragraphs.
We now give the definitions for the case when there is a minimal stratum Σ of
codimension 1. Only the changes to the argument are noted. This means that the
isotropy subgroup H corresponding to Σ contains a principal isotropy subgroup of
index two. If r is the distance from Σ, then DN has the form

1 S
DN = Z ∇∂r + D
E
= Z∇E ∂r
r
where Z = −iσ (DN ) (∂r ) is a local bundle isomorphism and the map DS = 0.
In this case, there is no reason to modify the metric inside Bε (Σ). The “desin-
gularization” of M along Σ is the manifold with boundary M # = M Bδ (Σ) for
some 0 < δ < ε; the singular stratum is replaced by the boundary ∂ M # = Sδ (Σ),
which is a two-fold cover of Σ and whose normal bundle is necessarily oriented (via
∂r ). The double M is identical to the double of M # along its boundary, and M
contains one less stratum.
4.3. Discussion of operator product assumption. We now explain spe-

cific situations that guarantee that, after a foliated homotopy, D+ may be written
locally as a product of operators as in (4.1) over the tubular neighborhhood B4ε (Σ)
over a singular stratum Σ. This demonstrates that this assumption is not overly
restrictive. We also emphasize that one might think that this assumption places
conditions on the curvature of the normal bundle N Σ; however, this is not the
case for the following reason. The condition is on the foliated homotopy class of
the principal transverse symbol of D. The curvature of the bundle only effects the
zeroth order part of the symbol. For example, if Y → X is any fiber bundle over
a spinc manifold X with fiber F , then a Dirac-type operator D on Y has the form
D = ∂X ∗ DF + Z, where DF is a family of fiberwise Dirac-type operators, ∂X is
the spinc Dirac operator on X, and Z is a bundle endomorphism.
First, we show that if D+ is a transversal Dirac operator at points of Σ, and
if either Σ is spinc or its normal bundle N Σ → Σ is (fiberwise) spinc , then it
has the desired form. Moreover, we also remark that certain operators, like those
resembling transversal de Rham operators, always satisfy this splitting condition
with no assumptions on Σ.
Let N F be normal bundle of the foliationFΣ = F|Σ , and let N Σ be the normal
bundle of Σ in M . Then the principal transverse symbol of D+ (evaluated at
ξ ∈ Nx∗ FΣ ⊕ Nx∗ Σ) at points x ∈ Σ takes the form of a constant multiple of Clifford
multiplication. That is, we assume there is an action c of Cl (N FΣ ⊕ N Σ) on E
and a Clifford connection ∇ on E such that the local expression for D is given by
the composition
∇ proj
Γ (E) → Γ (E ⊗ T ∗ M ) → Γ (E ⊗ (N ∗ FΣ ⊕ N ∗ Σ))
∼
= c
→ Γ (E ⊗ (N FΣ ⊕ N Σ)) → Γ (E) .
64
The principal transverse symbol σ (D+ ) at ξx ∈ Tx∗ Σ is

q
σ D +
(ξx ) = ic (ξx ) : Ex+ → Ex−
j=1
Suppose N Σ is spin ; then there exists a vector bundle S = S + ⊕ S − → Σ that

c
is an irreducible representation of Cl (N Σ) over each point of Σ, and we let E Σ =

EndCl(N Σ) (E) and have
E∼ Σ
= S ⊗E
as a graded tensor product, such that the action of
Cl (N FΣ ⊕ N Σ) ∼ (N FΣ )
= Cl (N Σ) ⊗Cl
(as a graded tensor product) on E + decomposes as
+ −
c (x) ⊗ 1 −1 ⊗ c (y)∗ S ⊗ E Σ+ S ⊗ E Σ+
∗ : →
1 ⊗ c (y) c (x) ⊗ 1 S − ⊗ E Σ− S + ⊗ E Σ−
(see [5], [42]). If we let the operator ∂ N denote the spinc transversal Dirac operator
on sections of π ∗ S → N Σ, and let DΣ be the transversal Dirac operator defined by
the action of Cl (N FΣ ) on E Σ , then we have
+
D+ = ∂ N ∗ DΣ
up to zeroth order terms (coming from curvature of the fiber).
The same argument works if instead we have that the bundle N FΣ → Σ is spinc .
In this case a spinc Dirac operator ∂ Σ on sections of a complex spinor bundle over
Σ is transversally elliptic to the foliation FΣ , and we have a formula of the form
+
D+ = DN ∗ ∂ Σ ,
again up to zeroth order terms.
Even if N Σ → Σ and N FΣ → Σ are not spinc , many other first order operators
have splittings as in Equation (4.1). For example, if D+ is a transversal de Rham
operator from even to odd forms, then D+ is the product of de Rham operators in
the N Σ and N FΣ directions.
In [27], where a formula for the basic index is derived, the assumptions dictate
that every isotropy subgroup is a connected torus, which implies that N Σ → Σ
automatically carries a vertical almost complex structure and is thus spinc , so that
the splitting assumption is automatically satisfied in their paper as well.
5. The equivariant index theorem

We review some facts about equivariant index theory and in particular make
note of [14, Theorem 9.2]. Suppose that a compact Lie group G acts by isometries
on a compact, connected Riemannian manifold W . In the following sections of the
paper, we will be particularly interested in the case where W is the basic manifold
−
associated to (M, F) and G = O (q). Let E = E ⊕ E be a graded, G-equivariant
+
Hermitian vector bundle . We consider a first order G-equivariant differential

over W
operator D = D+ : Γ W , E + → Γ W , E − that is transversally elliptic, and let
D− be the formal adjoint ofD+ .

The group G acts on Γ W , E ± by (gs) (x) = g · s g −1 x , and the (possibly
infinite-dimensional) subspaces ker (D+ ) and ker (D− ) are G-invariant subspaces.
27
Let ρ : G → U (Vρ ) be an irreducible

unitary representation of G, and let χρ = tr (ρ)
ρ

denote its character. Let Γ W , E ±
be the subspace of sections that is the direct

sum of the irreducible G-representation subspaces of Γ W , E ± that are unitarily
equivalent to the representation ρ. It can be shown that the extended operators
ρ ρ
Dρ,s : H s Γ W , E + → H s−1 Γ W , E −
are Fredholm and independent of s, so that each irreducible representation of G ap-

pears with finite multiplicity in ker D± (see [14]). Let a±
ρ ∈ Z≥0 be the multiplicity
of ρ in ker (D± ).
The study of index theory for such transversally elliptic operators was initiated
by M. Atiyah and I. Singer in the early 1970s ([1]). The virtual representation-
valued index of D is given by

−
indG (D) := ρ − aρ [ρ] ,
a+
ρ
where [ρ] denotes the equivalence class of the irreducible representation ρ. The
index multiplicity is
1
−
ρ − aρ =
indρ (D) := a+ ind D|Γ(W ,E − )ρ .
,E + )ρ →Γ(W
dim Vρ
In particular, if ρ0 is the trivial representation of G, then

ρ0
ind (D) = ind D|Γ W ,E − )G ,
( ,E + ) →Γ(W
G
where the superscript G implies restriction to G-invariant sections.

There is a clear relationship between the index multiplicities and Atiyah’s equi-
variant distribution-valued index indg (D); the multiplicities
ρ determine the distri-
butional index, and vice versa. The space Γ W , E ±
is a subspace of the λρ -
eigenspace of C. The virtual character indg (D) is given by (see [1])
indg (D) : = “tr ( g|ker D+ ) − tr ( g|ker D− ) ”

= indρ (D) χρ (g) .
ρ
Note that the sum above does not in general converge, since ker D+ and ker D− are
in general infinite-dimensional, but it does make sense as a distribution on G. That
is, if dg is the normalized, biinvariant Haar measure on G, and if φ = β + cρ χρ ∈
C ∞ (G), with β orthogonal to the subspace of class functions on G, then

ind∗ (D) (φ) = “ φ (g) indg (D) dg”

G

ρ
= ind (D) φ (g) χρ (g) dg = indρ (D) cρ ,
ρ ρ
an expression which converges because cρ is rapidly decreasing and indρ (D) grows
at most polynomially as ρ varies over the irreducible representations of G. From
this calculation, we see that the multiplicities determine Atiyah’s distributional
66
index. Conversely, let α : G → U (Vα ) be an irreducible unitary representation.

Then

ind∗ (D) (χα ) = indρ (D) χα (g) χρ (g) dg = indα D,
ρ
so that complete knowledge of the equivariant distributional index is equivalent to

ρ
knowing all of the
multiplicities
ρ ind
(D).Because the operator D induces a Fred-
ρ
holm operator Γ W , E +
→Γ W , E −
, all of the indices indG (D) , indg (D),
and indρ (D) depend only on the stable homotopy class of the principal transverse
symbol of D.
The equivariant index theorem ([14, Theorem 9.2]) expresses indρ (D) as a sum
of integrals over the different strata of the action of G on W , and it involves the
eta invariant of associated equivariant elliptic operators on spheres normal to the
strata. The result is

+
r

indρ (D) = Aρ0 (x) |dx| β Σαj ,

0
GW j=1
1 1
S+,b
β Σαj = ( − η Dj
2 dim Vρ nb rank W b
b∈B

+h DjS+,b ) ,
Aρj,b (x) |dx|

GΣ αj
(The notation is explained in [14]; the integrands Aρ0 (x) and Aρj,b (x) are the familar
Atiyah-Singer integrands corresponding to local heat kernel supertraces of induced
elliptic operators over closed manifolds.)
6. The basic index theorem

Suppose that E is a foliated Cl (Q) module with basic Cl (Q) connection ∇E
over a Riemannian foliation (M, F). Let

DbE : Γb E + → Γb E −

be the corresponding basic Dirac operator, with basic index indb DbE .
In what follows, if U denotes an open subset of a stratum of (M, F), U denotes
the desingularization of U very similar to that in Section 4, and U denotes the
fundamental domain of U inside U . We assume that near each component Mj
of a singular stratum of (M, F), DbE is homotopic (through basic, transversally
elliptic operators) to the product DN ∗ DMj , where DN is an F-basic, first order
differential operator on a tubular neighborhood of Σαj that is elliptic and Zhas
constant coefficients on the fibers and DMj is a global transversally elliptic, basic,
first order operator on the Riemannian foliation (Mj , F). In polar coordinates, the
fiberwise elliptic operator DN may be written

1 S
DN = Zj ∇∂r + Dj
E
,
r
where r is the distance from Mj , where Zj is a local bundle isometry (dependent
on the spherical parameter), the map DjS is a family of purely first order operators
that differentiates in directions tangent to the unit normal bundle of Mj .
29
Theorem 6.1. (Basic Index Theorem for Riemannian foliations) Let M0 be the
principal stratum of the Riemannian foliation (M, F), and let M1 , ... , Mr denote
all the components of all singular strata, corresponding to O (q)-isotropy types [G1 ],
... ,[Gr ] on the basic manifold. With notation as in the discussion above, we have

+
r
indb DbE = A0,b (x) |dx| β (Mj ) ,

M 0 F j=1
1 1
β (Mj ) = −η D S+,τ
+ h D S+,τ ,
Aτj,b (x) |dx|
j j
2 τ nτ rank W τ Mj F
where the sum is over all components of singular strata and over all canonical
isotropy bundles W τ , only a finite number of which yield nonzero Aτj,b , and where
(1) A0,b (x) is the Atiyah-Singer integrand, the local supertrace of the ordi-
nary heat kernel associated to the elliptic operator induced from D E (a
b
desingularization of DbE ) on the quotient M#0 F , where the bundle E is
replaced
by
the space
of basic
sections of over each leaf closure;
(2) η DjS+,b and h DjS+,b are the equivariant eta invariant and dimension
of the equivariant kernel of the Gj -equivariant operator DjS+,b (defined in
a similar way as in [14, formulas (6.3), (6.4), (6.7)]);
(3) Aτj,b (x) is the local supertrace of the ordinary heat kernel associated to

the elliptic operator induced from 1 ⊗ DMj (blown-up and doubled from
1 ⊗ DMj , the twist of DMj by the canonical isotropy bundle W τ from
Definition 3.24) on the quotient M #j F , where the bundle is replaced by
the space of basic sections over each leaf closure; and
(4) nτ is the number of different inequivalent Gj -representation types present
in a typical fiber of W τ .
Proof. Using Proposition 3.1, we have

indb DbE = ind DG ,
where D = D + is defined in (3.1). Let Σα1 , ..., Σαr denote the components of
the strata of the basic manifold W relative to the G-action corresponding to the
1 , ..., Mr. Near each Σαj , we write D = DN ∗ D , and write
αj
components M
DN = Zj ∇∂r + r Dj in polar coordinates. By the Invariant Index Theorem
E 1 S
[14, Theorem 9.6], a special case of the Equivariant Index Theorem stated in the
last section, we have

+
r

ind DG = AG0 (x) |dx| β Σαj ,

0
GW j=1
1 1
−η S+,τ S+,τ
β Σαj = Dj + h Dj j,τ (x) |dx|,
AG
2 nτ rank W τ
GΣ
τ ∈B αj
where τ ∈ B only if W τ corresponds to irreducible isotropy representations whose

# #
duals are present in E αj , the bundle on which Dαj acts. First, GW 0 = M0 F ,
#
and GΣαj = Mj F . By definition, A0 (x) is the Atiyah-Singer integrand, the lo-
G
cal supertrace of the ordinary heat kernel associated to the elliptic operator induced
from D (blown-up and doubled from D) on the quotient GW 0 , where the bundle
68
E →W is replaced by the bundle of invariant sections of E over each orbit (corre-

#
sponding to a point of GW 0 ). This is precisely the the space of basic sections of
over the corresponding leaf closure (point of M #0 F ), and the operator is the same
E G
as D by construction. Similarly, A is the local supertrace of the ordinary heat
b j,τ
kernel associated to the elliptic operator induced from (1 ⊗ Dαj ) (blown-up and
doubled from 1⊗Dαj , the twist of Dαj by the canonical isotropy bundle W τ → Σαj
) on the quotient GΣαj , where the bundle is replaced by the space of invariant
sections over each orbit. Again, this part of the formula
is exactly
that shown
in the statement of the theorem. The quantities −η DjS+,τ + h DjS+,τ in the
equivariant and basic formulas are the same, since the spherical operator on the
normal bundle to the stratum in the basic manifold is the same as the spherical
operator defined on the normal bundle to the stratum of the Riemannian foliation.
The theorem follows. z
7. The representation-valued basic index theorem

In order to retain the complete information given by Atiyah’s distributional
index of the transversal differential operator D, we need to consider the equivariant
indices indρ (D) associated to any irreducible representation ρ of O (q).
Definition 7.1. The representation-valued basic index of the transversal Dirac
E
operator Dtr is defined as
E
indρb Dtr = indρ (D) .
Using [14, Theorem 9.2], we have the following result. The proof is no different
than that of Theorem 6.1.
Theorem 7.2. (Representation-valued Basic Index Theorem for Riemannian foli-
ations) Let M0 be the principal stratum of the Riemannian foliation (M, F), and
let M1 , ... , Mr denote all the components of all singular strata, corresponding to
O (q)-isotropy types [G1 ], ... ,[Gr ] on the basic manifold. With notation as in the
previous section, we have

ρ E
r
indb Dtr = +
Aρ0 (x) |dx| β (Mj ) ,

M 0 F j=1
1 1
β (Mj ) = S+,τ
−η Dj S+,τ
+ h Dj
Aρj,τ (x) |dx|,
2 τ nτ rank W τ
Mj F
where Aρ0 (x) and Aρj,τ (x) are the local Atiyah-Singer integrands of the operators
#0 and
induced on the leaf closure spaces by extracting the sections of type ρ from M
#
Mj .
8. The basic index theorem for foliations given by suspension

One class of examples of Riemannian foliations are those constructed by sus-
pensions. Let X be a closed manifold with fundamental group π1 (X) , which acts
by deck transformations. Let φ : π1 (X) → Isom (Y ) be
on the universal cover X
a homomorphism to the group of isometries of a closed Riemannian manifold Y .
The suspension is defined to be
X ×φ Y = X × Y ∼,
31

where the equivalence relation is defined by (x, y) ∼ x · g −1 , φ (g) y for any g ∈

π1 (X). The foliation F associated to this suspension is defined by the X-parameter

submanifolds, so that T F agrees with T X over each fundamental domain of X ×φ Y
in X × Y . This foliation is Riemannian, with transverse metric given by the metric
on Y . A transversally-elliptic operator that preserves the foliation is simply an
elliptic operator DY on Y that is G-equivariant, where G = φ (π1 (X)) ⊂ Isom (Y ).
It follows that DY is also equivariant with respect to the action of the closure G, a
compact Lie group. Then we have that the basic index satisfies

Y Y G
indb Db = ind D .
We wish to apply the basic index theorem to this example. Observe that the strata
of the foliation F are determined by the strata of the G-action on Y . Precisely, if
Σα1 , ..., Σαr are the components of the strata of Y relative to G, then each
× Σα ∼
Mj = X j
is a component of a stratum of the foliation (X ×φ Y, F). Similarly, the desingular-

izations of the foliation correspond exactly to the desingularizations of the group
action in the Equivariant Index Theorem ([14]), applied to the G action on Y . By
the basic index theorem,

r

r
indb DbY = A0,b (x) |dx|+ β (Mj ) = 0 (x) |dx|+
AG β (Mj ) ,

M 0 F

Y 0 G
j=1 j=1
where AG Y
0 (x) is the Atiyah-Singer integrand of the operator D on the (blown up)
quotient of the principal stratum of the G-action, where the bundle is the space of
invariant sections on the corresponding orbit. Similarly, the singular terms β (Mj )
are exactly the same as those in the Equivariant Index Theorem, applied to the G
action on Y . Thus, the basic index theorem gives precisely
the same formula as the
Y G
Equivariant Index Theorem calculating the index ind D .
We remark that in this particular case, the basic index may be calculated in an
entirely different way, using the Atiyah-Segal fixed point formula for
G-equivariant
elliptic operators (see [4]). Their formula is a formula for indg DY , the difference

of traces of $the action of g ∈ G on ker DY and ker DY ∗ , and the answer is
an integral Y g αg of characteristic classes over the fixed point set Y g ⊂ Y of the
element g. To extract the invariant part of this index, we would need to calculate

Y G Y
ind D = indg D dg = αg dg,
G G Yg
where dg is the normalized Haar measure. Since the fixed point set changes with g,
the integral above could not be evaluated as above. However, if G is connected, we
could use the Weyl integration formula to change the integral to an integral over a
maximal torus T , and we could replace Y g with the fixed point set Y T , since for
generic g ∈ T , Y g = Y T . Moreover, if G is not connected, one may construct a
suspension Y of the manifold on which a larger connected group G acts such that
G Y = GY .
70
9. An example of transverse signature

In this section we give an example of a transverse signature operator that arises
from an S 1 action on a 5-manifold. This is essentially a modification of an example
from [1, pp. 84ff], and it illustrates the fact that the eta invariant term may be
nonzero. Let Z 4 be a closed, oriented, 4-dimensional Riemannian manifold on which
Zp (p prime > 2) acts by isometries with isolated fixed points xi , i = 1, ..., N . Let
M = Z 4 ×Zp S 1 , where Zp acts on S 1 by rotation by multiples of 2π 1
p . Then S acts
1 ∼ 4
on M , and M S = Z Zp .
Next, let D+ denote the signature operator d + d∗ from self-dual to anti-self-
dual forms on Z 4 ; this induces a transversally elliptic operator (also denoted by
D+ ). Then the S 1 -invariant index of D+ satisfies

indρ0 D+ = Sign M S 1 = Sign Z 4 Zp .
By the Invariant Index Theorem [14, Theorem 9.6] and the fact that the Atiyah-
Singer integrand is the Hirzebruch L-polynomial 13 p1 ,

1
indρ0 (D) = p1
3 M S 1
1 S+,ρ0
N
+ −η Dj + h DjS+,ρ0 ,
2 j=1
where each DiS+,ρ0 is two copies of the boundary signature operator

p
B = (−1) (∗d − d∗)

on 2l-forms (l = 0, 1) on the lens space S 3 Zp . We have h DjS+,ρ0 = 2h (B) = 2
(corresponding to constants), and in [3] the eta invariant is explicitly calculated to
be

2
p−1
kmj π knj π
η DjS+,ρ0 = 2η (B) = − cot cot ,
p p p
k=1
where the action of the generator ζ of Zp on S 3 is
2m πi 2nj πi

j
ζ · (z1 , z2 ) = e p z1 , e p z2 ,
with (mj , p) = (nj , p) = 1. Thus,

1
N p−1
1 kmj π knj π
Sign M S = 1
p1 + cot cot +N
3 Z4 Zp p j=1 p p
k=1
Note that in [1, pp. 84ff] it is shown that

1
N p−1
1 kmj π knj π
Sign M S 1 = p1 + cot cot ,
3 Z 4 Zp p j=1 p p
k=1
which demonstrates that

1 1
p1 − p1 = N,
3 Z 4 Zp 3 4 Zp
Z
# and the original M .

illustrating the difference between the blowup M
33
10. The Basic Euler characteristic

10.1. The Basic Gauss-Bonnet Theorem. Suppose that a smooth, closed
manifold M is endowed with a smooth foliation F.
In the theorem that follows, we express the basic Euler characteristic in terms
of the ordinary Euler characteristic, which in turn can be expressed in terms of
an integral of curvature. We extend the Euler characteristic notation χ (Y ) for Y
any open (noncompact without boundary) or closed (compact without boundary)
manifold to mean
χ (Y ) if Y is closed
χ (Y ) =
χ (1-point compactification of Y ) − 1 if Y is open
Also, if L is a flat foliated line bundle over a Riemannian foliation (X, F), we define
the basic Euler characteristic χ (X, F, L) as before, using the basic cohomology
groups with coefficients in the line bundle L.
Theorem 10.1. (Basic Gauss-Bonnet Theorem, announced in [53]) Let (M, F) be

a Riemannian foliation. Let M0 ,..., Mr be the strata of the Riemannian foliation
(M, F), and let OMj F denote the orientation line bundle of the normal bundle
to F in Mj . Let Lj denote a representative leaf closure in Mj . With notation as
above, the basic Euler characteristic satisfies

χ (M, F) = χ Mj F χ Lj , F, OMj F .
j
Remark 10.2. In [27, Corollary 1], they show that in special cases the only term
that appears is one corresponding to a most singular stratum.
10.1.1. Proof using the basic Hopf index theorem. In this section, we prove
the basic Gauss-Bonnet Theorem using the Hopf index theorem for Riemannian
foliations ([6]).
To find a topological formula for the basic index, we first construct a basic,
normal, F -nondegenerate vector field V on (M, F) and then compute the basic
Euler characteristic from this information. The formula from the main theorem in
[6] is

χ (M, F) = ind (V, L) χ (L, F, OL ) .
L critical
We construct the vector field as follows. First, starting with i = 1 (where the
holonomy is largest, where Mi F is a closed manifold), we triangulate Mi F ∼ =

W (Gi )G, without changing the triangulation of Mi F \ Mi F (to construct
the triangulation, we may first apply the exponential map of Mi to the normal
space to a specific leaf closure of Mi and extend the geodesics to the cut locus,
and so on). The result is a triangulation of M F that restricts to a triangulation
of each Mi F . Next, we assign the value 0 to each vertex of the triangulation
and the value k to a point on the interior of each k-cell, and we smoothly extend
this function to a smooth basic Morse function on all of M whose only critical leaf
closures are each of the points mentioned above. The gradient of this function is a
a basic, normal, F -nondegenerate vector field V on M . Thus, letting Lk denote a
72
leaf closure corresponding to the value k,

χ (M, F) = ind (V, L) χ (L, F, OL )
L critical
k
= (−1) χ (Lk , F, OL )
k Lk

= χ Mi F χ (Li , F, OLi )
i

= χ Mi F χ Li , F, OMi F
i
where Li denotes a representative leaf closure of Mi , and OLi denotes its “neg-
ative direction orientation bundle”, which by the definition
of the vector field is
isomorphic to the orientation bundle OMi F of T Mi F .
10.1.2. Proof using the Basic Index Theorem. In this section, we prove the
basic Gauss-Bonnet Theorem using the Basic Index Theorem (Theorem 6.1).
As explained in Section 2.5 we wish to compute indb (Db ) = indb (Db ), with
1
Db = d + δb ; Db = Db − (κb ∧ +κb ) .
2
Let M0 be the principal stratum of the Riemannian foliation (M, F), and let M1 ,
... , Mr denote all the components of all singular strata, corresponding to O (q)-
isotropy types [G1 ], ... ,[Gr ] on the basic manifold. At each Mj , we may write the
basic de Rham operator (up to lower order perturbations) as
Db = DNj ∗ DMj ,
where DNj is in fact the de Rham
operator
on the vertical forms, and DMj is the
basic de Rham operator on Mj , F|Mj . Further, the spherical operator DjS in
the main theorem is simply
DjS = −c (∂r ) (d + d∗ ) , c (∂r ) = dr ∧ −dr ,
S
where (d + d∗ ) is a vector-valued de Rham operator on the sphere (normal to Mj )

S
and r is the radial distance from Mj . Weperformed

a similar calculation in [14,
Section 10.2], and the results are that η DjS+,σ = 0 for all Gj -representation
types [σ] and
⎧
⎪ 2 if σ = 1 and Gj preserves orientation
⎪⎨
1 if σ = 1 and Gj does not preserve orientation
S+,σ
(10.1) h Dj =
⎪
⎪ 1 if σ = ξGj and Gj does not preserve orientation
⎩
0 otherwise
Here, if some elements of Gj reverse orientation of the normal bundle, then ξGj
denotes the relevant one-dimensional representation of Gj as ±1. The orientation
line bundle OMi F → Mj of the normal bundle to Mj is a pointwise representation
space for the representation ξGj . After pulling back to and pushing forward
to the

basic manifold, it is the canonical isotropy G-bundle W b corresponding to j, ξGj .
We may also take it to be a representation bundle for the trivial Gj -representation
1 (although the trivial line bundle is the canonical one). The Basic Index Theorem
35
takes the form

+
r
indb DbE = A0,b (x) |dx| β (Mj )

M 0 F j=1

1 S+,ξGj

β (Mj ) = h Dj + +h DjS+,1 Aj,b x, OMj F |dx|
2 j j F
M

=
Aj,b x, OMj F |dx|.
j F
M
j
$
We rewrite Aj,b x, OMj F |dx|
j F
as $ Kj x, O before taking
|dx|
Mj F
M

Mj
it to the quotient. We see that Kj x, OMj F is the Gauss-Bonnet integrand on

the desingularized stratum M #j , restricted to O
Mj F -twisted

basic forms. The
result is the relative Euler characteristicχ Lj , F, OMj F

Kj x, OMj F |dx| = χ Mj , lower strata, F, O
Mj F ,
j
M
Here, the relative basic Euler characteristic is defined for X a closed subset of a
manifold Y as χ (Y, X, F, V) = χ (Y, F, V) − χ (X, F, V), which is also the alternat-
ing sum of the dimensions of the relative basic cohomology groups with coefficients
in a complex vector bundle V → Y . Since Mj is a fiber bundle over Mj F with
fiber Lj (a representative leaf closure), we have

= χ Lj , F, O
Kj x, OMj F |dx| Mj F χ Mj F, lower strataF ,
j
M
by the formula for the Euler characteristic on fiber bundles, which extends naturally
to the current situation. The Basic Gauss-Bonnet Theorem follows.
10.1.3. The representation-valued basic Euler characteristic. Taking into con-
sideration the Representation-valued Basic Index Theorem (Theorem 7.2), we may
use the arguments in the previous section to derive a formula for the basic Euler
characteristic of basic forms twisted by a representation of O (q). Since the proof
is nearly the same, we simply state the result.
Theorem 10.3. (Representation-valued Basic Gauss-Bonnet Theorem) Let (M, F)
be a Riemannian foliation. Let M0 ,..., Mr be the strata of the Riemannian foliation
(M, F), and let OMj F denote the orientation line bundle of the normal bundle
to F in Mj . Let Lj denote a representative leaf closure in Mj . For (X, FX ) a
Riemannian foliation of codimension q, let χρ (X, FX , V) denote the index of the
basic de Rham operator twisted by a representation ρ : O (q) → U (Vρ ) with values
in the flat line bundleV. Then the basic Euler characteristic satisfies

χρ (M, F) = χ Mj F χρ Lj , F, OMj F .
j
10.2. Examples of the basic Euler characteristic. In addition to the

examples in this section, we refer the reader to [29], where in some nontaut Rie-
mannian foliations, the basic Euler characteristic and basic cohomology groups and
twisted basic cohomology groups are computed using the theorems in this paper.
74
The first example is a codimension 2 foliation on a 3-manifold. Here, O(2)

acts on the basic manifold, which is homeomorphic to a sphere. In this case, the
principal orbits have isotropy type ({e}), and the two fixed points obviously have
isotropy type (O(2)). In this example, the isotropy types correspond precisely to
the infinitesimal holonomy groups.
Example 10.4. (This example is taken from [51] and [55].) Consider the one
dimensional foliation obtained by suspending an irrational rotation on the stan-
dard unit sphere S 2 . On S 2 we use the cylindrical
) coordinates (z,
) θ), related to
the standard rectangular coordinates by x = (1 − z 2 ) cos θ, y = (1 − z 2 ) sin θ,
z = z. Let α be an irrational multiple of 2π, and let the three–manifold M =
S 2 × [0, 1] / ∼, where (z, θ, 0) ∼ (z, θ + α, 1). Endow M with the product metric on
Tz,θ,t M ∼= Tz,θ S 2 × Tt R. Let the foliation F be defined by the immersed submani-
folds Lz,θ = ∪n∈Z {z} × {θ + α} × [0, 1] (not unique in θ). The leaf closures Lz for
|z| < 1 are two dimensional, and the closures corresponding to the poles (z = ±1)
are one dimensional. *
The stratification of (M, F) is M (H1 ) M (H2 ), where M (H1 ) is the union of the
two “polar” leaves (z = ±1), and M (H2 ) is the complement of M (H1 ). Note that
each orientation bundle OM (Hi )F is trivial. Next,

χ M (H2 ) F = χ (open interval) = −1,

and χ M (H1 ) F = χ (disjoint union of two points) = 2, while

χ L1 , F, OM (H1 )F = χ (L1 , F) = χ S 1 , S 1 = 1.

However, χ L2 , F, OM (H2 )F = χ (L2 , F) = 0, since every such leaf closure is
a flat torus, on which the foliation restricts to be the irrational flow and since the
vector field ∂θ is basic, nonsingular, and orthogonal to the foliation on this torus.
By our theorem, we conclude that

χ (M, F) = χ M (Hi ) F χ Li , F, OM (Hi )F
i
= 2 · 1 + (−1) · 0 = 2.
We now directly calculate the Euler characteristic of this foliation. Since the
foliation is taut, the standard Poincare duality works [35] [36] , and Hb0 (M ) ∼ =
Hb2 (M ) ∼= R . It suffices to check the dimension h1 of the cohomology group
Hb1 (M ). Then the basic Euler characteristic is χ (M, F) = 1 − h1 + 1 = 2 − h1 .
Smooth basic functions are of the form f (z), where f (z) is smooth in z for −1 <
f1 12 − z near z = 1 for a smooth function
2
z < 1 and is of the form f (z) =
f1 and is of the form f (z) = f2 1 − z near z = −1 for a smooth function f2 .
Smooth basic one forms are of the form α = g (z) dz + k (z) dθ, where g (z) and
k (z) are smooth functions for −1 < z < 1 and satisfy

g (z) = g1 1 − z 2 and

(10.2) k (z) = 1 − z 2 k1 1 − z 2
near z = 1 and

g (z) = g2 1 − z 2 and

k (z) = 1 − z 2 k2 1 − z 2
37
near z = −1 for smooth functions g1 , g2 , k1 , k2 . A simple calculation shows that

ker d1 = im d0 , so that h1 = 0. Thus, χ (M, F) = 2. This example shows that the
orbit space can be dimension 1 (odd) and yet have nontrivial index.
The next example is a codimension 3 Riemannian foliation for which all of
the infinitesimal holonomy groups are trivial; moreover, the leaves are all simply
connected. There are leaf closures of codimension 2 and codimension 1. The codi-
mension 1 leaf closures correspond to isotropy type (e) on the basic manifold, and
the codimension 2 leaf closures correspond to an isotropy type (O(2)) on the ba-
sic manifold. In some sense, the isotropy type measures the holonomy of the leaf
closure in this case.
Example 10.5. This foliation is a suspension of an irrational rotation of S 1 com-
posed with an irrational rotation of S 2 on the manifold S 1 × S 2 . As in Exam-
ple 10.4, on S 2 we use the cylindrical
) coordinates (z,
)θ), related to the standard
rectangular coordinates by x = (1 − z 2 ) cos θ, y = (1 − z 2 ) sin θ, z = z. Let
α be an irrational multiple of 2π, and let β be any irrational number. We con-
sider the four–manifold M = S 2 × [0, 1] × [0, 1] / ∼, where (z, θ, 0, t) ∼ (z, θ, 1, t),
(z, θ, s, 0) ∼ (z, θ + α, s + β mod 1, 1). Endow M with the product metric on
Tz,θ,s,t M ∼= Tz,θ S 2 × Ts R × Tt R. Let the foliation F be defined by the immersed
submanifolds Lz,θ,s = ∪n∈Z {z} × {θ + α} × {s + β} × [0, 1] (not unique in θ or s).
The leaf closures Lz for |z| < 1 are three–dimensional, and the closures correspond-
ing to the poles (z = ±1) are two–dimensional. The basic forms in the various
dimensions are:
Ω0b = {f (z)}

Ω1b = g1 (z) dz + 1 − z 2 g2 (z)dθ + g3 (z) ds

Ω2b = h1 (z) dz ∧ dθ + 1 − z 2 h2 (z)dθ ∧ ds + h3 (z) dz ∧ ds
Ω3b = {k (z) dz ∧ dθ ∧ ds} ,
where all of the functions above are smooth in a neighborhood of [0, 1]. An elemen-
tary calculation shows that h0 = h1 = h2 = h3 = 1, so that χ (M, F) = 0.
We now compute the basic
* Euler characteristic using our theorem. The stratifi-
cation of (M, F) is M (H1 ) M (H2 ), where M (H1 ) is the union of the two “polar”
leaf closures (z = ±1) , and M (H2 ) is the complement
of M(H1 ). Note that each
orientation bundle OM (Hi )F is trivial. Next, χ M (H2 ) F = χ (open interval) =

−1, and χ M (H 1 ) F = χ (disjoint
union of two points) = 2.
Observe that χ L1 , F, OM (H1 )F = χ (L1 , F) = 0, since this is a taut, codimension-

1 foliation. Also, χ L2 , F, OM (H2 )F = χ (L2 , F) = 1 − 2 + 1 = 0, since
the basic forms restricted to L2 consist of the span of the set of closed forms
{1, dθ, ds, dθ ∧ ds}. Thus,

i
= 2 · 0 + (−1) · 0 = 0,
as we have already seen.
Note that taut foliations of odd codimension will always have a zero Euler char-
acteristic, by Poincare duality. Open Question: will these foliations always have a
zero basic index?
76
The following example is a codimension two transversally oriented Riemannian

foliation in which all the leaf closures have codimension one. The leaf closure
foliation is not transversally orientable, and the basic manifold is a flat Klein bottle
with an O(2)–action. The two leaf closures with Z2 holonomy correspond to the
two orbits of type (Z2 ), and the other orbits have trivial isotropy.
Example 10.6. This foliation is the suspension of an irrational rotation of the
flat torus and a Z2 –action. Let X be any closed Riemannian manifold such that
π1 (X) = Z ∗ Z , the free group on two generators {α, β}. We normalize the volume
of X to be 1. Let X be the universal cover. We define M = X × S 1 × S 1 π1 (X),
where π1 (X) acts and by α (θ, φ) = (2π − θ, 2π − φ)
by deck√ transformations
on X
and β (θ, φ) = θ, φ + 2π on S 1 × S 1 . We use the
standard product–type
metric.
The leaves of F are defined to be sets of the form (x, θ, φ)∼ | x ∈ X . Note that
the foliation is transversally oriented. The leaf closures are sets of the form

Lθ = (x, θ, φ)∼ | x ∈ X, φ ∈ [0, 2π] φ ∈ [0, 2π]
(x, 2π − θ, φ)∼ | x ∈ X,
The basic forms are:

Ω0b = {f (θ)}
Ω1b = {g1 (θ) dθ + g2 (θ)dφ}
Ω2b = {h(θ)dθ ∧ dφ} ,
where the functions are smooth and satisfy
f (2π − θ) = f (θ)
gi (2π − θ) = −gi (θ)
h (2π − θ) = h (θ) .
A simple argument shows that h0 = h2 = 1 and h1 = 0. Thus, χ (M, F) = 2. The
basic manifold W is an O(2)–manifold, defined by W = [0, π] × S 1 ∼ , where the
circle has length 1 and (θ = 0 or π, γ) ∼ (θ = 0 or π, −γ). This is a Klein bottle,
since it is the connected sum of two projective planes. O(2) acts on W via the usual
1
action on S .
Next, we compute the basic
* Euler characteristic using our theorem. The strati-
fication of (M, F) is M (H1 ) M (H2 ), where M (H1 ) is the union of the two leaf
closures θ2 = 0 and θ2 = π, and M (H2 ) is the complement of M (H1 ). Note
that the orientation bundle OM (H2 )F is trivial since an interval is orientable, and
OM (H1 )F is trivial even though those leaf closures are not transversally oriented
(since the points are oriented!). Next,

χ M (H2 ) F = χ (open interval) = −1,
and

χ M (H1 ) F = χ (disjoint union of two points) = 2.

Observe that χ L2 , F, OM (H2 )F = χ (L2 , F) = 0, since each representative leaf
L2 is a taut (since it is a suspension), codimension 1 foliation, and thus ordinary
dim HB (L2 , F) = dim HB (L2 , F) = 1. On the
0 1
Poincare duality
holds ([58],[49]):
other hand, χ L1 , F, OM (H1 )F = χ (L1 , F) = 1, since each such leaf closure has
39
dim HB0
(L1 , F) = 1 but dim HB
1
(L1 , F) = 0 since there are no basic one-forms. By
our theorem, we conclude that

χ (Y, F) = χ M (Hi ) F χ Li , F, OM (Hi )F
i
= 2 · 1 + (−1) · 0 = 2,
as we found before by direct calculation.
The next example is a codimension two Riemannian foliation with dense leaves,
such that some leaves have holonomy but most do not. The basic manifold is a
point, the fixed point set of the O (2) action. The isotropy group O(2) measures
the holonomy of some of the leaves contained in the leaf closure.
Example 10.7. This Riemannian foliation is a suspension of a pair of rotations of
the sphere S 2 . Let X be any closed Riemannian manifold such that π1 (X) = Z ∗ Z ,
that is the free group on two generators {α, β}. We normalize the volume of X to
be 1. Let X be the universal cover. We define M = X × S 2 π1 (X). The group

π1 (X) acts by deck transformations on X and by rotations on S 2 in the following
ways. Thinking of S 2 as imbedded in R3 , let α act by an irrational rotation around
the z–axis, and let β act by an irrational rotation around the x–axis. We use the
standard
product–type metric.
As usual, the leaves of F are defined to be sets of
the form (x, v)∼ | x ∈ X . Note that the foliation is transversally oriented, and a
generic leaf is simply connected and thus has trivial holonomy. Also, the every leaf
is dense. The leaves {(x, (1, 0, 0))∼ } and {(x, (0, 0, 1))∼ } have nontrivial holonomy;
the closures of their infinitesimal holonomy groups are copies of SO(2). Thus, a
leaf closure in M covering the leaf closure M has structure group SO(2) and is thus

all of M , so that W is a point. The only basic forms are constants and 2 forms
of the form CdV , where C is a constant and dV is the volume form on S 2 . Thus
h0 = h2 = 1 and h1 = 0, so that χ (M, F) = 2.
Our theorem in this case, since there is only one stratum, is

i
= χ (point) χ (M, F)
= χ (M, F) ,
which is perhaps not very enlightening.
The following example is a codimension two Riemannian foliation that is not
taut. This example is in [16].
Example 10.8. Consider the flat torus T 2 = R2 Z2 . Consider the map F : T 2 →
T 2 defined by
x 2 1 x
F = mod 1
y 1 1 y
Let M = [0, 1] × T 2 ∼, where (0, a) ∼ (1, F (a)). Let v, v be orthonormal
√ √
eigenvectors of the matrix above, corresponding to the eigenvalues 3+2 5 , 3−2 5 ,
respectively. Let the linear foliation F be defined by the vector v on each copy of
T 2 . Notice that every leaf is simply connected and that the leaf closures are of the
form {t} × T 2 , and this foliation is Riemannian if we choose a suitable metric.
78
For example, we choose the metric along [0, 1] to be standard and require each
torus to be orthogonal to this direction. Then we define the vectors v and v to be
of v and v vary smoothly over [0, 1] so
orthogonal in this√metric and let the lengths √
that v(0) = 2 v(1) and v (0) = 2 5 v (1). Let v = a (t) v, v = b (t) v
3+ 5 3−
be the resulting renormalized vector fields. The basic manifold is a torus, and the
isotropy groups are all trivial. We use coordinates (t, x, y) ∈ [0, 1] × T 2 to describe
points of M . The basic forms are:
Ω0b = {f (t)}
Ω1b = {g1 (t) dt + g2 (t)v∗ }
Ω2b = {h(t)dt ∧ v ∗ } ,

where all the functions are smooth. Note that dv ∗ = − aa(t)
(t)
dt∧v ∗ By computing the
cohomology groups, we get h0 = h1 = 1, h2 = 0. Thus, the basic Euler characteristic
is zero.
We now compute the basic Euler characteristic using our theorem. There is
only one stratum, and the leaf closure space is S 1 . The foliation restricted to each
leaf closure is an irrational flow on the torus. Thus,

i

= χ S 1 χ {t} × T 2 , F
= 0 · 0 = 0,
as we have already seen.
Following is an example of using the representation-valued basic index theorem,
in this case applied to the Euler characteristic (Theorem 10.3).
2 of the torus T π= R Z ,
Example 10.9. Let M = R ×φ T 2 be the suspension 2 2 2
constructed as follows. The action φ : Z → Isom T is generated by a 2 rotation.

The Riemannian
foliation F is given by the R-parameter curves. Explicitly, k ∈ Z
y1
acts on by
y2
k
y1 0 −1 y1
φ (k) = .
y2 1 0 y2
Endow T 2 with the standard flat metric. The basic harmonic forms have basis
{1, dy1 , dy2 , dy1 ∧ dy2 }. Let ρj be the irreducible character defined by k ∈ Z →
ρ
eikjπ/2 . Then the basic de Rham operator (d + δb ) 0 on Z-invariant basic forms has
ρ
kernel {c0 + c1 dy1 ∧ dy2 : c0 , c1 ∈ C}. It is also not hard to see that ker (d + δb ) 1 =
ρ2 ρ3
span {idy1 + dy2 }, ker (d + δb ) = {0}, and ker (d + δb ) = span {−idy1 + dy2 }.
Then
χρ0 (M, F) = 2, χρ1 (M, F) = χρ3 (M, F) = −1, χρ2 (M, F) = 0.
This illustrates the point that it is not possible to use the Atiyah-Singer integrand
on the quotient of the principal stratum to compute even the invariant index alone.
Indeed, the Atiyah-Singer integrand would be a constant times the Gauss curvature,

0
which is identically zero. In these cases, the three singular points a1 = , a2 =
0
41
1

0
1 , a3 = 2
1 certainly contribute to the index. The quotient M F is an
2 2
orbifold homeomorphic to a sphere.
We now compute the Euler characteristics χρ (M, F) using Theorem 10.3. The
strata of the foliation are as follows. The leaves corresponding to a1 and a3 comprise
the most singular stratum Ms with isotropy Z4 , and the leaf correspondng to a2 is
its own stratum Ml with isotropy isomorphic to Z2 . Then

χ Ms F = 2,

χ Ml F = 1,

χ M0 F = χ S 2 {3 points} = −1.
In each stratum (M0 , Ml , or Ms ), the representative leaf closure is a circle,
a single leaf, and each stratum is transversally oriented. The Euler characteristic
χρ (Lj , F) is one if there exists a locally constant section of the line bundle associ-
ated to ρ over Lj , and otherwise it is zero. We see that
⎧
⎪
⎪ 1 if Mj = M0 and ρ = ρ0 , ρ1 , ρ2 , or ρ3
⎨
1 if Mj = Ms and ρ = ρ0
χρ Lj , F, OMj F = χρ (Lj , F) = .
⎪
⎪ 1 if Mj = Ml and ρ = ρ0 or ρ2
⎩
0 otherwise
Then Theorem 10.3 implies
+ +
ρ 1 if ρ = ρ0 , ρ1 , ρ2 , or ρ3 1 if ρ = ρ0 or ρ2
χ (M ) = (−1) + (1)
0 otherwise 0 otherwise
+
1 if ρ = ρ0
+ (2)
0 otherwise
⎧
⎪
⎪ −1 if Mj = M0 and ρ = ρ1 or ρ3
⎨
2 if Mj = Ms and ρ = ρ0
= ,
⎪
⎪ 0 if Mj = Ml and ρ = ρ2
⎩
0 otherwise
which agrees with the previous direct calculation.
References
[1] M. F. Atiyah, Elliptic operators and compact groups, Lecture Notes in Mathematics 401,
Berlin: Springer-Verlag, 1974.
[2] M. F. Atiyah, V. K. Patodi, and I. M. Singer, Spectral asymmetry and Riemannian geometry.
I, Math. Proc. Camb. Phil. Soc. 77 (1975), 43–69.
[3] M. F. Atiyah, V. K. Patodi, and I. M. Singer, Spectral asymmetry and Riemannian geometry.
II, Math. Proc. Camb. Phil. Soc. 78 (1975), 405–432.
[4] M. F. Atiyah and G. B. Segal, The index of elliptic operators: II, Ann. of Math. (2) 87(1968),
531–545.
[5] M. F. Atiyah and I. M. Singer, The index of elliptic operators: III, Ann. of Math. (2)
87(1968), 546–604.
[6] V. Belfi, E. Park, and K. Richardson, A Hopf index theorem for foliations, Diff. Geom. Appl.
18(2003), no. 3, 319-341.
[7] B. Booss-Bavnbek, Bernhelm and M. Lesch, The invertible double of elliptic operators, Lett.
Math. Phys. 87 (2009), no. 1-2, 19–46.
[8] B. Booss-Bavnbek and K. P. Wojciechowski, Elliptic boundary problems for Dirac operators,
Boston: Birkhäuser, 1993.
[9] G. Bredon, Introduction to Compact Transformation Groups, New York:Academic Press,
1972.
80
[10] T. Bröcker and T. tom Dieck, Representations of Compact Lie Groups, Graduate Texts in
Math. 98, Springer-Verlag, New York 1985.
[11] J. Brüning, A localization method for asymptotic expansions on manifolds with singularities,
Humboldt University, Berlin, preprint 2000.
[12] J. Brüning and E. Heintze, Representations of compact Lie groups and elliptic operators, Inv.
Math. 50 (1979), 169-203.
[13] J. Brüning and E. Heintze, The asymptotic expansion of Minakshisundaram–Pleijel in the
equivariant case, Duke Math. Jour. 51 (1984), 959-979.
[14] J. Brüning, F. W. Kamber, and K. Richardson, The eta invariant and equivariant index of
transversally elliptic operators, preprint arXiv:1005.3845v1 [math.DG].
[15] J. Brüning and R. Seeley, Regular singular asymptotics, Adv. in Math. 58(1985), 133-148.
[16] Y. Carrière, Flots riemanniens, in Transversal structure of foliations (Toulouse, 1982),
Astérisque 116(1984), 31–52.
[17] A. Connes and H. Moscovici, Hopf algebras, cyclic cohomology and the transverse index
theorem, Comm. Math. Phys. 198 (1998), no. 1, 199–246.
[18] D. Domı́nguez, A tenseness theorem for Riemannian foliations, C. R. Acad. Sci. Sér. I 320
(1995), 1331-1335.
[19] H. Donnelly, Eta invariants for G-spaces, Indiana Univ. Math. J. 27(1978), no. 6, 889-918.
[20] R. G. Douglas, J. F. Glazebrook, F. W. Kamber, and G. L. Yu, Index formulas for geometric
Dirac operators in Riemannian foliations, K-Theory 9 (1995), no. 5, 407–441.
[21] A. El Kacimi-Alaoui, Opérateurs transversalement elliptiques sur un feuilletage riemannien
et applications, Compositio Math. 73(1990), 57–106.
[22] A. El Kacimi, G. Hector, V. Sergiescu, La cohomologie basique d’un feuilletage riemannien
est de dimension finie, Math. Z. 188(1985), 593–599.
[23] A. El Kacimi-Alaoui and M. Nicolau, On the topological invariance of the basic cohomology,
Math. Ann. 295 (1993), no. 4, 627–634.
[24] D. B. A. Epstein, Foliations with all leaves compact, Ann. Inst. Fourier (Grenoble) 26 (1976),
no. 1, viii, 265–282.
[25] E. Ghys, Un feuilletage analytique dont la cohomologie basique est de dimension infinie,
Publ. de l’IRMA, Lille 7(1985).
[26] J. F. Glazebrook and F. W. Kamber, Transversal Dirac families in Riemannian foliations,
Comm. Math. Phys. 140(1991), 217-240.
[27] A. Gorokhovsky and J. Lott, The index of a transverse Dirac-type operator: the case of
abelian Molino sheaf, preprint arXiv:1005.0161v2 [math.DG].
[28] G. Habib and K. Richardson, A brief note on the spectrum of the basic Dirac operator, Bull.
London Math. Soc. 41 (2009), 683-690.
[29] G. Habib and K. Richardson, Modified differentials and basic cohomology for Riemannian
foliations, preprint.
[30] S.D. Jung, The first eigenvalue of the transversal Dirac operator, J. Geom. Phys. 39 (2001),
253–264.
[31] S. D. Jung and K. Richardson, Transverse conformal Killing forms and a Gallot-Meyer
Theorem for foliations, preprint arXiv:0805.4187v1 [math.DG].
[32] F. W. Kamber and K. Richardson, G-equivariant vector bundles on manifolds of one orbit
type, preprint.
[33] F. W. Kamber and Ph. Tondeur, Foliated bundles and characteristic classes, Lecture Notes
in Math. 493, Springer-Verlag, Berlin–New York 1975.
[34] F. W. Kamber and Ph. Tondeur, Foliations and metrics, Differential geometry (College Park,
Md., 1981/1982), 103–152, Progr. Math., 32, Birkhäuser Boston, Boston, MA, 1983.
[35] F. W. Kamber and Ph. Tondeur, Dualité de Poincaré pour les feuilletages harmoniques, C.
R. Acad. Sci. Paris 294 (1982), 357–359.
[36] F. W. Kamber and Ph. Tondeur, Duality for Riemannian foliations, Proc. Symp. Pure Math.
A.M.S. 40/1 (1983), 609–618.
[37] F. W. Kamber and Ph. Tondeur, De Rham-Hodge theory for Riemannian foliations, Math.
Ann. 277 (1987), 415–431.
[38] K. Kawakubo, The Theory of Transformation Groups, Oxford University Press, 1991.
[39] T. Kawasaki, The signature theorem for V -manifolds, Topology 17 (1978), 75–83.
[40] T. Kawasaki, The index of elliptic operators over V -manifolds, Nagoya Math. J. 84 (1981),
135–157.
43
[41] Y. Kordyukov, Index theory and noncommutative geometry on manifolds with foliations,
Uspekhi Mat. Nauk 64 (2009), no. 2(386), 73–202; translation in Russian Math. Surveys 64
(2009), no. 2, 273–391.
[42] H. B. Lawson and M.-L. Michelsohn, Spin Geometry, Princeton University Press, Princeton,
1989.
[43] J. Lee and K. Richardson, Lichnerowicz and Obata theorems for foliations, Pacific J. Math.
206 (2002), 339-357.
[44] J. Lee and K. Richardson, Riemannian foliations and eigenvalue comparison, Ann. Global
Anal. Geom. 16 (1998), 497-525.
[45] P. March, M. Min-Oo, E. Ruh, Mean curvature of Riemannian foliations, Canad. Math. Bull.
39 (1996), no. 1, 95–105.
[46] X. Masa, Duality and minimality in Riemannian foliations, Comment. Math. Helv. 67 (1992),
no. 1, 17–27.
[47] A. Mason, An application of stochastic flows to Riemannian foliations, Houston J. Math. 26
(2000), 481-515.
[48] P. Molino, Riemannian foliations, Progress in Mathematics 73, Birkhäuser, Basel, 2000.
[49] E. Park and K. Richardson, The basic Laplacian of a Riemannian foliation, Amer. J. Math.
118(1996), 1249–1275.
[50] B. Reinhart, Differential geometry of foliations — the fundamental integrability problem,
Ergebnisse der Mathematik und ihrer Grenzgebiete 99, Springer-Verlag, Berlin, 1983.
[51] K. Richardson, Asymptotics of heat kernels on Riemannian foliations, Geom. Funct. Anal.
8(1998), 356–401.
[52] K. Richardson, The transverse geometry of G-manifolds and Riemannian foliations, Illinois
J. Math. 45 (2001), no. 2, 517–535.
[53] K. Richardson, Generalized equivariant index theory, Foliations 2005, World Sci. Publ., Hack-
ensack, NJ, 2006, 373–388.
[54] K. Richardson, Asymptotics of heat kernels on Riemannian foliations, Geom. Funct. Anal.
8 (1998), 356–401.
[55] K. Richardson, Traces of heat operators on Riemannian foliations, Trans. Amer. Math. Soc.
362 (2010), no. 5, 2301-2337.
[56] H. Rummler, Quelques notions simples en géométrie riemannienne et leurs applications aux
feuilletages compacts, Comm. Math. Helv. 54 (1979), 224–239.
[57] I. Satake, The Gauss-Bonnet theorem for V –manifolds, J. Math. Soc. Japan 9 (1957), 464–
492.
[58] Ph. Tondeur, Geometry of foliations, Monographs in Mathematics 90, Birkhäuser Verlag,
Basel 1997.
[59] Winkelnkemper, H. E., The graph of a foliation, Ann. Global Anal. Geom. 1(1983), 51-75.
Institut für Mathematik, Humboldt Universität zu Berlin, Unter den Linden 6,

D-10099 Berlin, Germany
E-mail address, J. Brüning: bruening@mathematik.hu-berlin.de
Department of Mathematics, University of Illinois, 1409 W. Green Street, Urbana,

IL 61801, USA
E-mail address, F. W. Kamber: kamber@math.uiuc.edu
Department of Mathematics, Texas Christian University, Fort Worth, Texas 76129,

USA
E-mail address, K. Richardson: k.richardson@tcu.edu
Volume 546, 2011
The Witt construction in characteristic one and

Quantization
Alain Connes
Dedicated to Henri Moscovici, with admiration and friendship
Abstract. We develop the analogue of the Witt construction in characteristic

one. We construct a functor from pairs (R, ρ) of a perfect semi-ring R of
characteristic one and an element ρ > 1 of R to real Banach algebras. We find
that the entropy function occurs uniquely as the analogue of the Teichmüller
polynomials in characteristic one. We then apply the construction to the
semi-field Rmax
+ which plays a central role in idempotent analysis and tropical
geometry. Our construction gives the inverse process of the “dequantization”
and provides a first hint towards an extension Run of the field of real numbers
relevant both in number theory and quantum physics.
1. Foreword
The celebration of Henri’s 65th birthday gives me a long awaited occasion to
express my deep gratitude to him, for his indefectible friendship in our long journey
through mathematics, since the first time we met in Princeton in the fall of 1978.
Besides the great enlightening moments, those that I cherish most are the times
when we both knew we were close to some “real stuff” but also knew that we could
get there only at the price of time consuming efforts which we shared so happily
over the years.
Contents
1. Foreword
2. Introduction
3. Sum of Teichmüller representatives
4. Characteristic one
5. Entropy and the w(α)
6. Analogue of the Witt construction in characteristic one
7. Towards Run
References
2010 Mathematics Subject Classification. 13F35, 28D20, 58D30.
Key words and phrases. Witt construction, Quantization, Entropy.
c 2011
XXXX
1
83
84
2 ALAIN CONNES
2. Introduction
The goal of this paper is to develop an analogue of the Witt construction in
the case of characteristic one, which was initiated in [1]. Our starting point is a
formula which goes back to Teichmüller and which gives an explicit expression for
the sum of the multiplicative lifts in the context of strict p-rings. A ring R is a
strict p-ring when R is complete and Hausdorff with respect to the p-adic topology,
p is not a zero-divisor in R, and the residue ring K = R/pR is perfect. The ring R
is uniquely determined by K up to canonical isomorphism and there exists a unique
multiplicative section τ : K → R of the residue morphism : R → K = R/pR
(1) τ : K → R, ◦ τ = id, τ (xy) = τ (x)τ (y) , ∀x, y ∈ K.
Every element x of R can be written uniquely in the form

(2) x= τ (xn )pn xn ∈ K
which gives a canonical bijection τ̃ : K[[T ]] → R such that

(3) τ̃ ( xn T n ) = τ (xn )pn .
The formula which goes back to Teichmüller [20] allows one to express the sum of
two (or more) multiplicative lifts in the form
⎛ ⎞

(4) τ (x) + τ (y) = τ̃ ⎝ α
wp (α, T ) x y 1−α ⎠
α∈Ip
In this equation the sum inside the parenthesis in the right hand side takes place
in K[[T ]], the variable α ranges in
(5) Ip = {α ∈ Q ∩ [0, 1], ∃n, pn α ∈ Z}
so that, since K is perfect, the terms xα y 1−α make sense. Finally the terms
(6) wp (α, T ) ∈ Fp [[T ]] , ∀α ∈ Ip
only depend on the prime p and tend to zero at infinity in Fp [[T ]], for the discrete
topology in Ip , so that the sum (4) is convergent. The formula (4) easily extends
to express the sum of n multiplicative lifts as
⎛ ⎞
α
(7) τ (xj ) = τ̃ ⎝ wp (α1 , . . . , αn , T ) xj j ⎠

αj ∈Ip , αj =1
As is well known the algebraic structure of R was functorially reconstructed from

that of K by Witt who showed that the algebraic rules in R are polynomial in terms
of the components Xn
−n n
(8) x= τ (Xnp )pn , Xn = xpn
which makes sense since K is perfect. One can in fact, as already noted by Te-
ichmüller in [20], also reconstruct the full algebraic structure of R as a deformation
of K depending upon the parameter T using the above formula (7) but the corre-
sponding algebraic relations are not so simple to handle, mostly because the map
x → xα from K to K is an automorphism only when α is a power of p. As we
shall now explain, this difficulty disappears in the limit case of characteristic one.
The framework we use for characteristic one is that of semi-rings (cf. [11]). For
multiplicatively cancellative semi-rings S of characteristic one the maps x → xn
THE WITT CONSTRUCTION IN CHARACTERISTIC ONE AND QUANTIZATION 85
3
are injective endomorphisms for any positive integer n (cf. [11]) and it is natural
to say that S is perfect when these maps are surjective. It then follows that the
fractional powers x → xα make sense for any α ∈ Q∗+ and define automorphisms
θα ∈ Aut(S). We shall show in §5 how to solve the functional equation on the
coefficients w(α) defined for α ∈ I¯ = Q ∩ [0, 1] which ensures that the following
analogue of (4) defines a deformation of S into a semi-ring of characteristic zero

(9) x +w y = w(α) xα y 1−α
α∈I¯
We let w(0) = w(1) = 1 and I = Q∩(0, 1) so that I¯ = I ∪{0, 1}. The commutativity
of the operation +w means that
(10) w(1 − α) = w(α) ,
and the associativity means that the equality
α2 1−α1
(11) w(3) (α1 , α2 , α3 ) = w(α1 )w( ) , ∀αj ∈ I | αj = 1
1 − α1
defines a symmetric function on the simplex Σ3 where, more generally,

(12) Σn = {(α1 , . . . , αn ) ∈ I n | αj = 1}
Note that (10) means that the function w(2) (α1 , α2 ) = w(α1 ) is symmetric on Σ2 .
The equations of symmetry of w(n) , n = 2, 3 only use the multiplicative structure
of S and thus continue to make sense for any map w : I → G where G is a
uniquely divisible abelian group (denoted multiplicatively). We take G = S × the
multiplicative group of invertible elements of S. We show in Theorem 5.2 that all
solutions of these equations (the symmetry of w(n) , n = 2, 3) are of the form
(13) w(α) = χ(α)α χ(1 − α)1−α , ∀α ∈ I
where χ is a homomorphism Q× + → G. Thus in this generality one can give arbi-
trarily the value of χ(p) ∈ G for all primes p. But the group of invertible elements
of S admits an additional structure: the partial order coming from the additive
structure of S. We show in Theorem 5.4 that, provided the θs extend by continuity
to s ∈ R∗+ , all solutions which fulfill the inequality
(14) w(α) ≥ 1, ∀α ∈ Q ∩ [0, 1] ,
are of the form
(15) w(α) = ρS(α) , S(α) = −α log(α) − (1 − α) log(1 − α).
where ρ ∈ S, ρ ≥ 1. One recognizes the entropy function
(16) S(α) = −α log(α) − (1 − α) log(1 − α)
which is familiar in thermodynamics, information theory and ergodic theory. In §6
we construct a functor W from pairs (R, ρ) of a multiplicatively cancellative perfect
semi-ring R of characteristic one and an invertible element ρ > 1 in R to algebras
over R. The construction of the algebra W (R, ρ) involves several operations
• A completion with respect to a ρ-adic distance canonically associated to
ρ.
• A deformation of the addition involving the w(α) = ρS(α) .
• A symmetrization to obtain a ring from a semi-ring.
86
4 ALAIN CONNES
We also show in §6.3 that the algebra W (R, ρ) over R naturally yields a Banach
algebra W (R, ρ) obtained by completion and still depending functorially on (R, ρ).
The Gelfand spectrum of the complexified algebra W (R, ρ)C is a non-empty com-
pact space canonically associated to (R, ρ).
In the last section 7, we return to a more algebraic set-up and to the analogy
with the Witt construction in characteristic p. The need for the construction of
an extension Run of R playing a role similar to the maximal unramified extension
of the p-adic fields appears both in number theory and in quantum physics. In
number theory we refer to the introduction of [22] for the need of an interpretation
of the connected component of identity in the idèle class group of the global field of
rational numbers as a Galois group involving a suitable refinement of the maximal
abelian extension of Q (cf. also the last section of [2]). This global question admits
a local analogue whose solution requires constructing Run . We show in §7 that the
analogue of the Witt construction in characteristic one gives a first hint of what Run
could look like. One obtains the (completion of the) maximal unramified extension
of p-adic integers by applying the Witt construction Wp (at the prime p) to the
extension of the finite field Fp given by an algebraic closure F̄p . The analogue, in
characteristic one, of the latter extension is the extension
(17) B ⊂ Rmax
+
of the only finite semi-field B which is not a field, by the semi-field Rmax + which
plays a central role in idempotent analysis (cf. [13], [16]) and tropical geometry
([10], [12]). The semi-field Rmax+ is defined as the locally compact space R+ =
[0, ∞) endowed with the ordinary product and a new idempotent addition x + y =
max{x, y}. It admits a one parameter group of automorphisms θλ ∈ Aut(Rmax + ),
θλ (x) = xλ for all x ∈ Rmax
+ , which is the analogue of the arithmetic Frobenius.
Since the above construction of W (R, ρ) depends upon the choice of the element
ρ ∈ R, ρ > 1, the W (θλ ) give canonical isomorphisms
(18) W (θλ ) : W (R, ρ) → W (R, θλ (ρ))
To eliminate the dependence upon ρ in the case of Rmax
+ , we consider all ρ’s simul-
T
taneously, i.e. we let ρ = e where T > 0 is a parameter. One then has
(19) w(α, T ) = α−T α (1 − α)−T (1−α)
which depends on the parameter T a bit like the wp (α, T ) of (6). When one
computes using the formula (9), i.e.

(20) x +w y = w(α, T ) xα y 1−α
α∈I
the sum of n terms xj independent of T one obtains

T
1/T
(21) x1 +w · · · +w xn = xj
In particular if all xj = 1, this gives the function T → nT . In fact more generally

the functions of the form T → xT form a sub semi-ring isomorphic to R+ with its
ordinary operations (of addition and multiplication) and are the fixed points of the
natural extension of the θλ as automorphisms given, using (18), by
(22) αλ (f )(T ) = θλ (f (T /λ)) = f (T /λ)λ .
5
In general for elements given by functions f (T ) the addition coming from (20) is
given by
T
(23) (f +w g)(T ) = f (T )1/T + g(T )1/T , ∀T.
This shows that for each T there is a uniquely associated character χT with values
in real numbers with their usual operations, which is given by
(24) χT (f ) = f (T )1/T
and one can use the characters χT to represent the elements of the extension Run as
functions of T with the ordinary operations of pointwise sum and product. In this
representation the functions τ (x) which were independent of T now are represented
by
T → χT (τ (x)) = x1/T
Such functions τ (x) are the analogues of the Teichmüller lifts and they generate
the field formed of fractions of the form

(25) χT (f ) = λj e−sj /T / μj e−tj /T
where the coefficients λ, μ are rational numbers and the exponents s, t are real
numbers. Such fractions, with the coefficients λ, μ real, give a first hint towards
Run but one should relax the requirement that the sums involved are finite. We
briefly discuss in §7.6 the role of divergent series ([18]) in the representation of
elements of Run . The key examples to be covered come from quantum physics and
are given by functional integrals which are typically of the form (cf. §7.7)

S(φ) − J, φ S(φ)
(26) Z(J) = exp(−

)D[φ] / exp(−

) D[φ] .
This suggests that, in quantum physics, the parameter T should be related to

and the above expression (26) should define for each value of the source parameter
J an element f ∈ Run such that Z(J) = χT (f ) for = T . Moreover the elements
of Run obtained in this manner have asymptotic expansions of the form

(27) f (T ) = χT (f )T ∼ an T n ,
where one sets T = . This property should be satisfied by all elements of Run and is
compatible with the fact that the “numbers” which are obtained after quantization
can often only be described as sums of formal perturbative series in powers of .
Basing the definition of Run on the asymptotic expansion (27) is however losing
too much information as shown in the simple case (25) where the expansion only
depends upon the terms with lowest values of sj and tj . We briefly relate this issue
with the theory of Borel summation of divergent series in §7.6 and then show, at the
end of §7, that the above construction is intimately related to idempotent analysis
(cf. [13], [16]). In fact our analogue in characteristic one of the Witt construction
provides the inverse process of the “dequantization” of idempotent analysis, a fact
which justifies the word “quantization” appearing in the title of this paper.
In conclusion we conjecture that the extension Run of R is the natural home
for the “values” of the many -dependent physical quantities arising in quantum
field theory. This fits with the previous understanding of renormalization from the
Riemann-Hilbert correspondence (cf. [4], [5], [8], [9]).
88
6 ALAIN CONNES
3. Sum of Teichmüller representatives

Our goal in this section is to recall a formula which goes back to Teichmüller
[20], for the sum of Teichmüller representatives in the context of strict p-rings.
We begin by recalling briefly the simplest instance of the polynomials with integral
coefficients which express the addition in the Witt theory (We refer to [19] and [17]
for a quick introduction to that theory). One defines the polynomials with integral
coefficients Sn (x, y) ∈ Z[x, y] by the equality
∞

(28) (1 − tx)(1 − ty) = (1 − Sn (x, y)tn )
n=1
The first few are of the form

S1 = x+y
S2 = −xy
S3 = −xy(x + y)
S4 = 3 −xy(x + y)2
S5 = −xy x + 2x y + 2xy 2 + y 3
2
S6 = −xy(x + y)2 x2 + xy + y 2
2
S7 = −xy(x
+ y) x2 + xy + y 2
S8 = −xy(x + y)2 x4 + 2x3 y + 4x2 y 2 + 2xy 3 + y 4
3
S9 = −xy(x + y) x2 + xy + y 2
S10 = −xy(x + y)2 x6 + 3x5 y + 7x4 y 2 + 8x3 y 3 + 7x2 y 4 + 3xy 5 + y 6
One has for all n
n
(29) xn + y n = d Sd (x, y) d
d|n
as follows by equating the coefficients of tn in −t dt

d
log of both sides of (28).
3.1. Strict p-rings. We now fix a prime p and recall well known notions.
Definition 3.1. A ring R is called a strict p-ring provided that R is complete
and Hausdorff with respect to the p-adic topology, p is not a zero-divisor in R, and
the residue ring S = R/pR is perfect.
Let S be a perfect ring of characteristic p.
1. There is a strict p-ring R with residue ring S, which is unique up to canonical
isomorphism.
2. There exists a unique multiplicative section (called the Teichmüller section),
(30) τ :S→R ◦ τ = id : R → S = R/pR
3. Every element x of R can be written uniquely in the form

(31) x= τ (xn )pn xn ∈ K
4. The construction of R and τ is functorial in S: if f : S → S is a homomorphism

of perfect rings of characteristic p, then there is a unique homomorphism F : R → R
given by

(32) F( τ (xn )pn ) = τ (f (xn ))pn
7
3.2. Universal coefficients wp (α). We consider a strict p-ring R with a per-

fect residue ring S = R/pR. We let τ : S → R be the multiplicative section. Since
S is of characteristic p and is perfect, we can define
(33) θα (x) = xα ∀x ∈ S, α = a/pn a ∈ N, n ∈ N
Note that this is an automorphism of S when α is a power of p but not in general.
Lemma 3.2. With Sk ∈ Z[x, y] defined above, one has for any pair (R, S) and
any x, y ∈ S
∞
n n
(34) τ (x) + τ (y) = τ (Spn (x1/p , y 1/p ))pn
0
Proof. This follows from [17] Theorem 1.5 and Lemmas 3.2 and 3.4 which
show that the polynomials Spn (x, y) used above agree with the Sn (x, y, 0, . . . , 0) of
[17] Theorem 1.5.
We now expand the polynomial Spn (x, y) in the form

n

p
n
−k
(35) Spn (x, y) = a(n, k)xk y p , a(n, k) ∈ Z
k=0
and we define a map wp from the set Ip of (5),

(36) Ip = {α ∈ Q ∩ [0, 1], ∃n, pn α ∈ Z}
to the local ring Fp [[T ]] of formal power series in T , by

(37) wp : Ip → Fp [[T ]], wp (α) = ā(n, k)T n ā(n, k) ≡ a(n, k) mod p
k/pn =α
Lemma 3.3. a) One has wp (α) ∈ Fp [[T ]] for all α ∈ Ip .

b) One has wp (0) = wp (1) = 1 and wp (1 − α) = wp (α) ∀α ∈ Ip .
c) For any fixed n, there are only finitely many α’s for which wp (α) = 0 modulo
T n+1 Fp [[T ]].
Proof. a) For each n there exists at most one k such that pn α = k ∈ N. Such
k exists iff n ≥ n0 where pn0 is the denominator of the reduced form of α. Thus
the series (37) is convergent in Fp [[T ]].
b) One has Spn (x, y) = Spn (y, x) so that a(n, pn − k) = a(n, k) and wp (1 − α) =
wp (α). Moreover (28) for y0 = 0 gives the vanishing of Sn (x, 0) for all n ≥ 1. Thus
one gets that a(n, 0) = a(n, pn ) = 0 for all n ≥ 1 and wp (0) = wp (1) = 1.
c) If wp (α) = 0 mod T n+1 Fp [[T ]], then pn α ∈ Z and thus pn α ∈ Z ∩ [0, pn ] which
is a finite set.
We now consider the ring S[[T ]] of formal series in T with coefficients in S.

Since S is of characteristic p, it contains Fp and one has
(38) Fp [[T ]] ⊂ S[[T ]]
We use τ to obtain a bijection of S[[T ]] with R,

(39) τ̃ ( sn T n ) = τ (sn )pn ∈ R
where the sum on the right is p-adically convergent.
90
8 ALAIN CONNES
Theorem 3.4. For any x, y ∈ S one has

(40) τ (x) + τ (y) = τ̃ ( wp (α)xα y 1−α )
α∈Ip
Proof. First, modulo T n the sum in the rhs is finite, thus

(41) s= wp (α)xα y 1−α
α∈Ip
is convergent and defines an element of S[[T ]]. By (37) one has

n n
s= a(n, k)T n xα y 1−α = Tn a(n, k)xk/p y 1−k/p
α∈Ip n n k
n n
which by (35) gives s = Spn (x1/p , y 1/p )T n . Thus
n n
(42) τ̃ (s) = τ (Spn (x1/p , y 1/p ))pn
and (34) gives the required equality (40).

Note that (40) extends to arbitrary sums τ (xi ) for xi ∈ S. One extends the
definition of the Sn (x, y) to k variables by

k ∞

(43) (1 − txj ) = (1 − Sn (x1 , . . . , xk )tn )
1 n=1
and one expands this polynomial for prime powers as

m
(44) Spn (x1 , . . . , xk ) = a(n, m1 , . . . , mk ) xj j

mj =pn
We let
ā(n, m1 , . . . , mk ) ≡ a(n, m1 , . . . , mk ) mod p
Then one considers the simplex

k
(45) Ip(k) = {(αj ) | αj ∈ Ip , αj = 1}
1
(k)
and one defines a map from Ip to Fp [[T ]] by

(46) wp : Ip(k) → Fp [[T ]], wp (α1 , . . . , αk ) = ā(n, m1 , . . . , mk )T n
mj /pn =αj
Theorem 3.5. For any xj ∈ S one has

k
1 · · · xk )
αk
(47) τ (xi ) = τ̃ ( wp (α1 , . . . , αk ) xα1
i=1 (k)
Ip
The proof is the same as for Theorem 3.4. Note that the coefficients wp (α1 , . . . , αk )
depend on T and thus one is forced to extend the perfect ring S in order to deform
its operations. Evaluation at T = 0, gives the usual rules of addition in S.
9
4. Characteristic one
We refer to [11] for the general theory of semi-rings. In a semi-ring R the
additive structure (R, +) is no longer that of an abelian group but is a commutative
monoid with neutral element 0. The multiplicative structure (R, ·) is a monoid with
identity 1 = 0 and distributivity holds while r · 0 = 0 · r = 0 for all r ∈ R. A semi-
ring R is called a semi-field when every non-zero element in R has a multiplicative
inverse, or equivalently when the set of non-zero elements in R is a (commutative)
group for the multiplicative law.
4.1. Perfect semi-rings of characteristic one. We let B = {0, 1} be the

only finite semi-field which is not a field. One has 1 + 1 = 1 in B ([11], [15]).
Definition 4.1. A semi-ring R is said to have characteristic 1 when
(48) 1+1=1
in R i.e. when R contains B as the prime sub-semi-ring.
It follows from distributivity that one then has
(49) a+a=a ∀a ∈ R
This justifies the term “additively idempotent” frequently used in semi-ring theory
as a synonym for “characteristic one”. Note first that (49) implies that
(50) x + y = 0 ⇒ x = 0, y = 0
Indeed one has x = x + (x + y) = (x + x) + y = x + y = 0.
A semi-ring R of characteristic one inherits a canonical partial order (cf. [11]).
Lemma 4.2. For x and y ∈ R one has
(51) y + x = y ⇔ ∃z ∈ R, y = x + z
Moreover this relation defines a partial order x ≤ y on R and

• xj ≤ yj , ∀j, implies xj ≤ yj .
• x ≤ y implies ax ≤ ay for all a ∈ R.
• For any x, y ∈ R the sum x + y is the least upper bound of x and y.
Proof. If y = x + z, then x + y = (x + x) + z = x + z = y. Thus the
equivalence (51) follows. The relation x ≤ y iff x + y = y is reflexive, transitive (if
x ≤ y and y ≤ z then x + z = x + (y + z) = (x + y) + z = y + z = z so that x ≤ z)
and anti-symmetric (x ≤ y, y ≤ x ⇒ x = x + y = y). The compatibility with
addition and multiplication is straightforward and finally if x ≤ z and y ≤ z one
gets x + y ≤ z + z = z so the last property follows.
By definition, a semi-ring R is multiplicatively cancellative when
(52) x = 0, xy = xz ⇒ y = z
Note that this condition means that the natural morphism
(53) R → S −1 R, S = R \ {0}
is an injection. Thus R embeds in a semi-field of characteristic one. This set-up
is thus closely related to that of a partially ordered group G. We recall, from [11]
Propositions 4.43 and 4.44 the following result which describes the analogue of the
Frobenius endomorphism in characteristic p.
92
10 ALAIN CONNES
Lemma 4.3. Let R be a multiplicatively cancellative semi-ring of characteristic

one. Then the map x → xn is an injective endomorphism of R for any n ∈ N.
Proof. Note first that for any m one has

m
(54) (x + y)m = xk y m−k
k=0
as follows from (49). To show that x + y n = (x + y)n we first show

n
(55) (xn + y n )(x + y)n−1 = (x + y)2n−1

which follows from (54) by noticing that if xa y b is a monomial with a + b = 2n − 1
then a ≥ n or b ≥ n. From (55) and the cancellative property one gets
(56) (x + y)n = xn + y n ∀x, y ∈ R
Let us show that xn = y n ⇒ x = y. We start with n = 2.
x2 = y 2 ⇒ x2 = x2 + y 2 = (x + y)2 = x2 + xy + y 2 = x2 + xy. Thus
xx = x(x + y) and canceling x one gets x = x + y. Similarly y = (x + y) and thus
y = x. In general one has, assuming xn = y n
xn = (x + y)n = xn + yxn−1 + · · · + y n−1 x + y n = xn + xn−1 y + · · · xy n−1 =
= x(xn−1 + xn−2 y + · · · + y n−1 ) = x(x + y)n−1
Thus since R is multiplicatively cancellative one gets xn−1 = (x + y)n−1 and by
induction this gives x = x + y and similarly y = x + y.
Definition 4.4. Let R be a multiplicatively cancellative semi-ring of charac-

teristic one. Then R is perfect when for any n the map x → xn is surjective.
Proposition 4.5. Let R be a multiplicatively cancellative perfect semi-ring of
characteristic one. The map x → xn defines an automorphism θn ∈ Aut(R) and
the equality
(57) θα = θa θb−1 , α = a/b ∈ Q∗+
defines an action of Q∗+ on R fulfilling
(58) θλλ = θλ ◦ θλ , θλ (x)θλ (x) = θλ+λ (x)
Proof. The first statement follows from Lemma 4.3. For α = a/b, θα (x) is
the unique solution of
z b = xa , z ∈ R
One has θa1 θa2 = θa1 a2 and the first part of (58) follows. For the second, note that
if
b
zj j = xaj , j = 1, 2
one gets (z1 z2 )b1 b2 = xa1 b2 +a2 b1 .
In the above context we shall use the notation

(59) θα (x) = xα , ∀α ∈ Q∗+ , x ∈ R
Lemma 4.6. Let R be a multiplicatively cancellative perfect semi-ring of char-
acteristic one. For α ∈ Q ∩ (0, 1), x, y ∈ R one has xα y 1−α ≤ x + y.
11
Proof. Let α = a
n, 1−α= b
n with a + b = n. Since θn is an automorphism,
(60) x ≤ y ⇔ xn ≤ y n .
Thus we just need to show that xa y b ≤ (x + y)n which follows from (54)
Example 4.7. We let X be a compact space and R = C̃(X, (0, ∞)) be the
semi-ring obtained by adjoining 0 to the set C(X, (0, ∞)) of continuous functions
from X to (0, ∞) endowed with the operations
(61) (f + g)(x) = Sup(f (x), g(x)), (f g)(x) = f (x)g(x) , ∀x ∈ X
For each x ∈ X the evaluation at x gives a homomorphism from R to Rmax+ , where
Rmax
+ = R+ = [0, ∞) endowed with the ordinary product and a new idempotent
addition x + y = max{x, y}. Note that it is not true in this example that an
increasing bounded sequence fn will have a least upper bound since in general the
latter will be given by a semi-continuous function. What is true however is that
the intervals of the form
(62) [ρ1 , ρ2 ] = {x ∈ R | ρ1 ≤ x ≤ ρ2 } , ρ1 = 0
are complete (i.e. every Cauchy sequence is convergent) for a suitable metric. We
shall see in §6 how to obtain a natural analogue of the p-adic metric in the context
of characteristic one.
4.2. Symmetrization. In this section we recall a well known operation, called
symmetrization, which associates a ring AΔ to a semi-ring (cf. [11]). Note that this
operation always gives a trivial result when A is of characteristic one and will only
be used below in the deformed semi-rings. An ideal in a semi-ring is an additive
subset stable under multiplication by any element of the semi-ring. Given a semi-
ring A, its symmetrization AΔ is obtained as follows:
Proposition 4.8. Let A be a semi-ring.
(1) The product A × A with the following operations is a semi-ring
(63) (a, b) + (c, d) = (a + c, b + d), (a, b)(c, d) = (ac + bd, ad + bc)
(2) The following subset J is an ideal of A × A
(64) J = {(a, a)|a ∈ A}
(3) The quotient (A × A)/J is a ring, denoted AΔ .
Proof. (1) The rules (63) are those of the group semi-ring A[Z/2Z] generated
by A and U commuting with A and fulfilling U 2 = 1.
(2) One has J + J ⊂ J. Let us check that JA ⊂ J. One has
(a, a)(c, d) = (ac + ad, ad + ac) ∈ J
(3) The quotient of a semi-ring by an ideal J is defined by the equivalence relation
(65) α ∼ β ⇔ ∃j, j ∈ J, α + j = β + j
It is true in general that the quotient by an ideal is still a semi-ring [11]. Let us
show that the quotient (A × A)/J is a ring. To see this it is enough to show that
any element (a, b) has an additive inverse. But
(a, b) + (b, a) = (a + b, a + b) ∈ J
so that (b, a) is the additive inverse of (a, b).
94
12 ALAIN CONNES
In the above case the equivalence relation (65) means

(66) (a, b) ∼ (a , b ) ⇔ ∃u, v ∈ A, a + u = a + v, b + u = b + v
This equivalence relation is the same as the one defining the Grothendieck group
of the additive monoid which is given by
(67) (a, b) ∼ (a , b ) ⇔ ∃c, a + b + c = a + b + c
One checks that (67) implies (66), taking u = b + c, v = b + c. Conversely (66)
implies (67), taking c = u + v.
5. Entropy and the w(α)

Let R be a multiplicatively cancellative perfect semi-ring of characteristic one,
and θλ (x) = xλ be the automorphisms θλ ∈ Aut(R) given by Proposition 4.5. We
let I = Q ∩ (0, 1) and consider a map
(68) w : I → R× , w(α) ∈ R× , ∀α ∈ I
We extend w to I¯ = Q ∩ [0, 1] by setting w(0) = w(1) = 1. We consider a sum of
the form

(69) x +w y = w(α)xα y 1−α
α∈I¯
where, by convention we let xα y 1−α = x for α = 1 and xα y 1−α = y for α = 0. We

first look formally at the associativity of the operation (69), disregarding the fact
that it is an infinite sum. The commutativity means that
(70) w(1 − α) = w(α) , ∀α ∈ I.
One has:
(71)

(x +w y) +w z = (x +w y) + z + w(α)(x +w y)α z 1−α
α∈I
⎛ ⎞α

=x+y+z+ w(β)xβ y 1−β + w(α) ⎝x + y + w(β)xβ y 1−β ⎠ z 1−α
β∈I α∈I β∈I

α αβ α(1−β) 1−α
= x + y + z + T+ w(α)w(β) x y z
α,β∈I
where T is the symmetric term given by

T = w(β)xβ y 1−β + w(α)xα z 1−α + w(α)y α z 1−α
β∈I α∈I α∈I
By a similar computation one has

x +w (y +w z) = x + y + z + T + w(u)w(v)1−u xu y v(1−u) z (1−v)(1−u)
u,v∈I
Thus it is enough to compare the two expressions

w(α)w(β)α xαβ y α(1−β) z 1−α ; w(u)w(v)1−u xu y v(1−u) z (1−v)(1−u) .
α,β∈I u,v∈I
13
Equating the exponents of x, y, z one gets the transformation u = αβ, v = α(1−β)

1−αβ .
This transformation gives a bijection of I × I to I × I whose inverse is given by
u
α = u + v − uv , β = .
u + v − uv
Thus the associativity holds provided

1−αβ
α(1 − β)
(72) w(α)w(β)α = w(αβ)w , ∀α, β ∈ I.
1 − αβ
5.1. General solutions. Both (70) and (72) only involve the multiplicative
group G = R× of the semi-ring R and we thus fix a multiplicative group G and
assume that it is uniquely divisible i.e. that xα ∈ G makes sense for any x ∈ G
and positive rational number α. We now look for solutions of (70) and (72) in this
framework. For each n we use the notation (12)

(73) Σn = {(α1 , . . . , αn ) ∈ I n | αj = 1}
Lemma 5.1. Let G be a uniquely divisible (multiplicative) group and w(α) ∈ G,

α ∈ I satisfy (70) and (72), then
(1) The following map from Σn to G is symmetric
(74)
w(α) = w(α1 )w (α2 /(1 − α1 ))1−α1 w (α3 /(1 − α1 − α2 ))1−α1 −α2 · · ·
1−α1 −···−αn−2
· · · w (αn−1 /(1 − α1 − · · · − αn−2 )) , ∀α = (α1 , . . . , αn ) ∈ Σn .
(2) Let (Jk ), k = 1, . . . m, be a partition of {1, . . . , n} = ∪Jk then one has

m
(75) w(α) = w(β) w(γk )βk , ∀α ∈ Σn
1

where βk = Jk αj and (γk )j = αj /βk for all j ∈ Jk .
1
(3) There exists a unique homomorphism χ : Q∗+ → G, such that

(76) χ(n−1 ) = w(n−1 , · · · , n−1 ) , ∀n ∈ N.
(4) One has
(77) w(β) = χ(β)β χ(1 − β)1−β , ∀β ∈ I
Proof. (1) Permuting α1 and α2 only affects the first two terms. Let us show
that
(78) w(α1 )w(α2 /(1 − α1 ))1−α1 = w(α2 )w(α1 /(1 − α2 ))1−α2
Using (72) for α = 1 − α1 , β = α2 /(1 − α1 ) one gets
α(1 − β) 1−α α1
αβ = α2 , 1 − = =
1 − αβ 1 − αβ 1 − α2
Thus (78) follows from (70) and (72). In the same way let us show that, for k < n−1,
the permutation αk ↔ αk+1 does not change (74). It only affects the following
two consecutive terms
T = w(αk /(1−(α1 +. . .+αk−1 )))1−α1 −···−αk−1 w(αk+1 /(1−(α1 +. . .+αk )))1−α1 −···−αk
1 when Jk has only one element w(γk ) = 1

96
14 ALAIN CONNES
Moreover one has

T = (w(a)w(b/(1 − a))1−a )β
where
β = 1−α1 −· · ·−αk−1 , a = αk /(1−(α1 +. . .+αk−1 )) , b = αk+1 /(1−(α1 +. . .+αk−1 ))
since
b/(1 − a) = αk+1 /(1 − (α1 + . . . + αk )) .
Thus the permutation αk ↔ αk+1 interchanges a ↔ b and the invariance follows
from (78). Finally the invariance under the permutation αn−1 ↔ αn follows from
(70).
(2) We first consider the special case of the partition {1, . . . , n} = {1, 2} ∪k {k}.
Then (75) takes the form
(79) w(α1 , α2 , . . . , αn ) = w(α1 + α2 , α3 , . . . , αn )w(α1 /(α1 + α2 ))α1 +α2
In that case using (74), equation (75) reduces to
w(u)w(v/(1 − u))1−u = w(u + v)w(u/(u + v))u+v
for u = α1 , v = α2 (so that u+v ≤ 1). In turns this follows from (72), for α = u+v,
β = u/(u + v).
To prove (75) in general, one proceeds by induction on n. If all the Jk have
only one element the equality is trivial. Let then Jk have at least two elements,
which using the symmetry proved in (1) can be assumed to be 1, 2 ∈ {1, . . . , n}.
One shows that the replacement (α1 , α2 ) → α1 + α2 in both sides of (75) has the
same effect. Using (79) one checks that both sides are divided by the factor
w(α1 /(α1 + α2 ))α1 +α2
Indeed for the left hand side this follows from (79) and in the right hand side only
the term w(γk )βk gets modified to w(γk )βk and one has, using (79),
w(γk ) = w(γk )w(α1 /(α1 + α2 ))−(α1 +α2 )/βk .
This proves (75) by induction on n since it holds for n = 2 as one checks directly
for the two possible partitions.
(3) Let α ∈ Σn and α ∈ Σm . Then the αij = αi αj belong to I and add up to
1. Then all the γk are the same and (75) gives
(80) w((αi αj )) = w((αi ))w((αj ))
One applies this for αi = 1/n, i = 1, . . . , n and αj = 1/m, j = 1, . . . , m and gets
using (76),
(81) χ((nm)−1 ) = χ(n−1 )χ(m−1 ) , ∀n, m ∈ N
Thus there exists a unique extension of χ to a homomorphism from Q∗+ , which is
determined by its value on 1/p, p a prime number.
(4) Let β = m n
. Take α ∈ Σm with αj = m 1
for all j, and the partition in two
sets J1 and J2 containing respectively n and m − n elements. Then apply (75).
1
The left hand side is χ( m ) by construction. The first term in the right hand side is
n
w( m ) = w(β). The term w(γ1 ) gives χ( n1 ) and similarly w(γ2 ) gives χ( m−n
1
). Thus
1 1 β 1 1−β n β m−n 1−β
(75) gives χ( m ) = w(β)χ( n ) χ( m−n ) and thus w(β) = χ( m ) χ( m ) =
χ(β)β χ(1 − β)1−β .
15
Theorem 5.2. Let χ be a homomorphism Q×

+ → G, then the function
(82) w(α) = χ(α)α χ(1 − α)1−α

fulfills (70) and (72). Moreover all solutions of (70) and (72) are of this form.
Proof. Let us first show that (82) implies (72). One has
α(1 − β) 1−αβ
w(αβ)w( ) = χ(αβ)αβ χ(1 − αβ)1−αβ
1 − αβ

α(1−β)
1−α
α(1 − β) 1−α
× χ χ
1 − αβ 1 − αβ
which using the multiplicativity of χ gives
χ(α)αβ χ(β)αβ χ(1 − αβ)(1−αβ−α(1−β)−(1−α)) χ(α)α(1−β) χ(1 − β)α(1−β)
× χ(1 − α)1−α = χ(α)α χ(1 − α)1−α χ(β)αβ χ(1 − β)α(1−β)
which agrees with
w(α)w(β)α = χ(α)α χ(1 − α)1−α χ(β)αβ χ(1 − β)α(1−β) .
Conversely by Lemma 5.1 (4) all solutions of (70) and (72) are of the form (82).
Remark 5.3. There is an interesting relation between Theorem 5.2 and [14].
In the additive notation for the function H of [14] one has the three conditions
(A) : H(1 − x) = H(x)
x x
(B) : H(x + y) = H(y) + (1 − y)H( ) + yH(− )
1−y y
(C) : xH(1/x) = −H(x)
Let us modify the last term of (B) using (A) and (C). One has
x x x+y x+y y
H(− ) = H(1 + ) = H( )=− H( )
y y y y x+y
x y
yH(− ) = −(x + y)H( )
y x+y
Thus one can rewrite (B) as
y x
(B ) : H(x + y) + (x + y)H( ) = H(y) + (1 − y)H( )
x+y 1−y
which is the same as (72) in additive notation i.e. with H(x) = log(w(x)). In our
case the function w(x) is only defined for x ∈ I = Q ∩ [0, 1] but one can show
by direct calculation that such a function fulfilling (70) and (72) can uniquely be
extended to Q to a function fulfilling (A), (B) and (C) in additive notation. The
dihedral group D of order 6 generated by the transformations s : x → 1 − x and
x → x1 admits [0, 1] as a fundamental domain for the coset space D/{1, s}. One
then deduces as in [14] that the function of two variables
x
(83) φ(x, y) = (x + y)H( )
x+y
is a two cocycle on the additive group of Q with coefficients in G. The symmetry and
homogeneity of this cocycle shows that the corresponding extension is an extension
of Q-vector spaces and thus φ is a coboundary, φ = bψ. Moreover the cochain ψ
is uniquely normalized by the condition ψ(1) = 0 and the homogeneity of φ then
shows that the equality χ(x) = x−1 ψ(x) defines a character χ of Q× . Since G is
98
16 ALAIN CONNES
uniquely divisible χ(−x) = χ(x) for all x and χ is determined by its restriction to
Q∗+ . Thus since H(x) = φ(x, 1 − x) one obtains an alternate proof of Theorem 5.2.
5.2. Positive solutions. We let, as above, R be a multiplicatively cancella-
tive perfect semi-ring of characteristic one. The uniquely divisible group G = R×
is a vector space over Q using the action θα (x) = xα and is partially ordered by
Lemma 4.2. We shall now make the stronger assumption that it is a partially or-
dered vector space over R. Thus G is a partially ordered group endowed with a one
parameter group of automorphisms θλ ∈ Aut(G), λ ∈ R× such that, with θ0 (x) = 1
for all x by convention,
(84) θλλ = θλ ◦ θλ , θλ (x)θλ (x) = θλ+λ (x)
We assume the following compatibility (closedness) of the partial order with the
vector space structure
(85) λn → λ, θλn (x) ≥ y ⇒ θλ (x) ≥ y
Theorem 5.4. Let w : I → G fulfill (70) and (72) and
(86) w(α) ≥ 1, ∀α ∈ I .
Then there exists ρ ∈ G, ρ ≥ 1 such that
(87) w(α) = ρS(α) , S(α) = −α log(α) − (1 − α) log(1 − α) , ∀α ∈ I.
Proof. We use additive notations so that for x ∈ G, one has log(x) ∈ E where
E is a partially ordered vector space over R. We let s(α) = log(w(α)) ≥ 0 and
L(α) = log(χ(α)) for α ∈ Q∗+ . Note that by (74) and (86) one gets
(88) w(α1 , . . . , αn ) ≥ 1
and thus by (76), χ(1/n) ≥ 1, so that
(89) L(1/n) ≥ 0 , ∀n ∈ N
One lets l(p) = L(1/p) ≥ 0, for each prime p. By (77) one has
(90) s(α) = αL(α) + (1 − α)L(1 − α), s(α) ≥ 0 , ∀α ∈ I.
n
p1 1
Let p1 and p2 be two primes and nj integers such that α = n
p2 2
< 1. One has:
n n
p2 2 −p1 1
1−α= n
p2 2
and if one lets
a
(91) pn2 2 − pn1 1 = qj j
be the prime factor decomposition of pn2 2 − pn1 1 > 0, one gets

(92) L(1 − α) = n2 l(p2 ) − aj l(qj ) ≤ n2 l(p2 )
Thus, since s(α) ≥ 0, αL(α) = s(α) − (1 − α)L(1 − α) fulfills
(93) αL(α) ≥ −(1 − α)L(1 − α) ≥ −(1 − α)n2 l(p2 )
But L(α) = n2 l(p2 ) − n1 l(p1 ) and thus dividing by n2 we get
n1
(94) α(l(p2 ) − l(p1 )) ≥ −(1 − α)l(p2 )
n2
which gives
n1
(95) l(p2 ) − α l(p1 ) ≥ 0 .
n2
17
Let
log(p2 )
a=
log(p1 )
then a ∈
/ Q. Taking rational approximations (using the density of Z + aZ in R),
one gets a sequence (n1 (j), n2 (j)) such that
n1 (j) − a n2 (j) < 0 , n1 (j) − a n2 (j) → 0 , when j → ∞
one then gets, when j → ∞ and with nk = nk (j),
n1 log(p2 ) pn1 pn1
(96) →a= , 1n2 < 1 , 1n2 → 1
n2 log(p1 ) p2 p2
and using (85) one gets from (95),
l(p2 ) l(p1 )
(97) − ≥ 0.
log(p2 ) log(p1 )
Exchanging the roles of p1 and p2 thus gives the equality
l(p2 ) l(p1 )
(98) = , ∀p1 , p2 .
log(p2 ) log(p1 )
This shows that λ = l(p)/ log(p) is a positive element of E independent of the prime
p. One then has
L(1/n) = λ log(n) , ∀n ∈ N,
and thus L(α) = −λ log(α) for α ∈ Q∗+ . One thus gets
log(w(α)) = λ(−α log(α) − (1 − α) log(1 − α)) = λS(α)
which gives (87), using the multiplicative notation.
Note that the function w automatically extends by continuity from I to [0, 1].
6. Analogue of the Witt construction in characteristic one

Let R be a multiplicatively cancellative perfect semi-ring of characteristic one.
We look at the meaning of an expression of the form

(99) w(α)xα y 1−α
α∈I¯
Here, I is Q ∩ (0, 1), I¯ = Q ∩ [0, 1] = I ∪ {0, 1} and we use the notation xα = θα (x)
(cf. (59)). Note that this notation does not make sense for α ∈ {0, 1}. Thus we let
(100) xα y 1−α = x for α = 1, xα y 1−α = y for α = 0
The index set I¯ is countable and we treat (99) as a discrete sum.
nZ ∩ [0, 1]. Then the partial sums

1
Lemma 6.1. Let I(n) =

(101) s(n) = w(α)xα y 1−α
I(n)
form an increasing family for the partial order given by divisibility n|m.
Proof. We need to show that s(n) ≤ s(m) when n divides m. In that case
one has I(n) ⊂ I(m) and thus the conclusion follows from (51).
100
18 ALAIN CONNES
Moreover, assuming w(α) ≤ ρ for all α ∈ I and some fixed ρ ∈ R, we get using
Lemmas 4.2 and 4.6 that
(102) x + y ≤ s(n) ≤ (x + y)ρ ∀n.
The hypothesis that any increasing bounded sequence has a least upper bound is
too strong since it fails in example 4.7 for instance. We shall now show that the
s(n) form a Cauchy sequence and converge to the least upper bound of the s(n)
provided one passes to the completion for the ρ-adic metric which we now construct.
6.1. Completion for the ρ-adic metric. Let R be a multiplicatively can-
cellative perfect semi-ring of characteristic one. Let ρ ∈ R× , ρ ≥ 1. We want to use
the w(α) given by (87) and for this we first need to perform a suitable completion.
The function S(α) is positive and bounded by log 2 < 1 for α ∈ I. By (102) we just
need to complete the intervals of the form
(103) [ρ−n , ρn ] = {x ∈ R | ρ−n ≤ x ≤ ρn }, n∈N
We let

(104) Rρ = {0} ∪ [ρ−n , ρn ] ⊂ R.
n∈N
Lemma 6.2. (1) For any α, β ∈ Q∗+

(105) ρα + ρβ = ρα∨β , α ∨ β = Sup(α, β).
(2) Rρ is a perfect sub semi-ring of R.
Proof. (1) Let us show that ρα ≤ ρβ for α < β. It is enough to show that
1 + ρβ−α = ρβ−α , i.e. it is enough to show that
(106) 1 + ρ γ = ργ , ∀γ ∈ Q+
For γ = nk this follows from the additivity of θγ since 1 + ρ = ρ.
(2) It follows from (1) that Rρ is a sub semi-ring of R. Moreover for α ∈ Q∗+
and x ∈ [ρ−n , ρn ] one has θα (x) ∈ [ρ−nα , ρnα ] ⊂ Rρ using (1).
We want to construct a metric d on Rρ such that
(107) d(ρα , 1) → 0, when α ∈ Q, α → 0.
For α ∈ Q∗+ we let
(108) Uα = {(x, y) | x ≤ yρα , y ≤ xρα }
Lemma 6.3. One has
(109) Uα−1 = Uα , Uα ◦ Uβ ⊂ Uα+β
and

(110) (xj , yj ) ∈ Uα =⇒ ( xj , yj ) ∈ Uα .
Proof. For (x, y) ∈ Uα and (y, z) ∈ Uβ one gets

x ≤ yρα , y ≤ zρβ =⇒ x ≤ zρα+β
and one gets in the same way z ≤ xρα+β . This gives (109). Similarly

xj ≤ yj ρα =⇒ xj ≤ ( yj )ρα .
which gives (110).
19
We now define d(x, y) ∈ R+ for x, y ∈ Rρ \ {0} as follows

(111) d(x, y) = inf{α | (x, y) ∈ Uα }.
Note that d(x, y) < ∞ for all x, y ∈ Rρ \ {0} since if they both belong to some
interval [ρ−n , ρn ] one has d(x, y) ≤ 2n.
Lemma 6.4. One has
(112) d(x, y) = d(y, x) , d(x, z) ≤ d(x, y) + d(y, z) , ∀x, y, z = 0.

(113) d( xi , yi ) ≤ Sup(d(xi , yi )) , ∀xi , yi = 0.
(114) d(xy, zt) ≤ d(x, z) + d(y, t) , ∀x, y, z, t = 0.
(115) d(xα , y α ) ≤ α d(x, y) , ∀x, y = 0.

Proof. (112) follows from (109). Similarly (113) follows from (110). Note
that one has
x ≤ zρα , y ≤ tρβ =⇒ xy ≤ ztρα+β
which implies (114). Finally
x ≤ yρβ =⇒ xα ≤ y α ραβ
which implies (115).
Proposition 6.5. The separated completion R̄ρ obtained by adjoining 0 to the
separated completion of (Rρ \ {0}, d) is a perfect semi-ring of characteristic one.
The action θα , α ∈ Q∗+ on R̄ρ extends by continuity to an action of R∗+ by
automorphisms θλ ∈ Aut(R̄ρ ).
Proof. Lemma 6.4 shows that d defines a pseudo-metric on Rρ \ {0}. In fact
it also defines a pseudo-metric on the set of Cauchy sequences of Rρ \ {0}. The
quotient by the equivalence relation d(x, y) = 0 is naturally endowed with addition
using (113) which implies
d(x + y, z + t) ≤ Sup(d(x, z), d(y, t))
and multiplication using (114) so that
d(xy, zt) ≤ d(x, z) + d(y, t)
is still valid for Cauchy sequences. This shows that R̄ρ is a semi-ring. The inequal-
ity (115) shows that the θα extend to the completion (for α ∈ Q∗+ ) by uniform
continuity. The equality 1 + 1 = 1 continues to hold in R̄ρ which is thus a perfect
semi-ring of characteristic one. Let us show that the action θα , α ∈ Q∗+ on R̄ρ
extends by continuity to an action of R∗+ . First for any x ∈ [ρ−n , ρn ] and α ∈ Q∗+
one has d(x, 1) ≤ n and using (115)
(116) d(xα , 1) ≤ n α
Using (114) this implies
(117) d(xα , xβ ) ≤ n|α − β|
and shows that for αj → λ ∈ R∗+ the sequence xαj is a Cauchy sequence.
We continue to use the notation θλ (x) = xλ for λ ∈ R∗+ .

102
20 ALAIN CONNES
6.2. Construction of W (R, ρ). We can now show that the s(n) of Lemma
6.1 form a Cauchy sequence.
Lemma 6.6. Let R, ρ and R̄ρ be as above. Let w(α) = ρS(α) for all α ∈ I.
Then for any x, y ∈ R̄ρ the sequence

(118) s(n) = w(α)xα y 1−α
I(n)
is a Cauchy sequence which converges to the lowest upper bound of the w(α)xα y 1−α .
Proof. We can assume x, y = 0 since otherwise s(n) is constant. We estimate
d(s(n), s(nm)). For this we write the elements of I(nm) in the form
a k
(119) α= + , 0≤k≤m
n nm
Using (113) it is enough to estimate uniformly
a a a
(120) d(ρS(α) xα y 1−α , ρS( n ) x n y 1− n )
Thus it is enough to show that for any > 0, there exists δ > 0 such that
(121) d(ρS(α) xα y 1−α , ρS(β) xβ y 1−β ) ≤ , ∀α, β ∈ [0, 1], |α − β| < δ
This follows from (114) which allows one to consider each of the three terms sep-
arately. One uses (117) for the terms xα , y 1−α and the uniform continuity of the
function S(α), α ∈ [0, 1] for the term ρS(α) . By Lemma 6.1 the limit of the s(n)
gives the lowest upper bound of the s(n) and hence of the ρS(α) xα y 1−α .
Thus, under the hypothesis of Lemma 6.6, we get the convergence in R̄ρ . We
keep a notation closely related to the Witt case and let

(122) w(α)xα y 1−α = lim s(n) , s(n) = w(α)xα y 1−α
n→∞
α∈I¯ α∈I(n)
Before we go any further we shall evaluate the new operation on the powers of ρ.
Let us evaluate (122) for x = ρa , y = ρb . One has

s(n) = ρ(S(α)+αa+(1−α)b)
α∈I(n)
and thus, by (105), it is given by ρσn (a,b) where

(123) σn (a, b) = Supα∈I(n) (S(α) + αa + (1 − α)b)
Lemma 6.7. For a, b ∈ R, one has
(124) σn (a, b) → log(ea + eb ) when n → ∞
Proof. The function S(α) + αa + (1 − α)b = f (α) is uniformly continuous for
α ∈ [0, 1] and thus σn (a, b) tends to its maximum. One has
1−α
(125) f (α) = log( )+a−b
α
ea
and f (α) = − α(1−α)
1
< 0, so that f is strictly concave on [0, 1]. Take α = ea +eb
.
eb
Then 1 − α = ea +eb
, 1−α
α = e(b−a) and thus, by (125), f (α) = 0. Thus f is strictly
21
increasing in [0, α] and decreasing in [α, 1]. Its value at α is

− α log(α) − (1 − α) log(1 − α) + αa + (1 − α)b =
= −α log(ea ) − (1 − α) log(eb ) + αa + (1 − α)b + α log(ea + eb )+
+ (1 − α) log(ea + eb ) = log(ea + eb )

Proposition 6.8. (a) The following defines an associative operation in R̄ρ

(126) x +w y = w(α)xα y 1−α , 0 +w y = y +w 0 = y
α∈I¯
(b) One has (x +w y)z = xz +w yz, ∀x, y, z ∈ R̄ρ .

(c) One has
(127) ρa +w ρb = ρc , c = log(ea + eb ) , ∀a, b ∈ R.
Proof. By Lemma 6.6 (126) is well defined. The commutativity follows from
the symmetry w(1 − α) = w(α). For x = 0 all terms xα y 1−α vanish except when
α = 0 which gives y, thus the sum gives x +w y = y. To show associativity we
proceed as in (71) and we can now justify the equality
⎛ ⎞α

w(α) ⎝ w(β)xβ y 1−β ⎠ z 1−α = w(α)w(β)α xαβ y α(1−β) z 1−α
α∈I β∈I α,β∈I
using the continuity of the automorphisms θα (cf. (115)). The equality (b) follows
from z = z α z 1−α . (c) follows from Lemma 6.7.
Corollary 6.9. Assume ρ = 1. Then (R̄ρ , +w , ·) is a semi-ring and the map
(128) s ∈ R+ → r(s) = ρlog(s) , r(0) = 0,

is an injective homomorphism of the semi-ring R+ (with ordinary addition) in
(R̄ρ , +w , ·).
Proof. Proposition 6.8 shows that (R̄ρ , +w , ·) is a semi-ring. Using (127)
it just remains to show that ρa = ρb if a = b. One has ρα = 1 for α ∈ Q∗+ ,
α = 0 since ρ = 1 and θα is an automorphism. By Lemma 6.2, this shows that
d(ρα , ρβ ) = |α − β| for α, β ∈ Q and extends to
d(ρα , ρβ ) = |α − β| , ∀α, β ∈ R
which gives the injectivity of r.
We now define W (R, ρ).

Definition 6.10. Let R be a multiplicatively cancellative perfect semi-ring of
characteristic one and ρ ∈ R, ρ ≥ 1. We let
(129) W (R, ρ) = (R̄ρ , +w , ·)Δ
be the symmetrization of (R̄ρ , +w , ·).
104
22 ALAIN CONNES
It is a ring by construction (cf. §4.2). We need to ensure that 1 = 0 in W (R, ρ).

By (67) it is enough to show that an equation of the form 1 +w x = x where
x ∈ R̄ρ gives a contradiction. We have x ≤ ρn for some n. For x ≤ ρn one has
x1−α ≥ xρ−nα for α ∈ [0, 1], since ρnα ≥ xα ⇒ ρnα x1−α ≥ x. One has by
construction
1 +w x ≥ ρS(α) x1−α , ∀α ∈ [0, 1]
and hence using x1−α ≥ xρ−nα one gets
1 +w x ≥ ρS(α)−nα x , ∀α ∈ [0, 1].

e−n
Take α = 1+e−n one has S(α) − nα = log(e−n + 1) > 0 and one gets
(130) x ≤ ρn ⇒ 1 +w x ≥ ρβ x, β = log(e−n + 1)
Thus the equality 1 +w x = x implies that x ≥ ρβ x for some β > 0. Iterating one
gets x ≥ ρmβ x for all m ∈ N. But x ∈ [ρ−n , ρn ] for some n and one gets
ρn ≥ x ≥ ρmβ x ≥ ρmβ ρ−n , ∀m ∈ N
which gives a contradiction provided ρ = 1.

We define r(s) ∈ W (R, ρ) for s ∈ R by
(131) r(0) = 0, r(s) = (ρlog(s) , 0), s ≥ 0, r(s) = (0, ρlog |s| ), s < 0
Theorem 6.11. Assume ρ = 1. Then W (R, ρ) is an algebra over R.
Proof. It is enough to show that r is an injective homomorphism of the field

R of real numbers in the ring W (R, ρ). One has: r(s1 s2 ) = r(s1 )r(s2 ) using
log |s1 s2 | = log |s1 | + log |s2 | and the rule of signs. For sj > 0, one has: r(s1 ) +
r(s2 ) = r(s1 + s2 ) by Corollary 6.9. This extends to arbitrary signs by adding r(t)
to both terms. Moreover we have shown above that r(1) = 1 = 0 so that we get a
homomorphism of R in W (R, ρ). It is necessarily injective since R is a field.
Example 6.12. We take Example 4.7, R = C̃(X, (0, ∞)) for X a compact
space. We write
(132) ρ(x) = eT (x) T (x) ≥ 0
One then gets, using (124), the following formula for the addition +w of two func-
tions
(133) (f +w g)(x) = (f (x)β(x) + g(x)β(x) )T (x) , β(x) = 1/T (x)
Proposition 6.13. Let Ω = {x ∈ X, ρ(x) > 1} (i.e. the complement of

{x, T (x) = 0}). Then the map f → f β extends to an isomorphism
(134) W (R, ρ) Cb (Ω, R)
Proof. For ρ−n ≤ f ≤ ρn one has e−n ≤ f β ≤ en . Conversely, any h ∈

C(Ω, (e−n , en )) extends by f = hT (x) to X, with f (x) = 1 for x ∈/ Ω. This
is enough to show that the symmetrization process gives Cb (Ω, R) starting with
functions h(x), x ∈ Ω such that h(x) ∈ [e−n , en ] for some n.
23
6.3. The Banach algebra W (R, ρ). The above Example 6.12 suggests to
define a semi-norm on W (R, ρ) starting from
(135) ||f || = Inf{λ ∈ R+ | f ≤ ρlog λ } , ∀f ∈ R̄ρ
which is finite for elements of R̄ρ . One has
(136) ||f g|| ≤ ||f ||||g||
because a ≤ c, b ≤ d ⇒ ab ≤ cd in R̄ρ . Note also that, since f1α ≤ f2α if f1 ≤ f2
one gets
(137) f1 ≤ f2 ⇒ f1 +w g ≤ f2 +w g
Since ρlog λ1 +w ρlog λ2 = ρlog(λ1 +λ2 ) one thus gets
(138) ||f +w g|| ≤ ||f || + ||g||
We now extend ||.|| to the symmetrization W (R, ρ). Starting with a semi-ring A
and a semi-norm f → ||f || such that (136) and (138) hold, one can endow the
semi-ring A[Z/2Z] (cf. §4.2) with the semi-norm
(139) ||(a, b)||1 = ||a|| + ||b||
One has the compatibility with the product
(140) ||(a, b)(c, d)||1 ≤ ||(a, b)||1 ||(c, d)||1
since the left hand side is ||ac +w bd|| + ||ad +w bc|| ≤ ||a||||c|| + ||b||||d|| + ||a||||d|| +
||b||||c|| = ||(a, b)||1 ||(c, d)||1 . We now take the quotient by the ideal J = {(a, a), a ∈
A} as in §4.2. Thus the quotient semi-norm is
(141) ||(a, b)||1 = Inf{||(x, y)||1 , ∃u ∈ A, a +w y +w u = x +w b +w u}
It still satisfies (138) and (140).
We apply the above discussion to A = W (R, ρ). To show that, after completion
for the semi-norm (141), we get a Banach algebra W (R, ρ) over R we still need to
show that the quotient norm does not vanish.
Lemma 6.14. Assume ρ = 1. Let (a, b) be equivalent to (1, 0) modulo J, then
||a|| ≥ 1.
Proof. Since (a, b) is equivalent to (1, 0) there exists c ∈ R̄ρ such that
(142) a +w c = 1 +w b +w c
thus by (137) one has
(143) 1 +w c ≤ a +w c
Let us assume that ||a|| < 1. Then a ≤ ρ−s for some s > 0, and one can then find
t > 0 such that
e−s + e−t = 1
which implies, by (127), that ρ−s +w ρ−t = 1. We have, by (137),
1 +w c ≤ a +w c ≤ ρ−s +w c
and thus
(144) ρ−t +w ρ−s +w c ≤ ρ−s +w c
With x = ρ−s +w c this gives ρ−t +w x ≤ x and 1+w ρt x ≤ ρt x. Since 1+w ρt x ≥ ρt x
by (137), one gets 1+w ρt x = ρt x which shows that 1 = 0 in W (R, ρ) and contradicts
Theorem 6.11.
106
24 ALAIN CONNES
It follows that the Banach algebra obtained as the completion of W (R, ρ) is

non trivial since the norm of (1, 0) is equal to 1. Note that with the notation (131)
one has
(145) ||r(s)|| = |s| ∀s ∈ R
Indeed one has ||r(s)|| ≤ |s| by construction using (131) and (135) and by Lemma
6.14 one has ||r(1)|| = 1 so that by (140) one has
||r(s)r(1/s)|| ≤ ||r(s)||||r(1/s)||
and ||r(s)|| ≥ |s|.
Theorem 6.15. Assume ρ = 1. The completion W (R, ρ) is a unital Banach
algebra over R.
For any character χ of the complexification W (R, ρ)C = W (R, ρ) ⊗R C one has
(146) χ(ρ) = e
Proof. By construction W (R, ρ) is a real unital Banach algebra. The norm
on the complexification is defined by
(147) ||a + ib|| = ||a|| + ||b||
For a character χ of W (R, ρ)C , the restriction to C1 is the identity and thus one
has χ(r(s)) = s which gives (146) taking s = e = 2, 71828...
It follows from Gelfand’s theory that the characters of the complex Banach
algebra W (R, ρ)C form a non-empty compact space
(148) X = Spec(W (R, ρ)C ) = ∅
We shall not pursue further the study of this natural notion of spectrum associated
to the original pair (R, ρ).
Remark 6.16. Example 6.12 shows that the norm ||.||1 in W (R, ρ) is not
the spectral radius norm. In that example it gives ||f ||1 = ||f+ || + ||f− || where
f = f+ + f− is the decomposition of f in positive and negative parts. The ordinary
C(X, R) norm is given by Sup(||f+ ||, ||f− ||). From (135) one gets in general
(149) ||f n || = ||f ||n ∀f ∈ R̄ρ
However this holds only before passing to the symmetrization.
7. Towards Run
Our goal in this section is to show how to apply the above analogue of the
Witt construction to the semi-field Rmax
+ of idempotent analysis [13] [16] and show
that it gives in that case the inverse operation of the “dequantization”. It will
allow us to take a first speculative step towards the construction of the sought for
“unramified” extension Run of R.
In the case of the Witt construction, the functoriality allows one to apply the
functor Wp to an algebraic closure F̄p of Fp which yields the following diagram
Wp
F̄p → Wp (F̄p )
⊂
(150)
Wp
Fp → Zp = Wp (Fp )
25
In our case, the analogue of the extension Fp ⊂ F̄p is the extension of semi-rings
(151) B ⊂ Rmax
+
and the one-parameter group of automorphisms θλ ∈ Aut(Rmax λ

+ ), θλ (x) = x ,
max
plays the role of the Frobenius. But since our construction of W (R+ , ρ) depends
upon the choice of ρ, one first needs to eliminate the choice of ρ by considering
simultaneously all possible choices.
7.1. The w(α, T ). To eliminate the dependence on ρ it is natural to allow all

values of ρ, i.e. to introduce a parameter T ≥ 0,
(152) ρ = eT ∈ Rmax
+ , ρ≥1
With this notation, w(α) depends on T as it does in the Witt case, one has explicitly
(153) w(α, T ) = α−T α (1 − α)−T (1−α)
We view the w(α, T ) as the analogues of the wp (α, T ) of (6) of the Witt case. The
presence in w(α, T ) of the parameter T ≥ 0 means that even if one adds terms
which are independent of T the result will depend on T . Thus one works with
functions f (T ) ∈ Rmax
+ with the usual pointwise product and the new addition

(154) (f1 +w f2 )(T ) = w(α, T )f1 (T )α f2 (T )1−α
α∈I¯
Lemma 7.1. The addition (154) is given by
(155) (f1 +w f2 )(T ) = (f1 (T )1/T + f2 (T )1/T )T
for T > 0 and by
(156) (f1 +w f2 )(0) = sup(f1 (0), f2 (0))
Proof. This follows from Lemma 6.7 for T > 0 and from the equality w(α, 0) =
1 for T = 0. One then uses Lemma 4.6 to conclude in the case T = 0.
Corollary 7.2. The sum of n terms xj independent of T is given by

T
1/T
(157) x1 +w · · · +w xn = xj
In particular one can compute the sum of n terms all equal to 1 which will
necessarily be fixed under any automorphism of the obtained structure. One gets
(158) 1 +w 1 +w · · · +w 1 = nT
We expect more generally that the functions of the form
f (T ) = xT , ∀T ≥ 0
will be fixed by the lift of the θλ ∈ Aut(Rmax

+ ). We now review the analogy with
the Witt case.
108
26 ALAIN CONNES
7.2. Teichmüller lift. The constant functions T → x are the analogue of the
Teichmüller representatives
(159) τ (x)(T ) = x ∀T
One has

(160) τ (x) + τ (y) = w(α, T ) xα y 1−α
α∈I¯
where the sum in the right hand side is computed in Rmax

+ . We view this formula
as the analogue of the formula (40) of the usual Witt case.
7.3. Residue morphism. The evaluation at T = 0 is by construction a mor-

phism
(161) : f → f (0) ∈ Rmax
+
We view this morphism as the analogue of the canonical map which exists for any
strict p-ring
(162) p : Wp (K) → K = Wp (K)/pWp (K)
7.4. Lift of the automorphisms θλ ∈ Aut(Rmax + ). One has a natural one

parameter group of automorphisms αλ of our structure, which corresponds to the
θλ ∈ Aut(Rmax
+ ). It is given by
Proposition 7.3. The following defines a one parameter group of automor-

phisms
(163) αλ (f )(T ) = f (T /λ)λ ∀λ ∈ R×
+
One has
(164) ◦ αλ = θλ ◦ , ∀λ
and
(165) αλ ◦ τ = τ ◦ θλ , ∀λ.
Proof. One has for T > 0 using Lemma 7.1,
(T /λ)×λ
λ
αλ (f1 +w f2 )(T ) = ((f1 +w f2 )(T /λ)) = f1 (T /λ)λ/T + f2 (T /λ)λ/T
= (αλ (f1 )(T )1/T + αλ (f2 )(T )1/T )T = (αλ (f1 ) +w αλ (f2 ))(T )
thus αλ is additive. It is also multiplicative and defines an automorphism. The
equality (164) follows from (163) by evaluation at T = 0. The equality (165) also
follows from (163).
7.5. Fixed points. We now determine the fixed points of the αλ and show,
as expected from (158) that they form the semi-field R+ .
Proposition 7.4. The fixed points of αλ are of the form
(166) f (T ) = aT
and they form the semi-field R+ which is the positive part of the field R of real
numbers, endowed with the ordinary addition.
27
Proof. Assume that αλ (f ) = f for all λ > 0. Then one has, using (163) for
λ = T , f (T ) = f (1)T which gives (166). Moreover by Lemma 7.1 the addition
corresponds to the semi-field R+ which is the positive part of the field R of real
numbers.
7.6. Characters and representation by functions. For each T > 0 the
algebraic operations on the value f (T ) are the same as in the semi-field R+ using
the evaluation f (T )1/T . Thus there is a uniquely associated character χT which is
such that
(167) χT (f ) = f (T )1/T
and we now use the characters χT to represent the elements of the extension Run
as functions of T with the ordinary operations of pointwise sum and product.
Proposition 7.5. The following map χ is a homomorphism of semi-rings to
the algebra of functions from (0, ∞) to R+ with pointwise sum and product,
(168) χ(f )(T ) = f (T )1/T , ∀T > 0.
One has
(169) χ(τ (x))(T ) = x1/T , ∀T > 0
and
(170) χ(αλ (f ))(T ) = χ(f )(T /λ) , ∀T > 0.
Proof. These properties are straightforward consequences of (167).
In this representation χ, the residue morphism of §7.3 is given, under suitable
continuity assumptions by
(171) (f ) = lim χ(f )(T )T .
T →0
In this representation the algebraic operations are very simple and this suggests to
represent elements of Run as functions χ(f )(T ). Among them one should have the
fixed points (166) which give χ(f ) = a and the Teichmüller lifts which give (169).
We parameterize the latter in the form
(172) eξ (T ) = e−ξ/T , ∀T > 0.
After symmetrization and passing to the field of quotients, the fixed points (166)
and the Teichmüller lifts (159) generate the field of fractions of the form (in the χ
representation)

(173) χ(f )(T ) = aj e−ξj /T / bj e−ηj /T
where the coefficients aj , bj are real numbers and the exponents ξj , ηj ∈ R. While
such expressions give a first hint towards Run one should not be satisfied yet since,
as explained in §7.7 below, natural examples coming from quantum physics use
expressions of the same type but involving more elaborate sums. In all these ex-
amples, including those coming from the functional integral, it turns out that not
only limT →0 χ(f )(T )T exists as in (171) but in fact the function f (T ) = χ(f )(T )T
admits an asymptotic expansion for T → 0 of the form

(174) f (T ) = χ(f )(T )T ∼ an T n .
110
28 ALAIN CONNES
For functions of the form (173), this expansion only uses the terms with the lowest
values of ξj and ηj and is thus only a crude information on the element f . Moreover
as soon as one uses integrals instead of finite sums in (173) one obtains ∞general
asymptotic expansions (174). In fact, given a convergent series g(T ) = 0 bn T n
one has (cf. e.g. [18])
∞
(175) g(T ) = b0 + e−ξ/T φ(ξ)dξ
0
where φ(ξ) is the Borel transform

∞
ξn
φ(ξ) = bn+1
0
n!
Thus g(T ) is of the form (173) using integrals. Moreover, with b0 = 1, the asymp-
totic expansion of g(T )T for T → 0 is of the form

(176) g(T )T ∼ an T n ,
where the coefficients an are given by a0 = 1 and
a1 = 0
a2 = b1
b2
a3 = − 21 + b2
b21 b31
a4 =
3 4
2 + 3 − b1 b2 +2b3
b b b
a5 = − 21 − 41 + b1 b2 + b21 b2 − 22 − b1 b3 + b4
b31 11b41 b51 b2
a6 = 6 + 24 + 5 − 32 b21 b2 − b31 b2 + 22 + b1 b22 + b1 b3 + b21 b3 − b2 b3 − b1 b4 + b5
which shows how to determine the bn once the coefficients aj , j ≤ n + 1 are given.
Using the freedom to multiply g(T ) by ae−ξ0 /T for a, ξ0 ∈ R, a > 0, one gets in
this way any convergent series an T n with a0 > 0 as the asymptotic expansion of
T
g(T ) for g(T ) of the form
ξ1
g(T ) = ae−ξ0 /T + e−ξ/T φ(ξ)dξ
ξ0
One checks that the only restriction on the an is that a0 > 0 so that a0 = e−ξ0
for some ξ0 ∈ R. This suggests more generally to use the theory of divergent series
(cf. [18]) in the construction of Run . The simple point is that the action of R∗+
given by the αλ of (163), gives a grading which admits the fn , χ(fn )(T ) = T n as
eigenvectors. This is only formal since the fn (T ) = T nT do not have an asymptotic
expansion (176) but the χ(fn ) are integrals of the above form since
∞
n! T n+1 = e−ξ/T ξ n dξ .
0
7.7. Deformation parameter T ∼ . In idempotent analysis ([13], [16]),

the process that allows one to view Rmax+ as a result of a deformation of the usual
algebraic structure on real numbers is known as “dequantization” and can be de-
scribed as a semi-classical limit in the following way. First of all note that, for real
numbers wj , one has
h ln(ew1 /h + ew2 /h ) → max{w1 , w2 } as h → 0.
29
Thus, the natural map

Dh : R+ → R+ Dh (u) = uh
satisfies the equation
lim Dh (Dh−1 (u1 ) + Dh−1 (u2 )) = max{u1 , u2 }.
h→0
In the limit h → 0, the usual algebraic rules on R+ deform to become those of
Rmax
+ . In our context this corresponds to the residue morphism of §7.3 expressed
in the χ-representation. More specifically, with χ(f ) given by (173) one has
(177) (f ) = lim χ(f )(T )T = e− inf j (ξj )+inf j (ηj )
T →0
when the coefficients aj , bj are positive real numbers so that the exponentiation
makes sense. These elements form a semi-field and is a homomorphism to the
semi-field Rmax
+ . In fact, it extends to a homomorphism ˜ from the field K of
rational fractions of the form (173) to the hyperfield T R of tropical reals defined
by O. Viro (cf. [21]) by the following hyperaddition of real numbers:
⎧
⎨ a, if |a| > |b| or a = b;
(178) ab= b, if |a| < |b| or a = b;
⎩
[-a,a], if b = −a.
The extension is given by ˜(0) = 0 and for a reduced fraction

(179) χ(f )(T ) = aj e−ξj /T /
bj e−ηj /T
aj0 −ξj +ηk

(180) ˜(f ) = sign( )e 0 0 , ξ
j0 = inf (ξj ), ηk0 = inf (ηk ).
bk0 j k
One has χ(f )(T ) ∼ a ˜(f )1/T , for T → 0, for some a > 0 (with the notation
x = sign(x)|x|λ for x ∈ R and λ > 0). Thus one obtains the hyperfield T R of
λ
tropical reals as the quotient K/G of the field K by a subgroup of its multiplicative
group which, as already observed by M. Krasner, is the natural construction of
many hyperfields ([2], [3]).
The formalism of idempotent analysis, motivated by quantum physics, suggests
that the parameter T should be related to the Planck constant . Moreover in order
to use Run in the context of quantum physics, one should relax the requirement
that the sums involved in (173) only involve finitely many terms. The key example
is given by the functional integral in the Euclidean formulation of Quantum Field
Theory. Indeed the generating function of Euclidean Green’s functions is given by
(cf. e.g. [9])

S(φE ) − JE , φE
(181) Z(JE ) = N exp − D[φE ]

where S(φE ) is the Euclidean action, in terms of the Euclidean classical fields φE ,
the source JE is an element of the linear space dual to that of Euclidean classical
fields and the normalization factor N is the inverse of

S(φE )
exp − D[φE ]

Such integrals are the prototype of sums for +w where w is the function given by
(153) but since the sums are infinite one needs to extend the entropy from finite
partitions of 1 to infinite partitions. The formula for computing the sum is then
112
30 ALAIN CONNES
deeply related to the basic formula of thermodynamics using the entropy to express
the free energy from a variational principle involving the sum of the entropy and
the energy with suitable multipliers. It suggests that it might be worthwhile to
reconsider the functional integral from this angle, considering this formula as more
basic than the analogy with ordinary integrals.
Note finally that the expansion (174) still holds for the elements such as Z(JE )
of (181) and that this asymptotic expansion is in general much more involved than
in the simplest example of (173) since it is the “loop expansion” of quantum field
theory which is the basis of the concrete computations in quantum physics (cf.
e.g. [9]). In conclusion the above development suggests that the extension Run of
R is the proper receptacle for the “values” of many -dependent physical quanti-
ties arising in quantum field theory. Together with the previous understanding of
renormalization from the Riemann-Hilbert correspondence (cf. [4], [5], [8], [9]) this
should be an important piece of the puzzle provided by the quantum.
References
[1] A. Connes, C. Consani Characteristic 1, entropy and the absolute point; arXiv:0911.3537v1.
[2] A. Connes, C. Consani, The hyperring of adèle classes, to appear in Journal of Number
Theory; arXiv:1001.4260.
[3] A. Connes, C. Consani From monoids to hyperstructures: in search of an absolute arithmetic
Casimir Force, Casimir Operators and the Riemann Hypothesis, 147–198, de Gruyter 2010.
[4] A. Connes, D. Kreimer, Renormalization in quantum field theory and the Riemann-Hilbert
problem. I. The Hopf algebra structure of graphs and the main theorem. Comm. Math. Phys.
210 (2000), no. 1, 249–273.
[5] A. Connes, D. Kreimer, Renormalization in quantum field theory and the Riemann-Hilbert
problem. II. The β-function, diffeomorphisms and the renormalization group. Comm. Math.
Phys. 216 (2001), no. 1, 215–241.
[6] A. Connes, C. Consani, M. Marcolli, Noncommutative geometry and motives: the thermody-
namics of endomotives, Advances in Math. 214 (2) (2007), 761–831.
[7] A. Connes, C. Consani, M. Marcolli, The Weil proof and the geometry of the adeles class
space, to appear in “Algebra, Arithmetic and Geometry – Manin Festschrift”, Progress in
Mathematics, Birkhäuser (2008); arXiv0703392.
[8] A. Connes, M. Marcolli, Renormalization and motivic Galois theory, International Math.
Research Notices, (2004), no. 76, 4073–4091.
[9] A. Connes, M. Marcolli, Noncommutative Geometry, Quantum Fields, and Motives, Collo-
quium Publications, Vol.55, American Mathematical Society, 2008.
[10] A. Gathmann, Tropical algebraic geometry. Jahresber. Deutsch. Math.-Verein. 108 (2006),
no. 1, 3–32.
[11] J. Golan, semi-rings and their applications Updated and expanded version of The theory of
semi-rings, with applications to mathematics and theoretical computer science [Longman Sci.
Tech., Harlow, 1992. Kluwer Academic Publishers, Dordrecht, 1999.
[12] I. Itenberg, G. Mikhalkin, E. Shustin, Tropical algebraic geometry. Second edition. Oberwol-
fach Seminars, 35. Birkhauser Verlag, Basel, 2009. x+104 pp.
[13] V. Kolokoltsov, V. Maslov, Idempotent analysis and its applications. Translation of Idem-
potent analysis and its application in optimal control (Russian), “Nauka” Moscow, 1994.
Translated by V. E. Nazaikinskii. With an appendix by Pierre Del Moral. Mathematics and
its Applications, 401. Kluwer Academic Publishers Group, Dordrecht, 1997.
[14] M. Kontsevich The 1 12 -logarithm Friedrich Hirzebruchs Emeritierung, Bonn, November 1995.
[15] P. Lescot Algèbre absolue arXiv:0911.1989
[16] G. Litvinov Tropical Mathematics, Idempotent Analysis, Classical Mechanics and Geometry
arXiv:1005.1247
[17] J. Rabinoff, The theory of Witt vectors. Notes available at http://math.harvard.edu/ rabi-
noff/misc/witt.pdf
31
[18] J. P. Ramis, Séries divergentes et théories asymptotiques. (French) [Divergent series and
asymptotic theories] Bull. Soc. Math. France 121 (1993), Panoramas et Syntheses, suppl., 74
pp.
[19] J. P. Serre, Local fields, Graduate Texts in Mathematics, vol. 67, Springer-Verlag, New York,
1979, Translated from the French by Marvin Jay Greenberg.
[20] O. Teichmüller, Über die Struktur diskret bewerteter perfekter Körper, Nachr. Ges. Wiss.
Göttingen N.F. 1 (1936), 151-161.
[21] O. Viro Hyperfields for tropical geometry I, hyperfields and dequantization arXiv:1006.3034v2.
[22] A. Weil, Sur la théorie du corps de classes, J. math. Soc. Japan, t. 3, 1951, p. 1-35.
A. Connes: Collège de France, 3, rue d’Ulm, Paris, F-75005 France, I.H.E.S. and
Vanderbilt University
E-mail address: alain@connes.org
Volume 546, 2011
LIE PREALGEBRAS
Michel DUBOIS-VIOLETTE and Giovanni LANDI
Dedicated to Henri Moscovici
Abstract. We introduce a generalization of Lie algebras within the theory of

nonhomogeneous quadratic algebras and point out its relevance in the theory
of quantum groups. In particular the relation between the differential calculus
on quantum groups and the Koszul duality due to Positselski is made apparent.
Contents
1. Introduction
2. Lie algebras, PBW property, etc.
3. Associative algebras
4. Quadratic algebras
5. Examples I
6. Nonhomogeneous quadratic algebras
7. Duality
8. Examples II
9. Lie prealgebras
10. Complexes and homologies
11. Further prospects (tentative conclusion)
References
2010 Mathematics Subject Classification. Primary: 16S37; Secondary: 16E40, 17B37.

Key words and phrases. Quadratic algebras. Koszul algebras. Complexes and homologies.
Quantum groups. Differential calculi.
GL was partially supported by the ‘Italian project Cofin08 - Noncommutative geometry,
quantum groups and applications’.
2011
1
115
116
2 MICHEL DUBOIS-VIOLETTE AND GIOVANNI LANDI
1. Introduction
Our aim here is to describe a generalization of Lie algebras based on the theory
of nonhomogeneous quadratic algebras [15], [14], [4], [8]. Beside the usual Koszul
duality of quadratic algebras [12], [13], [14], a very powerful extension of it has been
developed by L. Positselski [15] to a duality between nonhomogeneous quadratic
algebras satisfying an extension of the Poincaré-Birkhoff-Witt (PBW) property and
curved graded differential algebras. Restricted to the quadratic-linear algebras this
duality leads to graded differential algebras and this is a natural framework to un-
derstand the differential calculi on quantum groups such as the ones introduced by
S.L. Woronowicz [20], [21]. In this paper we recall this theory of nonhomogeneous
quadratic algebras and we specify a subclass closer to the universal enveloping al-
gebras of Lie algebras. This subclass of quadratic-linear algebras is characterized
by a variant of the Poincaré duality property for the homogeneous part referred to
as the Gorenstein property [1] and by the Koszul property.
In order to render the paper intelligible for readers not acquainted with the
theory of nonhomogeneous quadratic algebras and to make it self-contained, an
important part of this article consists of a summary of the appropriate piece of the
theory of nonhomogeneous quadratic algebras. In Section 2, we recall some basic
facts on Lie algebras, associated complexes and the PBW property. Section 3 pro-
vides a similar discussion for associative algebras instead of Lie algebras. Quadratic
algebras, Koszul complexes, etc. are reviewed in Section 4. Section 5 introduces
some representative examples of quadratic algebras which will be used later as the
homogeneous part of examples of nonhomogeneous quadratic algebras. In Section 6
the nonhomogeneous quadratic algebras are introduced and the PBW property is
defined for these algebras. Section 7 is devoted to the Koszul duality of Positselski.
In Section 8 nonhomogeneous versions of the examples of Section 5 are discussed.
Lie prealgebras are defined in Section 9 and are shown to be in duality with the
differential quadratic Koszul Frobenius algebras. In Section 10 the corresponding
generalization of Chevalley-Eilenberg complexes is investigated.
Throughout the paper, K is a (commutative) field and all vector spaces and
algebras are over K. By an algebra without other mention, we always mean an
associative algebra and by a graded algebra we mean a N-graded algebra.
2. Lie algebras, PBW property, etc.
Let E be a finite-dimensional vector space equipped with an antisymmetric

bilinear product
(2.1) (x, y) → [x, y]
that is, one has a linear mapping
ψ : ∧2 E → E
and the product is given by
[x, y] + ψ(x ∧ y) = 0
for any x, y ∈ E.
LIE PREALGEBRAS 117
3
Let the tensor algebra T (E) over E be equipped with the filtration

E⊗
m
F n (T (E)) =
m≤n
associated to its gradation and let A be the algebra defined by

A = T (E)/({x ⊗ y − y ⊗ x − [x, y]|x, y ∈ E})
where (P ) denotes for any P ⊂ T (E) the two-sided ideal of T (E) generated by P .
Notice that one has
({x ⊗ y − y ⊗ x − [x, y]|x, y ∈ E}) = ({r + ψ(r)|r ∈ ∧2 E})
where in the right-hand side of this equality ∧2 E is considered as a subset of E ⊗ E.
The filtration of the tensor algebra induces a filtration F n (A) of A and one
defines the associated graded algebra

gr(A) = F n (A)/F n−1 (A)
n
(with the convention F n (A) = 0 whenever n < 0). One has a canonical surjective
homomorphism
(2.2) can : S(E) → gr(A)
of graded algebras of the symmetric algebra S(E) onto gr(A).
Consider the transpose ψ t of ψ, that is, the linear mapping
ψ t : E ∗ → ∧2 E ∗
where E ∗ denotes the dual vector space of E. This transpose ψ t has a unique
extension as an antiderivation
(2.3) d : ∧E ∗ → ∧E ∗
of the exterior algebra ∧E ∗ over E ∗ . One has the following theorem
Theorem 1. The following conditions (a), (b) and (c) are equivalent:
(a) the bracket (2.1) satisfies the Jacobi identity, i.e. one has
[[x, y], z] + [[y, z], x] + [[z, x], y] = 0
for any x, y, z ∈ E,
(b) the antiderivation (2.3) is a differential of ∧E ∗ , i.e. d2 = 0,
(c) the canonical homomorphism (2.2) is an isomorphism of S(E) onto gr(A).
This theorem is more or less classical and easy to prove (see for instance [7]).
Condition (a) means that (E, [•, •]) is a Lie algebra, Condition (b) means that
(∧E ∗ , d) is a graded differential algebra while Condition (c) means that one has the
Poincaré-Birkhoff-Witt (PBW) property.
In fact A is then the universal enveloping algebra U (E) of the Lie algebra E
while the graded differential algebra ∧E ∗ is the basic complex to construct the
Chevalley-Eilenberg complexes (d is the Koszul differential).
118
3. Associative algebras
Assume now that the finite-dimensional vector space E is equipped with an

arbitrary bilinear product
(3.1) (x, y) → xy
so one has a linear mapping
(3.2) ψ :E⊗E →E
and the product is given by
xy + ψ(x ⊗ y) = 0
for any x, y ∈ E. Thus E is an algebra which may be nonassociative.
Let then A be the (associative) algebra
A = T (E)/({x ⊗ y − xy|x, y ∈ E})
with the same conventions as before. One has
({x ⊗ y − xy|x, y ∈ E}) = ({r + ψ(r)|r ∈ E ⊗ E})
for the defining ideal of T (E).
The algebra A is filtered by the filtration induced from that of the tensor algebra
T (E) and one has the associated graded algebra
gr(A) = ⊗n F n (A)/F n−1 (A)
as in the last section. Let us investigate the structures of A and gr(A). The
projection of T (E) onto the degree 0 part induces homomorphisms
ε : T (E) → K
and
ε:A→K
of unital (associative) algebras. The kernels T+ (E) and A+ of these homomorphisms
are two-sided ideals and one has the decompositions
T (E) = T+ (E) ⊕ K1l
and
A = A+ ⊕ K1l
of these algebras. By construction and by the very universal property of T+ (E),
one has a surjective homomorphism of algebra
α : E → A+
which is such that any algebra homomorphism of E into an associative algebra A
factorizes through a unique homomorphism of A+ into A and α. It follows that A+
is the quotient of E by the two-sided ideal I generated by the associators
[x, y, z] = (xy)z − x(yz)
for x, y, z ∈ E. One has
F n (A) = A for n ≥ 1 and F 0 (A) = K1l
so the graded algebra gr(A) is given by
gr(A) = gr1 (A) ⊕ K1l
LIE PREALGEBRAS 119
5
where gr1 (A) = A+ = E/I as vector space but where the product of two elements
of gr1 (A) vanishes. Let A be the graded algebra
A = T (E)/(E ⊗ E) = E ⊕ K1l
with obvious notations. It is clear that one has a canonical surjective homomor-
phism
(3.3) can : A → gr(A)
of graded algebras which is an isomorphism if and only if E is an associative algebra.
Consider the transpose linear mapping
ψt : E ∗ → E ∗ ⊗ E ∗
of ψ. This transpose ψ t has a unique extension as an antiderivation
(3.4) d : T (E ∗ ) → T (E ∗ )
of the tensor algebra over E ∗ . One has the following analog of Theorem 1.
Theorem 2. The following conditions (a), (b) and (c) are equivalent:
(a) the product (3.1) is associative,
(b) the antiderivation (3.4) is a differential, i.e. d2 = 0,
(c) the canonical homomorphism (3.3) is an isomorphism.
We have explained above the equivalence (a) ⇔ (c). For the equivalence (a) ⇔
(b) as in Theorem 1 see for instance in [7].
In fact A is then the algebra
A = E ⊕ K1l = Ẽ
that is, the algebra obtained by adjoining a unit element to E; the algebra E itself
identifies with the two-sided ideal A+ of A, i.e. one has E = A+ . Notice that
then the graded differential algebra (T (E ∗ ), d), or more precisely (T+ (E ∗ ), d), is
the basic complex to construct the Hochschild complexes.
We shall see that the algebras A of Section 2 satisfying the conditions of Theo-
rem 1 and of this section satisfying the conditions of Theorem 2 are both examples
of Koszul quadratic-linear algebras but that the latter one does not fall into the
class of Lie prealgebras.
4. Quadratic algebras
A (homogeneous) quadratic algebra [12], [13], [14] is an associative algebra A

of the form
A = A(E, R) = T (E)/(R)
where E is a finite-dimensional vector space, R is a subspace of E ⊗ E and where
(R) denotes the two-sided ideal of the tensor algebra T (E) over E generated by
R(⊂ E ⊗ E). The space E is the space of generators of A and the subspace R of
E ⊗ E is the space
of relations of A. The algebra A = A(E, R) is naturally a graded
algebra A = n∈N An which is connected, i.e. such that A0 = K1l and generated
in degree 1, A1 = E.
120
To a quadratic algebra A = A(E, R) as above one associates another quadratic

algebra, its Koszul dual A! , defined by
A! = A(E ∗ , R⊥ )
where E ∗ denotes the dual vector space of E and R⊥ ⊂ E ∗ ⊗ E ∗ is the orthogonal
of R ⊂ E ⊗ E defined by
R⊥ = {ω ∈ E ∗ ⊗ E ∗ |ω, r = 0, ∀r ∈ R}
where, by using the finite-dimensionality of E, we have identified E ∗ ⊗ E ∗ with the
dual vector space (E ⊗ E)∗ of E ⊗ E. One has of course
(A! )! = A
n of the homogeneous components An of A are

and the dual vector spaces A!∗ ! !
A!∗
1 =E
and

E⊗ ⊗ R ⊗ E⊗
r s
(4.1) A!∗
n =
r+s+2=n
⊗ n
for n ≥ 2, as easily verified. In particular A!∗
2 = R and An ⊂ E
!∗
for any n ∈ N.
Consider the sequence of free left A-modules
b b b b
(4.2) → A ⊗ A!∗
n+1 → A ⊗ An → · · · → A ⊗ A2 → A ⊗ E → A → 0
!∗ !∗
n+1 → A ⊗ An is induced by the left A-module homomorphism of

where b : A ⊗ A!∗ !∗
A ⊗ E⊗ into A ⊗ E ⊗ defined by
n+1 n
b(a ⊗ (x0 ⊗ x1 ⊗ · · · ⊗ xn )) = ax0 ⊗ (x1 ⊗ · · · ⊗ xn )

⊗ n−2
for a ∈ A, xi ∈ E. It follows from (4.1) that A!∗
n ⊂ R⊗E for n ≥ 2, which
2
implies that b = 0. As a consequence the sequence (4.2) is a chain complex of free
left A-modules called the Koszul complex of the quadratic algebra A and denoted
by K(A). The quadratic algebra A is said to be a Koszul algebra whenever its
Koszul complex is acyclic in positive degrees, i.e. whenever Hn (K(A)) = 0 for
n ≥ 1. One shows easily that A is a Koszul algebra if and only if its Koszul dual
A! is a Koszul algebra.
It is important to realize that the presentation of A by generators and relations
is equivalent to the exactness of the sequence
b b ε
(4.3) A⊗R →A⊗E →A→K→0
so one always has
H1 (K(A)) = 0 and H0 (K(A)) = K
and
ε
(4.4) K(A) → K → 0
is a free resolution of the trivial module K whenever A is Koszul.This resolution is
then a minimal projective resolution of K in the graded category (i.e. the category
of graded modules) [5]. In the above sequences, ε is induced by the projection onto
degree 0.
LIE PREALGEBRAS 121
7
Let A = A(E, R) be a quadratic Koszul algebra such that one has A!D = 0
and A!n = 0 for n > D. It follows that the trivial (left) module K has projective
dimension D which implies that A has global dimension D (see [5]). It is worth
noticing that this also implies that the Hochschild dimension of A is D (see [3]).
By applying the functor HomA (•, A) to the Koszul chain complex K(A) of left
A-modules one obtains the cochain complex L(A) of right A-modules
b b b b
(4.5) 0 → A → · · · → A!n ⊗ A → A!n+1 ⊗ A →

b being the left multiplication by k θ k ⊗ ek in A! ⊗ A where (ek ) is a basis of E
with dual basis (θ k ). The algebra A is said to be Koszul-Gorenstein if it is Koszul
of finite global dimension D as above and if H n (L(A)) = KδD n
. Notice that this
implies that An AD−n as vector spaces (a version of Poincaré duality).
! !∗
5. Examples I
In this section we analyse some examples of quadratic algebras A = A(E, R)

which will appear as the “homogeneous parts” of representative examples of the
concepts described in the sequel.
1. Case R = ∧2 E ⊂ E ⊗ E.
In the case where R = ∧2 E considered as a subspace of E ⊗E, the quadratic algebra
A = A(E, R) is the symmetric algebra S(E) over E
A = A(E, ∧2 E) = S(E)
and one has
A! = ∧E ∗
for its Koszul dual.
In this case the Koszul complex is (S(E) ⊗ ∧• E, b) and the exterior differential
d is such that
db + bd = degree (in S • and ∧• )
so d gives an homotopy for b in positive degrees. Therefore A = S(E) is Koszul and,
since ∧D E = 0 and ∧n E = 0 for n > D = dim(E), A = S(E) has global dimension
D = dim(E). It is not hard to show that A = S(E) is in fact Koszul-Gorenstein.
The Koszul dual A! = ∧E ∗ is therefore also Koszul but is not of finite global
dimension since its Koszul complex (∧E ∗ ⊗ S • (E), b) has components of degree n
∧E ∗ ⊗ S n (E) = 0
for any n ∈ N.
2. Case R = E ⊗ E.
In the case where R = E ⊗ E, A = A(E, R) is the trivial quadratic algebra
A = A(E, E ⊗ E) = E ⊕ K1l
with A1 = E and An = 0 for n ≥ 2, and one has
A! = A(E ∗ , {0}) = T (E ∗ )
for its Koszul dual, i.e. it is the tensor algebra over E ∗ .
122
The Koszul complex for A! = T (E ∗ ) then reduces to
0 → T (E ∗ ) ⊗ E ∗ → T (E ∗ ) → 0
b
where b is the “product”, so A! = T (E ∗ ) is Koszul of global dimension 1 but is

obviously not Koszul-Gorenstein whenever dim(E) ≥ 2.
The quadratic algebra A = E ⊕ K1l is thus also Koszul but, since its Koszul
complex ((E ⊕ K1l) ⊗ T • (E), b) has components of degree n,
(E ⊕ K1l) ⊗ E ⊗ = 0
n
for any n ∈ N, it is not of finite global dimension.

3. Example connected with the calculus [20] on the twisted SU (2)
group.
Let A be the quadratic algebra generated by 3 elements ∇0 , ∇1 , ∇2 of degree 1 with
relations
⎧
⎪ 1
⎪
⎪ ν∇2 ∇0 − ∇0 ∇2 = 0
⎪
⎪ ν
⎪
⎪
⎪
⎨
1
(5.1) ν 2 ∇1 ∇0 − 2 ∇0 ∇1 = 0
⎪
⎪ ν
⎪
⎪
⎪
⎪
⎪
⎪
⎩ ν 2 ∇2 ∇1 − 1 ∇1 ∇2 = 0
ν2
where ν ∈ K\{0}. By setting
(5.2) E = K∇0 ⊕ K∇1 ⊕ K∇2
and
1
R = K(ν∇2 ⊗ ∇0 − ∇0 ⊗ ∇2 )
ν
1
⊕ K(ν 2 ∇1 ⊗ ∇0 − 2 ∇0 ⊗ ∇1 )
ν
1
(5.3) ⊕ K(ν ∇2 ⊗ ∇1 − 2 ∇1 ⊗ ∇2 )
2
ν
one has A = A(E, R).
The Koszul dual of A is the algebra A! generated by 3 elements ω0 , ω1 , ω2 of
degree 1 with relations
⎧
⎪
⎪ (ωα )2 = 0, ∀α ∈ {0, 1, 2}
⎨
ω2 ω0 + ν 2 ω0 ω2 = 0
(5.4)
⎪
⎪ ω1 ω0 + ν 4 ω0 ω1 = 0
⎩
ω2 ω1 + ν 4 ω1 ω2 = 0
The algebra A is a deformation of S(E) with dim(E) = 3, it is Koszul [9] of
global dimension 3 and is Koszul-Gorenstein (see however Remark c) below). The
argument for the Koszul property is similar to the one used in the example 1 above
(see also [18]). The Koszul algebra A! has infinite global dimension.
LIE PREALGEBRAS 123
9
4. Generalization: q-deformed polynomial algebras.

Let A be the algebra generated by D elements X λ (λ ∈ {1, . . . , D}) with relations
(5.5) X μ X ν − q μν X ν X μ = 0
for μ, ν ∈ {1, . . . , D} where the q μν ∈ K are such that
(5.6) q μν q νμ = 1, q λλ = 1
for λ, μ, ν ∈ {1, . . . , D}. Again this quadratic algebra is Koszul [9] of global dimen-
sion D and is in fact Koszul-Gorenstein. Its Koszul dual A! is generated by the D
elements ωα (α ∈ {1, . . . , D}) with relations
(5.7) ωμ ων + q νμ ων ωμ = 0
for μ, ν ∈ {1, . . . , D}. It is again a Koszul algebra of infinite global dimension.
Remarks
a) Let A be a quadratic Koszul algebra with Koszul dual A! . Then their Poincaré
series satisfy
PA (t)PA! (−t) = 1
where the Poincaré series PA (t) is defined by

PA (t) = dim(An )tn
n
for any graded algebra A = ⊕n An . It follows that if A has a finite global dimension
D ≥ 1 then A! is of infinite global dimension.
b) Recall that a graded algebra A = ⊕n An is said to have polynomial growth
whenever there are positive K and N ∈ N such that
dim(An ) ≤ KnN −1
for any n ∈ N. The algebras A of Examples 1, 2, 3 and 4 above have polynomial
growth but the tensor algebra T (E) has exponential growth whenever dim(E) ≥ 2.
c) To complete the proof of the Koszul and the Koszul-Gorenstein properties for
the algebras of Examples 3 and 4 above, one must use in addition to the results of
[9] the fact that the Koszul and the Gorenstein properties are stable by the twists
[2]. This was proved in [16] in the more general context of homogeneous algebras.
6. Nonhomogeneous quadratic algebras
In this subsection we let E be as before a finite-dimensional vector space

and we
endow the tensor algebra T (E) with its natural filtration F n (T (E)) = m≤n E ⊗ ,
m
(n ∈ N).
A nonhomogeneous quadratic algebra [15], [4], [8] is an algebra A of the form
A = A(E, P ) = T (E)/(P )
where P is a subspace of F 2 (T (E)) and where (P ) denotes as above the two-sided
ideal of T (E) generated by P . The filtration of the tensor algebra T (E) induces a
filtration F n (A) of A and one associates as before to A the graded algebra
gr(A) = ⊕n F n (A)/F n−1 (A)
124
which is referred to as the associated graded algebra to the filtered algebra A.

Let R be the image of P under the canonical projection of F 2 (T (E)) onto E ⊗ E
and let A = A(E, R) be the homogeneous quadratic algebra T (E)/(R); A will
be referred to as the quadratic part of A. There is again the canonical surjective
homomorphism
can : A → gr(A)
of graded algebras and A is said to have the Poincaré-Birkhoff-Witt (PBW) property
whenever this homomorphism is an isomorphism. This terminology comes from the
example where A is the universal enveloping algebra U (g) of a Lie algebra g (see
Section 2).
One has the following theorem [4].
Theorem 3. Let A and A be as above. If A has the PBW property then the
following conditions (i) and (ii) are satisfied:
(i) P ∩ F 1 (T (E)) = 0,
(ii) (P.E + E.P ) ∩ F 2 (T (E)) ⊂ P .
Assume that A is a Koszul algebra, then conversely if conditions (i) and (ii) are
satisfied A has the PBW property.
Condition (i) means that P is obtained from R by adding to each non-zero

element of R terms of degrees 1 and 0. In other words, Condition (i) means that
there are linear mappings ψ1 : R → E and ψ0 : R → K such that one has
(6.1) P = {x + ψ1 (x) + ψ0 (x)1l|x ∈ R}
which gives P in terms of R. Condition (ii), which is a generalization of the Jacobi

identity, is then given more explicitly by the following proposition (see e.g. in [14]).
Remark. In the second part of Theorem 3 one can replace the condition of Koszulity
of A by a weaker condition, namely the condition of acyclicity in degrees 3 and 2
of the Koszul complex of A, to conclude that A has the PBW property whenever
conditions (i) and (ii) are satisfied.
In spite of the above remark, it turns out that the Koszul property for A is a
very natural and helpful property. A nonhomogeneous quadratic algebra A is said
to be Koszul whenever it has the PBW property and its (homogeneous) quadratic
part A is Koszul.
Proposition 4. Assume that Condition (i) of Theorem 3 is satisfied, let ψ1

and ψ0 be as above and set W3 = (R ⊗ E) ∩ (E ⊗ R). Then Condition (ii) of
Theorem 3 is equivalent to the following conditions (a), (b) and (c):
(a) (ψ1 ⊗ I − I ⊗ ψ1 )(W3 ) ⊂ R

(b) ψ1 ◦ (ψ1 ⊗ I − I ⊗ ψ1 ) − (ψ0 ⊗ I − I ⊗ ψ0 ) (W3 ) = 0
(c) ψ0 ◦ (ψ1 ⊗ I − I ⊗ ψ1 )(W3 ) = 0
where I is the identity mapping of E onto itself.
LIE PREALGEBRAS 125
11
7. Duality
In this section we recall the Koszul duality of [15] for the nonhomogeneous
quadratic Koszul algebras.
Let A = T (E)/(P ) be a nonhomogeneous quadratic algebra and let its qua-
dratic part be A = T (E)/(R). We assume that the condition (i) of Theorem 3
is satisfied and we let ψ1 : R → E and ψ0 : R → K be the corresponding linear
mappings as in (6.1).
Let ψ1t : E ∗ → R∗ and ψ0t :
K∗ → R∗ be the transpose linear mappings of ψ1
and ψ2 . The Koszul dual A = n A!n of A is such that
!

(E ⊗ ⊗ R ⊗ E ⊗ )∗
r s
A!n =
r+s+2=n
∗
for n ≥ 2 and A!1 = E , so in particular we can write
(7.1) ψ1t : A!1 → A!2
and
(7.2) ψ0t (1) = F
is an element of A!2 . One has A!3 = W3∗ and Condition (a) of Proposition 4 means
that ψ1t extends as an antiderivation
(7.3) d : A! → A!
of degree 1 of A! , Condition (b) then reads
(7.4) d2 α = [F, α]
for any α ∈ A! while Condition (c) reads
(7.5) dF = 0
for the element F = ψ0t (1) of A!2 .
A graded algebra equipped with an antiderivation d of degree 1 and an element
F of degree 2 satisfying the conditions (7.4) and (7.5) above is called a curved
graded differential algebra [15]. The above correspondence between nonhomoge-
neous quadratic algebras satisfying conditions (i) and (ii) of Theorem 3 and curved
graded differential algebras is an anti-equivalence of categories between the cate-
gory of nonhomogeneous quadratic algebras satisfying conditions (i) and (ii) and a
full subcategory of the category of curved graded differential algebras [14].
Let us introduce a more specific setting. A (curved) graded differential algebra
which is as graded algebra a quadratic algebra endowed with its natural gradation
will be referred to as a (curved) differential quadratic algebra. If furthermore the
underlying quadratic algebra is Koszul then it will be referred to as a curved differ-
ential quadratic Koszul algebra. Given two curved differential quadratic algebras,
A = (A(E, R), d, F ) and A = (A(E , R ), d , F ), a morphism of A to A is a linear
mapping
α : E → E
such that
(α ⊗ α)(R) ⊂ R
126
and such that the corresponding unital algebra homomorphism α̃ : A → A satisfies

α̃(F ) = F
and
α̃(dx) = d α̃(x)
for any x ∈ A. One has then the corresponding category of curved differential
quadratic algebras and the full subcategory of curved differential quadratic Koszul
algebras.
Given two nonhomogeneous quadratic algebras A = A(E, P ) = T (E)/(P ) and
A = A(E , P ) = T (E )/(P ), a morphism from A to A
α : A → A
is a linear mapping
α : E → E
such that
T (α)(P ) ⊂ P

where T (α) : T (E) → T (E ) is the corresponding unital algebra homomorphism.
One has the corresponding category of nonhomogeneous quadratic algebras and the
full subcategory of nonhomogeneous quadratic Koszul algebras.
It is clear that the correspondence A → (A! , d, F ) defines a contravariant func-
tor from the category of nonhomogeneous quadratic algebras satisfying the con-
ditions (i) and (ii) of Theorem 3 to the category of curved differential quadratic
algebras which is “one to one”. The duality is summarized by the following theorem.
Theorem 5. The above correspondence defines an anti-isomorphism between
the category of nonhomogeneous quadratic algebras satisfying Conditions (i) and
(ii) of Theorem 3 and the category of curved differential quadratic algebras which
induces an anti-isomorphism between the category of nonhomogeneous quadratic
Koszul algebras and the category of curved differential quadratic Koszul algebras.
Remark. It is important to realize that an object A of the category of nonhomoge-

neous quadratic algebras as defined above is not only the filtered algebra A but it
is A = A(E, P ) defined as a quotient of a tensor algebra T (E) as explained before.
In particular the vector space E is explicitly involved although its canonical image
in the filtered algebra is a quotient which is generically not isomorphic to E; see for
instance in Section 3 the case where ψ : E ⊗ E → E is a nonassociative product.
The morphisms of this category have been defined accordingly. Of course if A has
the PBW property E is mapped isomorphically onto the corresponding subspace
of the algebra A. More generally it is worth noticing here that if the conditions (i)
and (ii) of Theorem 3 are satisfied, then the canonical projection p : T (E) → A is
injective on E, i.e. it induces an isomorphism E p(E), [4] (see §3.2 in [4]).
8. Examples II
In this section we analyze some examples of nonhomogeneous quadratic Koszul

algebras having examples of Section 5 as homogeneous parts and we describe their
dual (curved) differential quadratic algebras.
LIE PREALGEBRAS 127
13
1. Universal enveloping algebra U (g).

Let g be a Lie algebra and let U (g) be its universal enveloping algebra. It is
clear that A = U (g) is a nonhomogeneous quadratic algebra with quadratic part A
which coincides with the symmetric algebra S(g), A = S(g). It follows that for the
Koszul dual one has A! = ∧g∗ . We know (see in Section 5, Example 1) that S(g)
is Koszul of global dimension D = dim(g) and is in fact Koszul-Gorenstein with
polynomial growth. Furthermore it is a classical result that U (g) has the PBW
property (see also in Section 2). Thus U (g) is a Koszul nonhomogeneous quadratic
algebra. In this case ψ0 = 0, i.e. there are only linear nonquadratic terms, so ∧g∗
is a differential quadratic Koszul algebra (curvature zero).
2. The case where E is an associative algebra A.
Let A be an associative algebra with product (x, y) → xy and let A be the nonho-
mogeneous quadratic algebra defined by
A = T (A)/({x ⊗ y − xy|x, y ∈ A})
as in Section 3. Then the quadratic part is given by
A = T (A)/(A ⊗ A) = A ⊕ K1l
i.e. A1 = A as vector space and An = 0 for n ≥ 2. The quadratic algebra A is
Koszul but is of infinite global dimension (see in Section 5, Example 2). In view of
Theorem 2, A has the PBW property and therefore A is a Koszul nonhomogeneous
quadratic algebra. Since ψ0 = 0, the Koszul dual A! of A, which is the tensor
algebra over the dual vector space A∗ of A, i.e. A! = T (A∗ ), is a differential
quadratic Koszul algebra, (see also Theorem 2 b)).
3. Calculus on the twisted SU (2) group [20].
Let A be the nonhomogeneous quadratic algebra generated by 3 elements ∇0 , ∇1 , ∇2
with relations
⎧
⎪ 1
⎪
⎪ ν∇2 ∇0 − ∇0 ∇2 = ∇1
⎪
⎪ ν
⎪
⎪
⎪
⎨
1
(8.1) ν 2 ∇1 ∇0 − 2 ∇0 ∇1 = (1 + ν 2 )∇0
⎪
⎪ ν
⎪
⎪
⎪
⎪
⎪
⎪
⎩ ν 2 ∇2 ∇1 − 1 ∇1 ∇2 = (1 + ν 2 )∇2
ν2
with ν ∈ K\{0}. The quadratic part of A is the quadratic algebra A of Example 3
in Section 5 with relations given by (5.1), (i.e. the above relations where the non-
quadratic terms of the right-hand sides have been set equal to 0). This quadratic
algebra is Koszul of global dimension 3 and is Koszul-Gorenstein with polynomial
growth. It is not hard to verify that A has the PBW property by using Theorem 3
and Proposition 4 with ψ0 = 0 (since the nonquadratic terms are linear). It follows
that the Koszul dual A! of A, which is as explained in Section 5 the quadratic
algebra generated by ω0 , ω1 , ω2 in degree 1 with relations (5.4), is a differential
quadratic Koszul algebra; its differential is the antiderivation d given by
⎧
⎨ dω0 = ν 2 (1 + ν 2 )ω0 ω1
(8.2) dω1 = νω0 ω2
⎩
dω2 = ν 2 (1 + ν 2 )ω1 ω2
128
on the generators, which implies d2 = 0 on A! . This is the left covariant differential

calculus on the twisted SU (2) group of [20]. More precisely, it is a calculus on the
twisted SL(2) group and it becomes a calculus for the twisted SU (2) group if one
equips it with the involution
∇1 → ∇∗1 = ∇1 , ∇0 → ∇∗0 = −ν −1 ∇2 , ∇2 → ∇∗2 = −ν∇0
and the corresponding one on the dual basis of the ω’s.
4. Clifford algebra and generalization.
Given a symmetric bilinear form g on E, the Clifford algebra of g
A = T (E)/({x ⊗ y + y ⊗ x − 2g(x, y)1l|x, y ∈ E})
gives another example of nonhomogeneous quadratic Koszul algebra. In this case
the quadratic part A of A is the exterior algebra
A = ∧E
∗
with Koszul dual A = S(E ), so it is of infinite global dimension. The algebra A!
!
is a curved differential quadratic Koszul algebra with curvature F = −2g ∈ S 2 (E ∗ )

but with trivial differential d = 0.
More generally it is worth noticing that an antiderivation d of S(E ∗ ) vanishes
in degrees different from 1 since
d(αβ) = d(α)β − αd(β) = −(d(β)α − βd(α)) = −d(βα) = −d(αβ) = 0
for any α, β ∈ E ∗ . Furthermore
d(α)β = d(β)α
for any α, β ∈ E ∗ implies that there is a θ ∈ E ∗ such that one has
d(α) + θα = 0
∗
for any α ∈ E whenever d is of degree 1. Using Theorem 5 it follows that the
algebras of the form
A = T (E)/({x ⊗ y + y ⊗ x − θ(x)y − θ(y)x − 2g(x, y)1l})
with θ ∈ E and g ∈ S 2 (E ∗ ) are all the nonhomogeneous quadratic Koszul algebras
∗
having quadratic part A = ∧E.
9. Lie prealgebras
Throughout the following, A = T (E)/(P ) is a nonhomogeneous quadratic alge-

bra with homogeneous part A = T (E)/(R) and we assume that the conditions (i)
and (ii) of Theorem 3 are satisfied with ψ1 : R → E and ψ0 : R → K as before but
now we take ψ0 = 0. This latter assumption is equivalent to saying that A is an
augmented algebra, its augmentation ε : A → K being induced by the projection of
T (E) onto the degree zero component ( K). In this case F = 0 so A! is a graded
differential algebra in view of (7.4), with differential d induced by the transpose of
ψ1 . An algebra A as above is referred to as a quadratic-linear algebra, [14]. Notice
that then A is an augmented filtered algebra and that the subspace of F 1 (A) which
is in the kernel of the augmentation, i.e. p(E) = F 1 (A) ∩ Ker(ε) is the image of
E under the canonical projection p : T (E) → A. In the case where A has the
PBW property, E can be identified with its image p(E) and this is in fact already
LIE PREALGEBRAS 129
15
true here since we have assumed that the conditions (i) and (ii) of Theorem 3 are
satisfied [4] (see the remark at the end of Section 7). Theorem 5 has the following
counterpart.
Theorem 6. The anti-isomorphism of Theorem 5 induces an anti-isomorphism
between the category of quadratic-linear algebras and the category of differential
quadratic algebras which restricts to an anti-isomorphism between the category of
quadratic-linear Koszul algebras and the category of differential quadratic Koszul
algebras.
Example. Let g be a Lie algebra, then A = U (g) is of the above type with A = S(g)
so A! = ∧g∗ . As explained in Section 2 this gives an example of the above general
situation. Furthermore the correspondence g → ∧g∗ is an anti-isomorphism of
categories between the category of Lie algebras and the category of free graded-
commutative differential algebras generated in degree 1 (i.e. of the exterior algebras
equipped with differentials).
This example shows that the class of augmented nonhomogeneous quadratic
algebras A considered here generalizes the class of universal enveloping algebras of
Lie algebra. Notice that this class contains the algebra A = T (A)/({x⊗y−xy|x, y ∈
A}) associated to an associative algebra A as explained in Section 3. This latter
kind of quadratic-linear algebra is interesting but is clearly not a good analogue of
universal enveloping algebras of Lie algebras. This is why we shall restrict attention
to a smaller class of algebras where we impose the Poincaré duality corresponding
to the Gorenstein property and the assumption of Koszulity which is natural and
always satisfied by the examples we have in mind.
Let A be a Koszul nonhomogeneous quadratic algebra which is quadratic-linear
and such that its quadratic part A is Koszul-Gorenstein (see Section 4) and let E
be the vector space of elements of degree 1 of A identified with the corresponding
generating subspace of A, that is,
E = F 1 (A) ∩ Ker(ε)
where ε : A → K is the augmentation of A. Under these assumptions, E will be
referred to as a Lie prealgebra and A will be referred to as the enveloping algebra
of E . If, furthermore the quadratic part A has polynomial growth (see remark
b) in Section 5) we shall speak of a regular Lie prealgebra. Examples 1 and 3 of
the last section give examples of regular Lie prealgebras while Example 2 which is
quadratic-linear and Koszul does not correspond to a Lie prealgebra.
We formalize the notion of a Lie prealgebra with the following definition.
Definition 1. A Lie prealgebra is a triple (E, RE , ψE ) where E is a finite-
dimensional vector space, RE is a vector subspace of E ⊗ E and where
ψE : RE → E
is a linear mapping such that the quadratic algebra
AE = T (E)/(RE )
is Koszul-Gorenstein and such that the quadratic-linear algebra
AE = T (E)/({r + ψE (r)|r ∈ RE })
130
has the PBW property (so is Koszul). When no confusion arises concerning RE and
ψE the Lie prealgebra will be denoted simply by E; the quadratic-linear algebra AE is
its enveloping algebra. A morphism α from (E, RE , ψE ) to another Lie prealgebra
(E , RE , ψE ) is a linear mapping α : E → E such that (α ⊗ α)(RE ) ⊂ RE and

α ◦ ψE = ψE ◦ (α ⊗ α).
This defines the category of Lie prealgebras and E → AE is a covariant functor

from the category of Lie prealgebras to the category of quadratic-linear Koszul
algebras.
We now wish to describe completely the duality of Theorem 6 restricted to Lie
prealgebras. Let us recall that a Frobenius algebra is a finite-dimensional algebra A
such that one has the left A-module isomorphism A A∗ between A and its dual
vector space A∗ , where the left A-module structure of A∗ is induced by the right
A-module structure of A. This is equivalent to the existence of a nondegenerate
bilinear form (x, y) → B(x, y) on A such that
B(xy, z) = B(x, yz)
for x, y, z ∈ A. Then the linear mapping
x → B(•, x)
realizes the left A-module isomorphism A A∗ . Concerning the graded connected
case, one has the following result [17].
Proposition 7. Let A be a finite-dimensional graded connected algebra such
that AD = 0 and An = 0 for n > D. The following conditions (i) and (ii) are
equivalent:
(i) A is a Frobenius algebra,
(ii) dim(AD ) = 1 and (x, y) → (xy)D is nondegenerate,
where (xy)D is the component of xy in AD .
If A is a graded connected Frobenius algebra, there is an automorphism σ ∈ Aut(A)
of graded algebra (i.e. of degree 0) such that
xy = σ(y)x
for any x ∈ An and y ∈ AD−n with D as above and 0 ≤ n ≤ D.
Examples. The prototype of a graded Frobenius algebra as above is the exterior

algebra ∧E over a finite-dimensional vector space E with D = dim(E).
More generally, the Koszul dual A! of the quadratic algebra A of Example 4 in Sec-
tion 5, that is, the quadratic algebra generated by D elements ωα (α ∈ {1, . . . , D})
with relations given by (5.7) with the q’s satisfying (5.6) is a graded Frobenius
algebra as above. Indeed, a basis of this algebra is given by the ωi1 · · · ωik with
i1 < i2 < · · · < ik and one verifies that Condition (ii) of Proposition 7 is satisfied.
In this case, the automorphism σ of that proposition is given by

(9.1) σ(ωα ) = (−1)D−1 q αβ ωα
β=α
for α ∈ {1, . . . , D}.

A particular case of the above algebra is the exterior algebra over a D-dimensional
LIE PREALGEBRAS 131
17
vector space E which corresponds to q αβ = 1 for all α, β ∈ {1, . . . , D}. Formula

(9.1) then gives
σ(x) = (−1)(D−1)n x
for x ∈ ∧n E, 0 ≤ n ≤ D.
Another particular case of the above algebra with D = 3 is the quadratic algebra
A! with relations (5.4) in (Example 3) Section 5. Formula (9.1) reads
σ(ω0 ) = ν 6 ω0 , σ(ω1 ) = ω1 , σ(ω2 ) = ν −6 ω2
for the automorphism σ.
In the following we shall consider differential quadratic (Koszul) algebras which
are Frobenius and we shall call such an algebra a differential quadratic (Koszul)
Frobenius algebra. One has the corresponding category. It turns out that the
duality of Theorem 6 restricts as a duality between Lie prealgebras and differential
quadratic Koszul Frobenius algebras.
Theorem 8. Let E be a Lie prealgebra; then A!E is a differential quadratic
Koszul Frobenius algebra and this defines an anti-isomorphism between the cate-
gory of Lie prealgebras and the category of differential quadratic Koszul Frobenius
algebras.
This follows from the fact proved in [17] that a quadratic Koszul algebra A
of finite global dimension is Koszul-Gorenstein if and only if its Koszul dual A! is
Frobenius.
10. Complexes and homologies
In this section we define the generalization of the Chevalley-Eilenberg com-

plexes for a Lie prealgebra and relate their (co)homologies to the corresponding
Hochschild (co)homologies of its enveloping algebra generalizing thereby the rela-
tion between the Chevalley-Eilenberg (co)homologies of a Lie algebra g and the
corresponding Hochschild (co)homologies of its universal enveloping algebra U (g).
It turns out that several results of these developments are valid in the more general
context of quadratic-linear algebras. Accordingly, let us complete the definitions of
last section.
Definition 2. A prealgebra is a triple (E, RE , ψE ) where E is a finite dimen-
sional vector space, RE is a vector subspace of E ⊗ E and ψE : RE → E is a linear
mapping such that
AE = T (E)/({r + ψE (r)|r ∈ RE })
is a quadratic-linear algebra, that is, such that ψ1 = ψE and ψ0 = 0 satisfy the
conditions of Proposition 4. When no confusion arises concerning RE and ψE
the prealgebra will be denoted simply by E; the quadratic linear algebra AE is its
enveloping algebra. The morphisms of prealgebras are defined as in the case of
Definition 1 for Lie prealgebras.
One has therefore a category of prealgebras and the full subcategory of Lie pre-
algebras. Notice that the functor E → AE is such that one has E ⊂ AE canonically
and that one can recover the prealgebra (E, RE , ψE ) from the (augmented filtered)
quadratic-linear algebra AE .
132
Given a prealgebra (E, RE , ψE ) we let AE = A(E, RE ) be the quadratic part of

its enveloping algebra AE . One can reformulate Theorem 6 in the following manner
similar to Theorem 8.
Theorem 9. Let E be a prealgebra; then A!E is a differential quadratic algebra
and this defines an anti-isomorphism between the category of prealgebras and the
category of differential quadratic algebras.
A representation of the prealgebra E = (E, RE , ψE ), also called a left E-

module, is a vector space V equipped with a linear mapping π of E into End(V )
such that
(10.1) m(π ⊗ π(r)) + π(ψE (r)) = 0
for any r ∈ RE , where m : End(V ) ⊗ End(V ) → End(V ) denotes the product of
endomorphisms. This is the same thing as a left AE -module since it extends as a
representation of T (E) which vanishes on {r + ψE (r)|r ∈ RE } in view of (10.1).
One defines similarily a right representation of E or a right E-module as cor-
responding to a right AE -module; for instance if V is a left E-module, the dual
vector space V ∗ is canonically a right E-module.
Let (π, V ) be a representation of E as above and let
δπ(0) : V → V ⊗ E ∗
be the linear mapping defined by
δπ(0) (v)(e) = π(e)v
(0)
for e ∈ E and v ∈ V . The linear mapping δπ extends canonically as a right
A!E -module endomorphism
δπ : V ⊗ A!E → V ⊗ A!E
which is in fact an endomorphism of degree 1 of the free graded right A!E -module
V ⊗ A!E .
Let d be the differential of A!E and let us denote again by d the linear mapping
of degree 1 of V ⊗ A!E into itself defined by
d(v ⊗ w) = v ⊗ dω
for v ∈ V and ω ∈ A!E . One has the following result.
Lemma 10. The linear mapping of degree 1
δπ + d : V ⊗ A!E → V ⊗ A!E
satisfies
(δπ + d)2 = 0
i.e. the graded vector space V ⊗ A!E endowed with dπ = δπ + d is a cochain complex
of vector spaces.
Proof. One has

(δπ + d)v ⊗ ω = δπ (v)ω + v ⊗ dω
LIE PREALGEBRAS 133
19
for v ∈ V and ω ∈ A!E . This implies
(δπ + d)2 v ⊗ ω = δπ2 (v)ω + δπ (v)dω + d(δπ (v))ω − δπ (v)dω + v ⊗ d2 ω

= (δπ2 (v) + d(δπ (v)))ω.
On the other hand equation (10.1) is equivalent to
(10.2) δπ2 (v) + dδπ (v) = 0
for any v ∈ V .
The cochain complexes (V ⊗A!E , dπ ) generalize the Chevalley-Eilenberg cochain
complexes. Indeed, if E is a Lie algebra g then A!E = ∧g∗ and if (π, V ) is a
representation of g then (V ⊗ ∧g∗ , dπ ) is a Chevalley-Eilenberg cochain complex
and all these cochain complexes are of this form.
By duality one obtains a chain complex (V ∗ ⊗ A!∗ ∗
E , dπ ) from the complex (V ⊗
!∗ ∗
A!E , dπ );
in particular one has the chain complex (AE , d ) which is a differential
coalgebra. More generally, given a right representation (W, π) one has the chain
∗
complex (W ⊗ A!∗ E , dπ ). These chain complexes generalize the Chevalley-Eilenberg
chain complexes.
We now assume in the sequel of this section that E = (E, RE , ψE ) is a Lie
prealgebra, i.e. that the enveloping algebra AE is Koszul and that its quadratic
part AE is Koszul-Gorenstein. This is equivalent to saying that A!E is a differential
quadratic Koszul Frobenius algebra in view of Theorem 8. In this case, one has
(10.3) H • (V ⊗ A!E ) = Ext•AE (K, V )
for a representation V of E and
(10.4) H• (W ⊗ A!∗
E ) = Tor• (W, K)
AE
for a right representation of E [14], where H • (V ⊗ A!E ) is the cohomology of

∗ !∗ ∗
(V ⊗A!E , dπ ) and H• (W ⊗A!∗
E , dπ ) is the homology of (W ⊗AE , dπ ). This generalizes
what happens in the case where E is a Lie algebra g, [6], [11], [19].
Recalling that for an algebra A and a (A, A)-bimodule M, that is, an Ae -module
M (with Ae = A ⊗ Aopp ), one has
H • (A, M) = Ext•Ae (A, M)
for the Hochschild cohomology H • (A, M) of A and

e
H• (A, M) = TorA
• (M, A)
for the Hochschild homology H• (A, M) of A with values in M, and remembering

that K can be considered as a left or as a right AE -module by using the aug-
mentation ε : AE → K, one deduces from (10.3) and (10.4) relations between the
above (co)homologies (of E) and the Hochschild (co)homologies of the enveloping
algebra AE of E which generalize the relations between the Chevalley-Eilenberg
(co)homologies of a Lie algebra g and the Hochschild (co)homologies of its univer-
sal enveloping algebra U (g).
134
11. Further prospects (tentative conclusion)
It is natural to expect that the differential calculi on quantum groups fall in

the framework described above, namely the Koszul duality of Positselski starting
from Lie prealgebras. In fact, it is usually not difficult to show that the quantum
tangent space E (see e.g. [10], Chapter 14 for the definition) is a prealgebra in
the sense of Definition 2 and that the corresponding differential quadratic algebra
A!E is a Frobenius algebra. Indeed, in most cases A!E is explicitly given as a
differential quadratic algebra which is by inspection a Frobenius algebra. However,
it is generally more involved to prove the Koszulity of the enveloping algebra AE ,
that is here the Koszulity of its quadratic part AE .
In other words, it is generally easy to show that the quantum tangent spaces
are weak Lie prealgebras in the sense of the following definition.
Definition 3. A weak Lie prealgebra is a prealgebra E such that the differential
quadratic algebra A!E is a Frobenius algebra.
Given such a weak Lie prealgebra E, it is a Lie prealgebra if and only if AE is

Koszul, or equivalently if and only if A!E is Koszul.
For instance, we have shown that relations (8.1) define a Lie prealgebra; it
turns out that this is the quantum tangent space corresponding to the left covari-
ant differential calculus on the twisted SL(2) group of [20]. On the other hand one
shows easily that the quantum tangent spaces corresponding to the 4D± bicovari-
ant differential calculi on the same twisted (quantum) SL(2) group are weak Lie
prealgebras. We conjecture that they are in fact Lie prealgebras, but to show this
one has to prove the Koszulity of the corresponding algebras.
It is worth noticing that the Koszul property is a very nice and desirable prop-
erty which here is furthermore natural in view of the duality described in Theorem
8 as well as in view of the (co)homological results described in last section. In con-
trast, Definition 3 is somehow artificial and corresponds to intermediate properties.
Thus an important task remains to show that the usual quantum tangent spaces
corresponding to the differential calculi on quantum groups are Lie prealgebras.
Finally we mention that we are aware that there are many very interesting
related papers on so-called quantum Lie algebras. However, to the best of our
knowledge the description given in the present paper of the basic structures which
are relevant is new.
References
[1] M. Artin and W.F. Schelter. Graded algebras of global dimension 3. Adv. Math., 66:171–216,
1987.
[2] M. Artin, J. Tate, and M. Van den Bergh. Modules over regular algebras of dimension 3.
Invent. Math., 106:335–388, 1991.
[3] R. Berger. Dimension de Hochschild des algèbres graduées. C.R. Acad.Sci. Paris, Ser. I,
341:597–600, 2005.
[4] A. Braverman and D. Gaitsgory. Poincaré-Birkhoff-Witt theorem for quadratic algebras of
Koszul type. J. Algebra, 181:315–328, 1996.
[5] H. Cartan. Homologie et cohomologie d’une algèbre graduée. Séminaire Henri Cartan,
11(2):1–20, 1958.
LIE PREALGEBRAS 135
21
[6] H. Cartan and S. Eilenberg. Homological algebra. Princeton University Press, 1973.
[7] M. Dubois-Violette. Lectures on graded differential algebras and noncommutative geometry.
In Y. Maeda et. al. editors, Noncommutative Differential Geometry and Its Applications to
Physics, pages 245–306. Shonan, Japan, 1999, Kluwer Academic Publishers, 2001.
[8] G. Fløystad. Koszul duality and equivalences of categories. Trans. Amer. Math. Soc.,
358:2373–2398, 2006.
[9] D.I. Gurevich. Algebraic aspects of the quantum Yang-Baxter equation. Algebra i Analiz,
2:119–148, 1990 (Transl. in Leningrad Math. J. 2:801–828 1991).
[10] A. Klimyk and K. Schmüdgen. Quantum groups and their representations. Springer, 1997.
[11] J.L. Loday. Cyclic homology. Springer Verlag, New York, 1992.
[12] Yu. I. Manin. Some remarks on Koszul algebras and quantum groups. Ann. Inst. Fourier,
Grenoble, 37:191–205, 1987.
[13] Yu. I. Manin. Quantum groups and non-commutative geometry. CRM Université de Montréal,
1988.
[14] A. Polishchuk and L. Positselski. Quadratic algebras, volume 37 of University Lecture Series.
Amer. Math. Soc., Providence, RI., 2005.
[15] L. Positselski. Nonhomogeneous quadratic duality and curvature. Func. Anal. Appl., 27:197–
204, 1993.
[16] A. Pottier. Stabilité de la propriété de Koszul pour les algèbres homogènes vis-à-vis du produit
semi-croisé. C.R. Acad. Sci. Paris, Série I, 343:161–164, 2006.
[17] S.P. Smith. Some finite dimensional algebras related to elliptic curves. CMS Conf. Proc.
Proc., 19:315–348, 1996.
[18] M. Wambst. Complexes de Koszul quantiques. Ann. Inst. Fourier, Grenoble, 43:1083–1156,
1993.
[19] C.A. Weibel. An introduction to homological algebra. Cambridge University Press, 1994.
[20] S.L. Woronowicz. Twisted SU(2) group. an example of noncommutative differential calculus.
Publ. RIMS, Kyoto Univ., 23:117–181, 1987.
[21] S.L. Woronowicz. Differential calculus on compact matrix pseudogroups (quantum groups).
Commun. Math. Phys., 122:129–170, 1989.
Laboratoire de Physique Théorique, UMR 8627, CNRS et Université Paris-Sud 11,

Bâtiment 210, F-91 405 Orsay Cedex
E-mail address: Michel.Dubois-Violette@u-psud.fr
Dipartimento di Matematica e Informatica, Università di Trieste, Via A. Valerio

12/1, I-34127 Trieste, Italy and INFN, Sezione di Trieste, Trieste, Italy
E-mail address: landi@units.it

Volume 546, 2011
On the Analogy Between Complex Semisimple Groups

and Their Cartan Motion Groups
Nigel Higson
To Henri, with admiration.
Abstract. In a 1975 article George Mackey examined analogies between the

representations of a semisimple Lie group and those of its Cartan motion group.
Alain Connes later pointed out that the analogies observed by Mackey harmo-
nize very well with the Connes-Kasparov conjecture in C ∗ -algebra K-theory.
Motivated by Connes’ observation, the present author analyzed the reduced
C ∗ -algebra of a complex semisimple group in a way that led to a new confir-
mation of the Connes-Kasparov conjecture for such groups, while at the same
time exhibiting a natural bijection between the tempered duals of the group
and its Cartan motion group. The purpose of this article is examine the same
issues algebraically rather than C ∗ -algebraically, and so obtain a Mackey-type
bijection between the admissible dual of a complex semisimple group and that
of its motion group.
1. Introduction
My title pays homage to the article of Mackey [Mac75] that I mentioned in
my abstract. Mackey examined the concept from mathematical physics of the con-
traction of a Lie group to a Lie subgroup [IW53], and argued that the irreducible
unitary representations of a semisimple group G should correspond to the irre-
ducible representations of its contraction to a maximal compact subgroup K. The
contraction is by definition the group
Gc = K Lie(G)/ Lie(K).
In the context of semisimple groups this is also called the Cartan motion group
associated to G.
Mackey didn’t specify what the correspondence between the unitary duals of G
and Gc ought to be, but it is clear from the text of his paper that he had in mind a
measure-theoretic equivalence. In interesting contrast to this, Connes later pointed
out that the Connes-Kasparov conjecture in C ∗ -algebra K-theory suggests that
2010 Mathematics Subject Classification. Primary 22E47; Secondary 16S30, 20C08, 46L80.
Key words and phrases. Representations, Complex semisimple groups, Mackey analogy.
The author was supported in part by a grant from the US National Science Foundation.
2011
c 0000
c Mathematical
holder)
1
137
138
2 NIGEL HIGSON
the duals of G and Gc ought to correspond with one another K-theoretically. See
[BCH94, p. 263]. Connes used the reduced duals of G and Gc , or equivalently, in
representation-theoretic language, their tempered duals. This is broadly consistent
with Mackey’s view, if “measure” is taken to mean “Plancherel measure.”
Nevertheless there is an interesting tension between the two ideas since of course
measure-theoretic correspondences need not preserve K-theory, nor vice versa. Be-
cause of this, the two views together might suggest to an optimistic mind that the
reduced duals of G and Gc ought to correspond to one another not only measure-
theoretically and K-theoretically, but exactly.
In a recent paper [Hig08] I examined this idea in the case where G is a con-
nected complex semisimple group. The classification of irreducible tempered repre-
sentations was completed around the time of Mackey’s work, and I pointed out that
with the classification in hand it is rather easy to construct a simple and natural
bijection between the reduced duals of G and Gc .
The bijection takes care of Mackey’s correspondence, but to go further and
understand Connes’ K-theoretic equivalence an additional idea is needed. This
is Vogan’s concept of minimal K-type, introduced in [Vog79]. The K-types of a
representation of G are its nonzero K-isotypical components. The usual highest
weight theory partially orders the K-types, and it turns out that the irreducible
representations of a complex semisimple group have unique minimal K-types. It
follows that the tempered dual is partitioned into locally closed subsets according
to minimal K-type. As locally closed subsets of the reduced duals, the minimal
K-type components correspond to subquotients of the reduced C ∗ -algebras of G
and Gc . I showed in [Hig08] that these subquotients are Morita equivalent to one
another. So the reduced C ∗ -algebras of G and Gc are assembled from the same
components (which are abelian C ∗ -algebras by a multiplicity one result) using the
operations of Morita equivalence, C ∗ -algebra extension and direct limit.
The C ∗ -algebra analysis can be taken an important step further. The motion
group Gc fits into a smooth one-parameter family {Gt } of Lie groups in which all the
other fibers apart from G0 = Gc are copies of G. This is a sort of deformation from
Gc to G; it is an instance of the deformation to the normal cone construction from
geometry. The corresponding reduced C ∗ -algebras Cλ∗ (Gt ) form a continuous field,
and the main theorem from [Hig08] is that this continuous field is assembled from
constant fields by Morita equivalences, extensions and a direct limit. This certainly
places the reduced duals of G and Gc in bijection with one another. Moreover it
leads immediately to a new proof of the Connes-Kasparov conjecture for G (the
original is due to Penington and Plymen [PP83]). So it unifies the perspectives of
Mackey and Connes.
Two of my students have studied extensions of [Hig08]: Chris George detailed
a natural bijection between the reduced duals of G = SL(n, R) and its motion
group [Geo09], while John Skukalek extended all the results of my paper to finite
extensions of complex semisimple groups [Sku10]. These and other calculations
make it clear that the Mackey/Connes phenomenon, including the minimal K-type
explanation for it that I have just sketched, is rather general. But my aim here
is to return to the case of connected complex semisimple groups in an effort to
understand the phenomenon a bit more fundamentally.
I shall analyze the admissible dual of G, comprised of equivalence classes of
irreducible (g, K)-modules, rather than the tempered dual, and I shall work with
COMPLEX SEMISIMPLE GROUPS AND THEIR MOTION GROUPS 139
3
convolution algebras of compactly supported distributions rather than with C ∗ -

algebras. But the conclusions will be more or less the same and moreover the
arguments will be in some ways more direct (for example the representation theories
of G and Gc will be derived from scratch). This is because the deformation from Gc
to G, which is fundamental to the whole story, has a very simple algebraic origin.
The convolution algebra of G that I shall be using is filtered by the usual concept
of order of a distribution, and the deformation {Gt } is a geometric counterpart
of the standard algebraic deformation from the associated graded algebra to the
algebra itself. I shall show that this algebraic deformation is assembled from trivial
deformations essentially as in the C ∗ -algebra case, with the theory of minimal K-
types playing the same role.
To get a sense of what is involved, consider the case of the trivial minimal
K-type. This is the smallest possible K-type, so every representation that includes
it automatically includes it as a minimal K-type. The problem is therefore to
understand all the irreducible representations of a complex semisimple group G
that contain the trivial representation of K, and to relate this space to the same for
the motion group Gc . The associated subquotient is actually a subalgebra of the
convolution algebra in this case (because the trivial K-type is the smallest possible
K-type). It is the algebra of K-bi-invariant distributions on G that are supported
on K, and it is isomorphic to the center of the enveloping algebra of g. The Harish-
Chandra isomorphism theorem computes the center of U(g), and in fact it identifies
the center with its associated graded algebra. So Harish-Chandra’s theorem plays
the role of a rigidity theorem, trivializing the algebraic deformation, as required.
For higher minimal K-types, algebras of the form
K
(1.1) R(g, σ) ∼
= U(g) ⊗ End(Vσ )op
come into play. Here Vσ is the representation space of a finite-dimensional irre-
ducible representation σ of K. The algebra (1.1), which incidentally has been stud-
ied in detail for other reasons by Kirillov [Kir01], corresponds to the open subset
of the dual consisting of all representations that contain the K-type σ. The alge-
bra that corresponds to the representations with minimal K-type σ is a quotient
of (1.1). A crucial theorem of Vogan [Vog79], reproduced here as Theorem 5.7,
identifies the quotient as the image of a generalized Harish-Chandra homomorphism
GHCσ : R(g, σ) −→ U(h).
Here h is a Cartan subalgebra of g. A generalization of Harish-Chandra’s theorem
(proved using essentially the same methods as in the classical case) computes the
image and identifies it with an associated graded counterpart, and this trivializes
the deformation associated to the minimal K-type σ.
Here is the grand conclusion: the order filtration on the Hecke algebra of a
complex semisimple group determines an algebraic deformation, the special fiber
of which is the Hecke algebra for the Cartan motion group of G. This deformation
is assembled from constant deformations, each associated to a minimal K-type. In
particular the admissible duals of G and Gc decompose into components according
to minimal K-type, and these components for G and Gc are isomorphic to one
another as affine varieties.
The main result includes in particular a calculation of the admissible dual of a
complex semisimple group. This problem has quite a long history, beginning with
Berezin [Ber62]. It was solved in full for the first time by Zhelobenko [Zhe74] and
140
4 NIGEL HIGSON
then solved in other ways in [Duf75, Enr81, BG80], among others. Certainly
Vogan’s more general classification results [Vog79] (for real as well as complex
groups) cover everything that is done here, and I will borrow heavily from his
approach. I do not claim any originality as far as the classification is concerned,
but the conceptual interpretation of the classification in terms of the Cartan motion
group is new, I believe, and I hope that the idea of viewing the classification problem
for representations as a sort of deformation (or rigidity) problem will prove to be of
some value. In any case I think it is interesting to reflect on the somewhat unlikely
origins of this point of view in measure theory and topological K-theory.
Here is an outline of the paper. Section 2 reviews basic ideas concerning Hecke
algebras and (g, K)-modules and Section 3 introduces the Harish-Chandra homo-
morphisms that we shall use throughout. Sections 4 and 5 mostly follow Vogan’s
work in [Vog79] and analyze minimal K-type representations in terms of Lie al-
gebra cohomology. Sections 6, 7 and 8 draw everything together by introducing
the Cartan motion group Gc and setting the results of the previous sections in the
context of the deformation from Gc to G. Finally Section 9 touches upon some of
the many open issues in this area.
2. Hecke Algebras
Let G be a Lie group and let K be a compact subgroup of G. I shall review the
construction of the Hecke algebra R(g, K) whose (nondegenerate) modules corre-
spond to (g, K)-modules. Then I shall examine certain subalgebras R(g, σ) whose
irreducible modules correspond to irreducible (g, K)-modules with a given non-zero
isotypical summand. These will be the focus of attention for the rest of the paper.
Hecke Algebra of a Compact Group. First, consider the compact group

K by itself. The theory here is simple and well known.
2.1. Definition. Let Z be a complex linear representation of the compact
group K. A vector z ∈ Z is K-finite if the set { kz : k ∈ K } spans a finite-
dimensional subspace of Z, and if the map k → kz is continuous (this requirement
makes sense since its range lies in a finite-dimensional vector space).
2.2. Remark. To be consistent with the definition, when dealing with finite-
dimensional representations of K I shall always assume that they are continuous
(and therefore smooth).
The K-finite terminology applies in the obvious way to functions that are de-
fined on spaces on which the group K acts, and we make the following definition.
2.3. Definition. Fix a Haar measure on K. The Hecke algebra R(K) of a
compact Lie group K is the convolution algebra of complex-valued functions on K
that are K-finite for the right translation action of K on itself.
The elements of R(K) are easy to characterize: they are precisely the matrix
coefficient functions of (continuous) finite-dimensional representations of K. They
are K-finite for both the right and left translation actions of K on itself.
2.4. Definition. A linear representation of K on a complex vector space Z is
locally finite if every vector in Z is K-finite, or equivalently, if it is a direct sum of
(continuous) finite-dimensional representations.
5
If Z is a locally finite representation of K, then the formula

fz = f (k) kz dk
K
equips W with the structure of a left R(K)-module (the integral makes sense be-
cause the integrand assumes continuously-varying values in a finite-dimensional
vector space).
2.5. Definition. Let σ : K → Aut(Vσ ) be a finite-dimensional and irreducible
representation of K. The isotypical projection associated to σ is the K-finite func-
tion
dim(Vσ )
pσ (k) = Trace σ(k−1 ) : Vσ → Vσ .
vol(K)
The function pσ is a central idempotent in R(K). The isotypical projections
corresponding to inequivalent irreducible representations are orthogonal to one an-
other: the product of any two such is zero.
If Z is a locally finite representation of K, then Z decomposes as a direct sum
of (locally finite) subrepresentations

Z= pσ Z
σ
(the sum is over representatives of the equivalence classes of finite-dimensional and
irreducible representations). The subrepresentation pσ Z is the σ-isotypical compo-
nent of Z; it can alternately be characterized as the range of the K-equivariant
evaluation homomorphism
(2.1) HomK (Vσ , Z) ⊗ Vσ −→ Z.
In fact (2.1) is an isomorphism onto pσ Z, so that a locally finite representation has
a canonical isotypical decomposition

Z∼= HomK (Vσ , Z) ⊗ Vσ .

σ∈K
Considered as an R(K)-module, Z is non-degenerate in the sense that

R(K)Z = Z.
This is clear from the fact that the pσ are idempotents. Conversely, if Z is a
nondegenerate left R(K)-module, then the formula
k · (f z) = (λ(k)f )z,
involving the left-regular representation λ of K on R(K), gives rise to a well-defined
locally finite representation of K on Z. In this way the category of locally finite
representations of K becomes isomorphic to the category of nondegenerate left
R(K)-modules.
Of course the algebraic structure of the Hecke algebra R(K) is very simple.
The isotypical decomposition

R(K) = pσ R(K)
σ
of the left regular representation is an algebra isomorphism. Moreover the action
of R(K) on the representation space Vσ induces an algebra isomorphism
∼
=
(2.2) pσ R(K) −→ End(Vσ )
142
6 NIGEL HIGSON
and we arrive at the isomorphism

∼
=

(2.3) R(K) −→ End(Vσ ).
σ
Hecke Algebra of a Noncompact Group. The noncompact case is a bit

more complicated. The appropriate Hecke algebra was defined by Flath [Fla79]
and is examined carefully in Chapter I of the book of Knapp and Vogan [KV95],
to which we refer the reader for details in what follows.1
Although it is possible to give the basic definitions analytically, using distribu-
tions as in my introduction, I shall instead take a more algebraic approach that is
better suited to the rest of the paper.
2.6. Definition. Denote by g the Lie algebra of G and by gC its complexifi-
cation. A (g, K)-module is a complex vector space Z that is simultaneously and
compatibly a complex gC -module and a locally finite representation of K. The
compatibility conditions are that
(a) the complex-linear map gC −→ End(Z) be K-equivariant, and
(b) its restriction to kC be the infinitesimal form of the K-action.
If Z is a (g, K)-module, then there are action maps
U(gC ) ⊗ Z −→ Z and R(K) ⊗ Z −→ Z,
and they combine to form an action map
(2.4) R(g, K) ⊗ Z −→ Z
where R(g, K) is the vector space
(2.5) R(g, K) = U(gC ) ⊗U(kC ) R(K).
The formula for (2.4) is the obvious one:
(S ⊗ T ) · z = S · (T · z).
The vector space R(g, K) is naturally a left U(gC )-module. In addition the adjoint
action of K on U(gC ) and the left regular representation of K on R(K) combine to
give a left action of K on R(g, K). So it carries the structure of a (g, K)-module,
and there is therefore an action map
(2.6) R(g, K) ⊗ R(g, K) −→ R(g, K).
This is an associative product. It is compatible with the natural left U(gC )- and
right R(K)-module structures on R(g, K).
2.7. Definition. The Hecke algebra R(g, K) is the associative algebra given
by (2.5) and (2.6).
The action map (2.4) gives each (g, K)-module a non-degenerate R(g, K)-
module structure. Suppose conversely that Z is a nondegenerate R(g, K)-module.
The Hecke algebra R(K) is a subalgabra of R(g, K), and the R(K)-module struc-
ture on Z that is obtained by restriction is again nondegenerate. So Z carries a
locally finite representation of K. In addition the formula
(2.7) S · (T · z) = (S · T ) · z,
1My notation differs a bit from [KV95], which uses (g , K) where I use (g, K).
C
7
where S ∈ U(gC ), T ∈ R(g, K) and z ∈ Z, provides Z with a compatible gC -module

structure and hence a (g, K)-module structure.
2.8. Theorem. The functor that associates to each nondegenerate left R(g, K)-
module the (g, K)-module above is an isomorphism of categories.
Spherical Hecke Algebras. Fix an irreducible and finite-dimensional rep-
resentation σ of K and let pσ ∈ R(K) be its isotypical projection. The adjoint
actions of the compact group K on U(gC ) and on R(K) induce an action of K on
the Hecke algebra R(g, K) by algebra automorphisms. This is the action used in
the following definition (compare [God52, Sec. 3] or Warner [War72, Chap. 4]).
2.9. Definition. The σ-spherical Hecke algebra is the subalgebra
R(g, σ) = R(g, K)K ∩ R(g, K)pσ
= R(g, K)K ∩ pσ R(g, K)pσ
of the K-fixed part of the Hecke algebra R(g, K).
If Z is a (g, K)-module, then the space HomK (Vσ , Z) carries a nondegener-
ate left action of R(g, σ), arising from the left R(g, K)-action on Z. Of course
HomK (Vσ , Z) is zero if the σ-isotypical component of Z is zero. However:
2.10. Proposition. The correspondence
W → HomK (Vσ , W )
induces a bijection from equivalence classes of irreducible (g, K)-modules with non-
zero σ-isotypical component to equivalence classes of irreducible R(g, σ)-modules.
To prove the proposition, consider the algebra
A(g, σ) = pσ R(g, σ)pσ .
If the matrix algebra End(Vσ ) is realized as R(K)pσ , as in (2.2), then End(Vσ ) and
R(g, σ) are commuting subalgebras of A(g, σ). Indeed, R(g, σ) is the commutant
of End(Vσ ) in A(g, σ). Multiplication gives an algebra homomorphism
(2.8) R(g, σ) ⊗ End(Vσ ) −→ A(g, σ).
and it is a general fact about commutants of matrix algebras that:
2.11. Lemma. The algebra homomorphism (2.8) is an isomorphism.
Proof. See [War72, Proposition 4.5.1.8] or [Jac64, p. 118].
Proof of Proposition 2.10. Because of the lemma, R(g, σ) and A(g, σ)

have isomorphic module categories via the operation of tensoring with Vσ . The
evaluation map (2.1) gives an isomorphism
HomK (Vσ , Z) ⊗ Vσ ∼
= pσ Z
and so it suffices to prove that the functor Z → pσ Z gives rise to a bijection between
equivalence classes of irreducible R(g, σ)-modules with pσ Z = 0 and equivalence
classes of irreducible A(g, σ)-modules.
This is again a general fact. Suppose given a ring R with an idempotent p and
a subring defined by A = p R p. The functor Z → pZ maps irreducible R-modules
144
8 NIGEL HIGSON
with pZ = 0 to irreducible A-modules. In the reverse direction, given an irreducible

A-module M , form the R-module
Z = R p ⊗A M.
This need not be irreducible, but denote by Z the submodule consisting of all z ∈ Z
such that p R z = {0}. Then Z/Z is an irreducible R-module. The correspondence
M → Z/Z is inverse to Z → pZ on irreducible modules.
The product on R(g, σ) is a bit easier to describe than the product on R(g, K).
First, since R(K)pσ ∼= End(Vσ ), it follows that

R(g, σ) = U(gC ) ⊗U(k ) R(K)pσ ]K ∼
C = U(gC ) ⊗U(k ) End(Vσ )]K
C
Second, the tensor product on the right-hand side carries an algebra structure given
by the formula
(2.9) (S1 ⊗ T1 ) · (S2 ⊗ T2 ) = S1 S2 ⊗ T2 T1
where Sj ∈ U(gC ) and Tj ∈ End(Vσ ) for j = 1, 2. This is a well-defined and
associative product on the K-fixed part of U(gC ) ⊗U(kC ) End(Vσ ).
2.12. Lemma. The linear isomorphism
K ∼
=
U (gC ) ⊗U(kC ) End(Vσ ) −→ R(g, σ)
described above is an algebra isomorphism for the product (2.9) on the left-hand
side.
Proof. Let S ⊗ T ∈ R(g, K), as in (2.5). If X ∈ R(g, K)K , then
T · X = X · T,
and moreover the right action of T is the obvious one from (2.5). Now use the
natural left U(gC )-module structure on R(g, K) to write S ⊗ T = S · T . We find
from (2.7) that
S·T ·X =S·X ·T
as required.
Complex Semisimple Groups. Assume now that G is a complex Lie group.
Denote √by J : g → g the complex structure on g, that is, the operator of multiplica-
tion by −1. Denote by ḡ the complex conjugate of g. This complex Lie algebra is
identical to g as a real Lie algebra, but it is given the complex structure −J instead
of J.
The complexification gC = g ⊗R C is a complex Lie algebra with respect to
scalar multiplication on the second tensor factor; the fact that g is itself a complex
Lie algebra is ignored here. However the complex structure on g causes gC to
decompose as a direct sum of two ideals: the complex-linear embeddings
j : g −→ gC and j̄ : ḡ −→ gC
defined by the formulas

j(X) = 1
2 X − iJX and j̄(X) = 1
2 X + iJX
(in which X and JX are shorthand for X ⊗ 1 and JX ⊗ 1) have commuting ranges
and induce an isomorphism
(2.10) (j, j̄) : g ⊕ ḡ −→ gC .
9
Now let G be a connected complex semisimple group and let K be a maximal

compact subgroup of G. Then
(2.11) g = k ⊕ Jk
(the right-hand side is a direct sum of real vector spaces only). The Cartan invo-
lution θ : g → g is the real Lie algebra involution defined by complex conjugation
relative to the decomposition (2.11):
θ : X + JY → X − JY,
where X, Y ∈ k. This is a conjugate-linear automorphism of the complex Lie algebra
g, or equivalently it is an isomorphism
∼
=
θ : g −→ ḡ
of complex Lie algebras.
From here on we shall use the identification
∼
=
(2.12) (j, j̄ ◦ θ) : g ⊕ g −→ gC .
The diagonal copy of g in g ⊕ g is mapped in this way onto the Lie subalgebra
kC ⊆ gC . Accordingly, a (g, K)-module becomes a (g ⊕ g)-module with the property
that it decomposes into finite-dimensional representations of the diagonal copy of
g (each of which exponentiates to a holomorphic representation of G).
It is occasionally helpful to write the first and second summands of (2.12) as g1
and g2 , and the diagonal as g unadorned. With this notation, g1 ⊕ g2 decomposes
as a semidirect product
(2.13) g1 ⊕ g2 = g1 g
This gives an isomorphism
∼
=
U(g1 ) U(g) −→ U(g ⊕ g)
that, combined with Lemma 2.12, determines an isomorphism of algebras
K ∼
=
U(g1 ) ⊗ End(Vσ )op −→ R(g, σ)
in which the tensor product is now over C. The superscript op denotes the opposite
algebra, in accordance with the multiplication law (2.9). To summarize:
2.13. Proposition. The complex-linear embedding j : g → gC induces an iso-
morphism
K ∼
=
U(g) ⊗ End(Vσ )op −→ R(g, σ)
of associative algebras.
2.14. Remark. The algebras appearing on the left-hand side in Proposition 2.13
have been studied by Kirillov in [Kir00] and [Kir01] for purposes different than
ours. He calls them quantum family algebras. He has also studied the classical
family algebras, for which the enveloping algebra U(g) is replaced by the symmetric
algebra Sym(g). We shall encounter these too, in Section 6.
When doing calculations it is often simpler to compute with U(g1 ⊕ g2 )K rather
than with R(g, σ). There is a natural homomorphism from one to the other,
(2.14) U(g1 ⊕ g2 )K −→ R(g, σ)
given by the formula T → T · pσ .
146
10 NIGEL HIGSON
2.15. Lemma. The algebra homomorphism

U(gC )K −→ R(g, σ)
given by the formula T → T · pσ is surjective.
Proof. Use (2.13) and Proposition 2.13. Since K is connected then the rep-
resentation σ, viewed as a homomorphism from U(kC ) into End(Vσ ), is surjective.
The lemma follows.
3. Harish-Chandra Homomorphisms
Throughout this section (and indeed the rest of the paper) G will be a connected
complex semisimple Lie group with maximal compact subgroup K. My aim is to
define the generalized Harish-Chandra homomorphisms
GHCσ : R(g, σ) −→ U(h)
that I mentioned in the introduction. Later I shall analyze these in enough detail to
determine the irreducible representations of G, and at the same time relate them,
following Mackey and Connes, to the irreducible representations of the Cartan
motion group Gc .
Classical Harish-Chandra Homomorphism. I shall begin by recalling the

classical construction of Harish-Chandra [HC51]; see for example [KV95, Sec.
IV.7] or [Dix96, Sec. 7.4] for expositions that take approaches that are appropriate
to our concerns here.
I need to review a few basic concepts from semisimple Lie algebra theory, as
follows (see for example Serre’s short text [Ser01] for further details). Denote by h
a Cartan subalgebra of g. A root for the pair (g, h) is a non-zero functional α ∈ h∗
for which the space

gα = X ∈ g : [H, X] = α(H)X, ∀H ∈ h
is nonzero. It is possible to partition the set of all roots into positive and negative
roots, in such a way that α is positive if and only if −α is negative, and so that if
α1 and α2 are positive roots, then α1 + α2 is also a positive root, if it is a root at
all. There is more than one way to do this, but fix one. The direct sum

(3.1) n+ = gα
α>0
indexed by the set of positive roots, is a Lie subalgebra of g. So is its opposite,

n− = gα ,
α<0
indexed by the negative roots, and h normalizes both. There is a vector space direct
sum decomposition
g = n− ⊕ h ⊕ n+
in which the summands are subalgebras but not ideals.
Now denote by Z(g) the center of the enveloping algebra U(g). Harish-Chandra
defined a homomorphism
hc : Z(g) −→ U(h)
11
as follows. By the Poincaré-Birkhoff-Witt theorem, there is a vector space direct

sum decomposition

(3.2) U(g) ∼= U(n− )⊗U(h) ⊕ U(g)n+
in which the first summand is included into U(g) by the obvious multiplication
map. There is an associated linear projection operator
(3.3) U(g) −→ U(n− )⊗U(h)
onto the first summand of (3.2).
3.1. Lemma. The restriction of the projection (3.3) to the commutant of h in
U(g) is an algebra homomorphism whose image lies in U(h).
3.2. Definition. The Harish-Chandra homomorphism
hc : Z(g) −→ U(h)
is the restriction of the projection (3.3) to Z(g).
3.3. Remark. The Harish-Chandra homomorphism depends on the choice of
a system of positive roots, which determines n± and hence the direct sum decom-
position (3.2). It is conventional to remove this dependency by incorporating a
“ρ-shift” into the definition of hc. But it seems better not to do so here.
The Harish-Chandra homomorphism can be characterized using representation
theory, as follows. Let V be a finite-dimensional, complex-linear, irreducible irre-
ducible representation of g. There is, up to scalar multiplication, a unique highest
weight vector, characterized by the condition n+ vhighest = 0. It is an eigenvector
for the action on h, so that there is a highest weight 2 φ ∈ h∗ with
Hvhighest = φ(H)vhighest ∀H ∈ h.
From now on I shall write vhighest = vφ . In fact I shall write V = Vφ , since
the highest weight determines the representation. The action of Z(g) on V is
determined by the highest weight as follows:
(3.4) Sv = hc(S)(φ)v ∀v ∈ V ∀S ∈ Z(g).
Here U(h) has been identified with the algebra of polynomials on the complex dual
vector space h∗ , allowing the element hc(S) to be evaluated at φ ∈ h∗ .
A crucial feature of the Harish-Chandra homomorphism is its symmetry. Let
W be the Weyl group of (g, h) (see for example [Ser01, Chap. 5]) and define as
usual

(3.5) ρ = 12 α.
α>0
This is the half-sum of the positive roots. Define a ρ-shifted affine action of W on
h∗ by
(3.6) w ·ρ φ = w(φ + ρ) − ρ
There is an induced action on the algebra of polynomials, and hence on U(h).
3.4. Theorem (Harish-Chandra). The range of the Harish-Chandra homomor-
phism lies within the W -invariant part of U(h) for the ρ-shifted action (3.6).
2In the current Lie algebra context the term “weight” is a synonym for “complex-linear
functional on h”.
148
12 NIGEL HIGSON
See [HC51]. Many proofs are possible. Most involve calculating the action of
Z(g) on representations of one sort or another (for example principal series, Verma
modules, cohomology spaces, etc) and then finding equivalences between spaces on
which Z(g) acts as φ and w(φ + ρ) − ρ. An argument along these lines will be
sketched in Section 4.
3.5. Remark. Harish-Chandra famously proved that in addition the Harish-
Chandra homomorphism maps Z(g) onto the W -invariant part of U(h). I shall
examine this issue in Section 7.
Generalized Harish-Chandra Homomorphisms. Fix an irreducible finite-
dimensional representation Vσ of K with highest weight σ ∈ h∗ . I want to define
the generalized Harish-Chandra homomorphism
(3.7) GHCσ : R(g, σ) −→ U(h).
Rather than give an algebraic formula for it (which is possible but not very enlight-
ening) I shall define the homomorphism (3.7) in representation-theoretic terms.
First, recall some standard terminology:
3.6. Definition. A weight φ ∈ h∗ is positive if it is a nonnegative linear
combination of positive roots. This notion of positivity determines a partial order
on the real linear span of the roots. A weight φ is dominant if it is positive and if
φ ≥ w(φ) for every element w of the Weyl group.
A weight is the highest weight of an irreducible finite-dimensional representa-
tion of K if and only if it is dominant and (analytically) integral in the sense that
it exponentiates to a character of the maximal torus of K with Lie algebra h ∩ k.
3.7. Definition. Let φ1 and φ2 be dominant integral weights. Denote by
Wφ1 ,φ2 the tensor product representation
Wφ1 ,φ2 = Vφ1 ⊗ Vφ∗2
of g ⊕ g. Here Vφ1 is the irreducible finite-dimensional representation with highest
weight φ1 , and Vφ∗2 is the contragredient of Vφ2 .
3.8. Lemma. Under the restriction to the diagonal g ⊆ g ⊕ g the representation
Wσ+φ,φ includes the representation Vσ with multiplicity one. Moreover if Vτ is
included in Wσ+φ,φ , then τ ≥ σ.
Proof. The multiplicity of Vτ in Wσ+φ,φ is equal to
dim HomK (Vτ , Vσ+φ ⊗ Vφ∗ ),
and this may be calculated using the adjunctions
HomK (Vτ , Vσ+φ ⊗ Vφ∗ ) ∼
= HomK (Vτ ⊗ Vφ , Vσ+φ )
∼
= HomK (Vσ+φ , Vτ ⊗ Vφ )∗
as follows. Suppose that τ = σ. Since Vσ ⊗ Vφ has a unique highest weight vector
of weight σ + φ, up to scalar multiplication, there is a unique inclusion of Vσ+φ into
Vσ ⊗ Vφ up to scalar multiplication. If Vσ+φ includes into Vτ ⊗ Vφ , then of course
there must be a (σ + φ)-weight vector in the tensor product. But the weights of the
tensor product are all less than or equal to τ + φ, which is the sum of the highest
weights of the factors. So σ ≤ τ , as required.
13
The Hecke algebra R(g, σ) acts on the space HomK (Vσ , Wσ+φ,φ ), and since that
space is one-dimensional the action is via some character of the algebra R(g, σ). I
can therefore (just about) make the following definition:
3.9. Definition. Define the homomorphism
GHCσ : R(g, σ) −→ U(h).
by the requirement that R(g, σ) act through the character
GHC σ+φ
R(g, σ) −−−−→
σ
U(h) −−−−→ C
on HomK (Vσ , Wσ+φ,φ ) (the last arrow is evaluation at σ + φ ∈ h∗ ).
To make this a real definition it must be checked that if S ∈ R(g, σ), then
GHCσ (S)(σ + φ), as defined above, is indeed a polynomial in φ. But this is straight-
forward (and in any case, in a moment I shall give an explicit formula for GHCσ in
terms of the isomorphism in Proposition 2.13).
It will require some preparation to even state the most important properties of
the generalized Harish-Chandra homomorphism. But I can at least record here its
symmetry properties. Define
(3.8) Wσ = { w ∈ W | w(σ) = σ }.
I shall prove the following result in the next section (like the corresponding result
for the classical Harish-Chandra homomorphism, there are several ways to prove
it; the one I’ll give is not the simplest, perhaps, but it arises while developing ideas
that will be crucial for something else).
3.10. Theorem. The range of the generalized Harish-Chandra homomorphism
lies within the Wσ -invariant part of U(h) for the ρ-shifted action (3.6) of Wσ on
U(h).
To finish this section, here is a formula for the generalized Harish-Chandra
homomorphism. It involves the coproduct map
Δ : U(g) −→ U(g) ⊗ U(g),
which is of course the algebra homomorphism defined on generators X ∈ g by
Δ(X) = X ⊗ 1 + 1 ⊗ X.
Sweedler’s standard notation for the coproduct is

Δ(S) = S (1) ⊗ S (2) .

The summation index is suppressed.
Let vσ be a highest weight vector in Vσ and let v σ ∈ Vσ∗ be the dual vector of
weight −σ, normalized so that v σ , vσ = 1. Define vφ ∈ Vφ and v φ ∈ Vφ∗ similarly.
3.11. Lemma. The character associated to the action of the algebra
K
R(g, σ) ∼
= U(g) ⊗ End(Vσ )op
on HomK (Vσ , Wσ+φ,φ ) is given by the formula

(3.9) S ⊗ T → v σ , S (1) T vσ v φ , S (2) vφ ,

where Δ(S) = S (1) ⊗ S (2) .
150
14 NIGEL HIGSON
Proof. As was noted in Lemma 3.8, there is (up to scale) a unique nonzero
K-equivariant map Vσ ⊗ Vφ → Vσ+φ . Denote by v1 ·v2 the image under this map of
a tensor v1 ⊗ v2 . The unique (up to scale) embedding of Vσ into Vσ+φ ⊗ Vφ∗ is then
given by the formula

(3.10) v → v·vμ ⊗ v μ ,
μ
where the sum is over a basis {vμ } for Vφ (for simplicity take it to be a basis of
weight vectors) and the dual basis {v μ }. The formula is obtained by calculating on
the image in Vσ+φ ⊗ Vφ∗ of vσ ∈ Vσ .
It follows that the generalized Harish-Chandra homomorphism is given by

(3.11) GHCσ : S ⊗ T → v σ , S (1) T vσ Proj(S (2) ),

where Proj : U(g) → U(h) is the composition
U(g) −→ U(n− ) ⊗ U(h) −→ U(h)
of the projection associated to (3.2) with the projection onto U(h) with kernel
n− U(n− ) ⊗ U(h).
4. Kostant’s Theorem
My aim in this section and the next is to analyze the generalized Harish-
Chandra homomorphism in more detail. The main goal, which will be reached
by the end of the next section, is the following theorem (some supporting defini-
tions will be given in the next section, but their meanings are easy to guess now).
4.1. Theorem. Let G be a connected complex semisimple group G with maximal
compact subgroup K.
(a) Every irreducible (g, K)-module has a unique minimal K-type.
(b) The irreducible (g, K)-modules with minimal K-type σ correspond, via Proposi-
tion 2.10, to the irreducible R(g, K) modules that factor through the generalized
Harish-Chandra homomorphism GHCσ .
The theorem is due to Vogan [Vog79]. The proof will involve techniques from
Lie algebra cohomology, most notably Kostant’s theorem. The main purpose of
this section is to cover the necessary preliminaries, but using them I shall at the
end of the section prove Theorem 3.10.
Let a be a finite-dimensional Lie algebra and let Z be an a-module. The
Lie algebra cohomology groups H p (a, Z) may be computed from the Chevalley-
Eilenberg complex
(4.1) Z −→ ∧1 a∗ ⊗ Z −→ ∧2 a∗ ⊗ Z −→ · · · −→ ∧d a∗ ⊗ Z
where d is the vector space dimension of a.
For the purposes of this paper it is probably simplest to define the differential
in (4.1) by forming the crossed product algebra

V(a) = U(a) ∧∗ a ,
equipping it with the order −1 differential that is zero on U(a) and maps X ∈ ∧1 a
to X ∈ U(a), and then invoking the natural isomorphisms
(4.2) ∧∗ a ⊗ Z ∼
= Hom(∧∗ a∗ , Z) ∼
= HomU(a) (V(a), Z)
15
to define the differential on ∧∗ a ⊗ Z. See for example [CE56, p. 287]. This makes
it straightforward to check that the various constructions below are cochain maps.
In any case, in degree zero the differential in (4.1) is

(4.3) z → ωj ⊗ Xj z,
with {Xj } a basis for a and {ωj } the corresponding dual basis for a∗ . So H 0 (a, Z)
is the space of a-invariants in Z, as it ought to be.
If Z carries a module structure (for some ring) that commutes with the action
of a, then the differentials in (4.1) are compatible with the module structure on
cochain spaces given by r · (ω ⊗ z) = ω ⊗ r · z. The cohomology spaces therefore
inherit a module structure.
Suppose for example that Z is a (g, K)-module, and consider only the locally
finite action of g on Z through the diagonal embedding of g into g ⊕ g. The action
of [U(g) ⊗ U(g)]K on Z commutes with this g-action. In particular the action of
the spherical Hecke algebra R(g, σ) commutes with g.
In particular the [U(g) ⊗ U(g)]K -action on Z commutes with the subalgebra
n+ ⊆ g, and so the cohomology spaces H p (n+ , Z), formed from the complex
(4.4) Z −→ ∧1 n∗+ ⊗ Z −→ ∧2 n∗+ ⊗ Z −→ · · · −→ ∧d n∗+ ⊗ Z
are naturally [U(g) ⊗ U(g)]K - and R(g, σ)-modules.

The Lie algebra h, and hence U(h), acts on each of the spaces ∧p n∗+ ⊗ Z in
(4.4) through the product of the coadjoint action on ∧p n∗+ and the given action on
Z. Once again the differentials are equivariant, and so there is an induced action
on the cohomology spaces H p (n+ , Z). This allows us to consider the weight spaces
H p (n+ , Z)ψ ⊆ H p (n+ , Z)
associated to various ψ ∈ h∗ .
According to (4.3) the bottom cohomology group H 0 (n+ , Z) is the space of
highest weight vectors in Z. If σ is a dominant integral weight, and if vσ ∈ Vσ is a
highest weight vector, then we obtain an isomorphism
∼
=
HomK (Vσ , Z) −→ H 0 (n+ , Z)σ
by mapping the intertwiner T ∈ HomK (Vσ , Z) to the σ-highest weight vector T vσ ∈

Z. This is an isomorphism of R(g, σ)-modules and of course it helps explain our
interest in cohomology since we are aiming to classify irreducible (g, K)-modules
through the R(g, σ)-modules HomK (Vσ , Z).
Before venturing further with (g, K)-modules, however, I shall review Kostant’s
theorem, which calculates the cohomology H ∗ (n+ , Vτ ) of the finite-dimensional ir-
reducible representations of g.
4.2. Definition. If w is an element of the Weyl group W , then denote by |w|

the number of positive roots that w−1 maps to negative roots (recall, for example
from [Ser01, Chap. 5], that W acts as a group of permutations on the set of all
roots).
4.3. Definition. If ψ ∈ h∗ , then denote by Cφ the one-dimensional represen-

tation of h determined by ψ.
152
16 NIGEL HIGSON
The famous Weyl character formula is the identity

(4.5) Vτ − ∧1 n∗+ ⊗ Vτ + ∧2 n∗+ ⊗ Vτ + · · · + (−1)n ∧d n∗+ ⊗ Vτ

= (−1)|w| Cw(τ +ρ)−ρ

w∈W
in the Grothendieck ring of finite-dimensional representations of h.3 The spaces on

the left-hand side of (4.5) are precisely those that appear in the complex (4.4) for
Z = Vτ . So, by the usual Euler characteristic principle, (4.5) is equivalent to

(4.6) (−1)p H p (n+ , Vτ ) = (−1)|w| Cw(τ +ρ)−ρ .

p=0 w∈W
Kostant’s theorem refines (4.6) to a formula for each of the spaces H p (n+ , Vτ )
individually:
4.4. Theorem (Kostant [Kos61]). If Vτ is an irreducible finite-dimensional
representation of g with highest weight τ , then as an h-module,

H p (n+ , Vτ ) ∼
= Cw(τ +ρ)−ρ .
|w|=p
For an exposition see [KV95, Sec. IV.9]. The proof of the theorem can be
broken into two steps, both of which are useful for quite a bit more. The first and
simplest is this:
4.5. Lemma. Let Vτ be the irreducible finite-dimensional representation of g
with highest weight τ . Let w ∈ W and let p = |w|. The (w(τ + ρ) − ρ)-weight
component of the Chevalley-Eilenberg complex
Vτ −→ ∧1 n∗+ ⊗ Vτ −→ ∧2 n∗+ ⊗ Vτ −→ · · · −→ ∧d n∗+ ⊗ Vτ
is zero in all degrees except p, where it has dimension one.
This means that any vector of weight (w(τ + ρ) − ρ) in the Chevalley-Eilenberg
complex must lie in degree p and is necessarily a cocycle; moreover unless it is zero
it cannot be a coboundary. So it represents the generator in degree p cohomology
that is guaranteed to exist by Kostant’s theorem.
The proof of Lemma 4.5 quickly reduces to a simple combinatorial calculation.
I shall need that calculation for other purposes and it appears below as Lemma 4.18.
The second step in the proof of Kostant’s theorem is to show that all other
weight components of the Chevalley-Eilenberg complex are acyclic. This is a con-
sequence of the following result, which calculates the Z(g)-module structure of any
cohomology space H p (n+ , Z).
4.6. Theorem (Casselman and Osborne). Let Z be any g-module. The U(h)-
and Z(g)-module actions on H p (n+ , Z) are compatible via the Harish-Chandra ho-
momorphism: if x ∈ H p (n+ , Z), then
S · x = hc(S) · x
for every S ∈ Z(g).
3The Grothendieck ring is an integral domain, so one can equivalently express the h-module

Vτ as a quotient of the right-hand side of (4.5) by (−1)p ∧p n∗+ . This is the customary presen-
tation of Weyl’s formula.
17
See [CO75]; for an exposition see [KV95, Sec. IV.10]. The Casselman-Osborne
theorem and the easy part of Kostant’s theorem imply that the image of the Harish-
Chandra homomorphism must lie in U(h)W . Harish-Chandra’s theorem that the
image is all of U(h)W implies that no other weight space can occur in cohomology,
and this completes the proof of Kostant’s theorem.
My aim in the remainder of the section is to provide a cohomological description
of the generalized Harish-Chandra homomorphism along the lines of the Casselman-
Osborne theorem. I shall use it in the next section to prove Theorem 4.1; I shall
use it here to prove that the image of GHCσ lies in U(h)Wσ , more or less along the
lines just sketched for the classical Harish-Chandra homomorphism.
Vogan’s version of the Casselman-Osborne theorem is a bit more complicated
than the original. Use the direct sum decomposition4
U(g) = U(h)⊗U(n− ) ⊕ n+ U(g).
to define a linear projection map

(4.7) U(g) ⊗ U(g) −→ U(h)⊗U(n− ) ⊗ U(h)⊗U(n− ) .
4.7. Lemma. The restriction of the projection (4.7) to the commutant of h in
U(g) ⊗ U(g) is an algebra homomorphism. The image of the restriction lies in
U(h) ⊗ U(h).
4.8. Definition. If ψ ∈ h∗ is any weight, then denote by
Shiftψ : U(h) → U(h)
the algebra homomorphism extending the map H → H + ψ(H)I from h into U(h).
In Vogan’s theorem the following map replaces the Harish-Chandra homomor-
phism on Z(g).
4.9. Definition. The homomorphism
K
GHC : U(g) ⊗ U(g) −→ U(h) ⊗ U(h)
is the restriction of the projection (4.7) to the K-invariant part of U(g) ⊗ U(g),
followed by Shift2ρ ⊗ Shift2ρ .
Now let Z be any (g ⊕ g)-module. Vogan defines a certain homomorphism
(4.8) δ : ∧d n∗ ⊗ H p (n, Z) −→ H p+d (n⊕n, Z)
and proves the following:
4.10. Proposition (Vogan). If x is any element of ∧d n∗ ⊗ H p (n, Z), then
δ(S · x) = GHC(S) · δ(x)
for every S ∈ [U(g) ⊗ U(g)]K .
The homomorphism δ in (4.8) is obtained from a cochain map between Chev-
alley-Eilenberg complexes. To describe it I shall write n in place of n+ , and I shall
4This is not the same as the direct sum decomposition (3.2), but it is the convenient one to
use here. However the price for using it is the later appearance of a “shift” in Definition 4.9.
154
18 NIGEL HIGSON
label the first and second coordinate copies in n ⊕ n as n1 and n2 . I shall reserve n
for the diagonal copy in the direct sum. The map of complexes is then as follows:
(4.9) ∧d n∗1 ⊗ Z / ∧1 n∗ ⊗ ∧d n∗ ⊗ Z / ∧2 n∗ ⊗ ∧d n∗ ⊗ Z /
1 1
δ δ δ

∧d (n∗1 ⊕n∗2 ) ⊗ Z / ∧1+d (n∗ ⊕n∗ ) ⊗ Z / ∧2+d (n∗ ⊕n∗ ) ⊗ Z /.
1 2 1 2
The top row is the Chevalley-Eilenberg complex for the n-cohomology of ∧d n∗1 ⊗ Z.
The bottom row is the Chevalley-Eilenberg complex for the (n1 ⊕n2 )-cohomology
of Z shifted in degree by d.
The vertical maps δ are tensor products of the identity on Z with embeddings
(4.10) ∧p n∗ ⊗ ∧d n∗1 ⊆ ∧p+d (n∗1 ⊕n∗2 )
that are defined as follows. The direct sum n1 ⊕ n2 is the internal direct sum of the
diagonal n with n1 . This induces an isomorphism
(4.11) ⊕p+q=n ∧p n ⊗ ∧q n1 ∼
= ∧n (n1 ⊕ n2 )
in the usual way, and therefore projections
∧p+q (n1 ⊕ n2 ) −→ ∧p n ⊗ ∧q n1 .
The embeddings (4.10) are adjoint to these projections.
The homomorphisms δ give a map between complexes, making possible the
following definition:
4.11. Definition (Compare Vogan [Vog79, p. 12]). Denote by
δ : ∧d n∗ ⊗ H p (n, Z) −→ H p+d (n⊕n, Z)
or equivalently
(4.12) δ : H p (n, ∧d n∗ ⊗ Z) −→ H p+d (n1 ⊕n2 , Z),
the homomorphism between cohomology spaces defined by (4.9).
Proof of Proposition 4.10. This is proved by first directly calculating in
the top-degree case, where p = d, using the formulas
H d (n, Z) ∼
= ∧d n∗ ⊗ Z/nZ
and
H 2d (n1 ⊕n2 , Z) ∼
= ∧2d (n∗1 ⊕n∗2 ) ⊗ Z/(n1 ⊕n2 )Z.
The shift that is built into the definition of GHC is there to cancel out the action
of h ⊕ h on ∧2d (n∗1 ⊕n∗2 ), so that the top degree case of the proposition is the
assertion that the action of [U(g) ⊗ U(g)]K on Z/nZ corresponds, through the map
in Lemma 4.7, to the action of U(h) ⊗U(h) on Z/(n1 ⊕n2 )Z. The proof is completed
by a standard cohomological dimension-shifting argument. See [Vog79, Theorem
3.5] for details.
My next aim is to relate the homomorphism GHC to the generalized Harish-

Chandra homomorphism GHCσ .
19
4.12. Definition. Denote by [U(h) ⊗ U(h)]σ the quotient of the algebra U(h) ⊗
U(h) by the ideal generated by the elements
(H, H) − (σ − 2ρ)(H)I ∀H ∈ h.
In addition, denote by
K
GHCσ : U(g) ⊗ U(g) −→ U(h) ⊗ U(h) σ
the composition of GHC with the projection to the quotient.
4.13. Lemma. If Z is any (g ⊕ g)-module, then the action of the algebra U(h) ⊗
U(h) on the weight space
H d (n ⊕ n, Z)σ−2ρ
factors through the quotient [U(h) ⊗ U(h)]σ−2ρ .
Proof. The element (H, H) belongs to the diagonal copy of h, which by defi-
nition acts via the character σ − 2ρ on any (σ − 2ρ)-weight space.
4.14. Definition. Denote by

Π : U(h) ⊗ U(h) σ −→ U(h)
the algebra homomorphism that extends the map
(H1 , H2 ) → H1 − H2 + (σ − 2ρ)(H2 )I
from h ⊕ h into U(h).
4.15. Lemma. The algebra homomorphism Π above is an isomorphism.
Proof. The map extending H → (H, 0) is an inverse.
With these definitions, the homomorphisms GHCσ and GHCσ are related as
follows:
4.16. Proposition. The diagram
K GHCσ
(4.13) U(g) ⊗ U(g) / U(h) ⊗ U(h)
σ
∼
= Π

R(g, σ) / U(h)
GHCσ
is commutative.
Let φ be a dominant integral weight. If we compose the diagram (4.13) with
the evaluation map
U(h)
σ+φ
/ C,
then the route around the bottom of the extended diagram gives the action of
[U(g) ⊗ U(g)]K on the space H 0 (n, Wσ+φ,φ )σ , as in Section 3. The proposition will
be proved by calculating enough about the map
(4.14) δ : ∧d n∗1 ⊗ H p (n, Wσ+φ,φ )σ −→ H p+d (n1 ⊕n2 , Wσ+φ,φ )σ−2ρ
to show that
(a) it is non-zero (and therefore injective, since the domain is one-dimensional);
and
156
20 NIGEL HIGSON
(b) the action of [U(h) ⊗ U(h)]σ on the image of (4.14) is given by the composition
of the map Π with evaluation at σ + φ. Equivalently, the action of U(h) ⊗ U(h)
on the image is given by evaluation at the weight (σ + φ, −σ − 2ρ).
The proof will then follow from Proposition 4.10.
The unique-up-to-scale σ-highest weight vector in Wσ+φ,φ is

(4.15) wσ = vσ ·vμ ⊗ v μ ,
μ
as in (3.10), so the left-hand side of (4.14) consists of scalar multiples of

η ⊗ wσ = η ⊗ vσ ·vμ ⊗ v μ ,
μ
where
η = η1 ∧ · · · ∧ ηd
∗
and where {ηj } is a basis for n . For convenience use a basis of weight vectors for
the coadjoint representation of h on n∗ ; the weights are precisely the negative roots
which we can list as −αj according to the indexing of the basis.
To compute δ(η ⊗ wσ ) (initially at the cochain level) it is necessary to calculate
the image of η under the embedding
∧d n∗1 −→ ∧d (n∗1 ⊕ n∗2 ).
in (4.10). The embedding is induced from the projection
n1 ⊕ n2 −→ n1 , (X1 , X2 ) → X1 − X2 ,
and as a result,
η → (η1 , −η1 ) ∧ · · · ∧ (ηd , −ηd ) ∈ ∧d (n∗1 ⊕ n∗2 ).
The image of η in ∧d (n∗1 ⊕ n∗2 ) therefore decomposes as a sum of 2d terms, indexed
by partitions
I J = {1, . . . , d}.
The term indexed by a given I J has weight (−αI , −αJ ) ∈ h∗ ⊕ h∗ , where αI , αJ
are the sums of the positive roots indexed by I and J.
This and (3.10) show that δ(η ⊗ wσ ) decomposes as a sum of nonzero terms
with weights
(4.16) (−αI + σ + μ, −αJ − μ)
∗ ∗
in h ⊕ h , where μ ranges over the weights of Vφ .
Proof of Proposition 4.16. The weights in h∗ ⊕ h∗ that occur as weights

of H d (n ⊕ n, Wσ+φ,φ ) may by computed from Kostant’s theorem. Writing

(4.17) H d (n ⊕ n, Wσ+φ,φ ) ∼
= H p (n, Vσ+φ ) ⊗ H q (n, Vφ∗ ).
p+q=d
and considering just the summand (p, q) = (0, d) we find that (σ + φ, −φ − 2ρ)
occurs. This is because
(4.18) Vφ∗ ∼
= V−w0 (φ) ,
21
where w0 is the longest element5 of W . Thus

H d (n, Vφ∗ ) = Cw0 (−w0 (φ)+ρ)−ρ = C−φ−2ρ .
It suffices to show that the composition of (4.14) with the projection onto this
(σ + φ, −σ − 2ρ)-weight space is nonzero.
Now, in the decomposition of δ(η ⊗ wσ ) into nonzero vectors with weights as
in (4.16), the weight (σ+φ, −φ−2ρ) occurs in a unique term. So it follows from
Lemma 4.5 that the projection of δ(η ⊗ wσ ) onto the (0, d)-summand in (4.17) is
nonzero, as required.
An elaboration of the preceding argument proves that the image of the gener-
alized Harish-Chandra homomorphism
lies within the Wσ -invariants.
4.17. Lemma. Assume that w ∈ W and w(σ) = σ. There is a unique term in
(4.16) having weight
(w(σ + φ + ρ) − ρ, −w(φ + ρ) − ρ)
The extremal case w = e that we used in the proof of Proposition 4.16 is easy;
the general case requires a little calculation:
4.18. Lemma. Let γ be a sum of distinct positive roots and let w ∈ W . Then
(4.19) w−1 (γ) + ρ − w−1 (ρ) ≥ 0
with equality if and only if γ is the sum of those positive roots that w−1 transforms
to negative roots.
Proof. The term ρ − w−1 (ρ) is the sum of all those positive roots that w−1
converts to negative roots. After the application of w−1 , some of the positive roots
in the sum γ may be transformed to negative roots, but of course only those that
occur both in the sum and in ρ − w−1 (ρ). So the negative roots in the first term of
the expression
w−1 (γ) + (ρ − w−1 (ρ))
are canceled out against positive roots in the second term, and overall the expression
is a sum of positive roots.
Proof of Lemma 4.17. We are to consider the equations
−αI + σ + μ = w(σ + φ + ρ) − ρ
−αJ − μ = −w(φ + ρ) − ρ.
Using w(σ) = σ the first equation rearranges to
w−1 (μ) − φ = ρ − w−1 (ρ) + w−1 (αI ).
But
w−1 (μ) − φ ≤ 0
since w−1 (μ) is a weight of Vφ and φ is the highest weight, while
0 ≤ ρ − w−1 (ρ) + w−1 (αI ),
5This is the unique element of length d. It is an involution and carries every positive root to
a negative root.
158
22 NIGEL HIGSON
with equality if and only if αI is the sum of the roots that w−1 transforms from
positive to negative, as in Lemma 4.18. We find therefore that the first equation
is solved exactly when αI is this sum and when μ = w(φ). And in this case the
second equation is solved too. Finally αI cannot be expressed in any other way as
a sum of distinct positive weights, so the lemma is proved.
Proof of Theorem 3.10. Let w ∈ Wσ and let p = |w|. Then

|w| + |ww0 | = d.
Using this, (4.18), the Künneth formula (4.17) and the easy part of Kostant’s
theorem, we find that the weight

w(σ + φ + ρ) − ρ, −w(φ + ρ) − ρ

= w(σ + φ + ρ) − ρ, ww0 (−w0 (φ) + ρ) − ρ
occurs as a weight of the cohomology space H d (n ⊕ n, Wσ+φ,φ ). By Lemma 4.17

and Lemma 4.5 the projection of δ(η ⊗ wσ ) onto this weight space is nonzero. We
find that U(h) ⊗ U(h) acts on the image of δ through the weight

w(σ + φ + ρ) − ρ, −w(φ + ρ) − ρ
From the commutative diagram it follows that R(g, σ) acts through the generalized
Harish-Chandra homorphism GHCσ and evaluation at the weight (w(σ + ρ) − ρ).
This being so for all w ∈ Wσ and all dominant integral φ it follows that the range
of GHCσ lies in U(h)Wσ , as required.
5. Minimal K-Types
In the following definition we shall identity finite-dimensional irreducible rep-
resentations of K with their highest weights, and carry over to representations the
given order on weights.
5.1. Definition (Compare [Vog79] or [Vog81]). Let Z be a (g, K)-module.
A finite-dimensional irreducible representation σ of K is a K-type of a (g, K)-
module Z if the σ-isotypical component of Z is nonzero. A minimal K-type is a
K-type that is minimal in the ordering on weights determined by our fixed choice
of positive roots. If σ is a minimal K-type of Z, then we shall call Z a σ-minimal
(g, K)-module.
5.2. Remark. Actually Vogan used a different definition, but for complex
semisimple groups the one above seems more natural (in any case for complex
semisimple groups it is equivalent to Vogan’s definition).
5.3. Example. The representations Wσ+φ,φ = Vσ+φ ⊗ Vφ∗ are all σ-minimal.
5.4. Definition. Call an irreducible R(g, σ)-module minimal if it corresponds

under the bijection of Proposition 2.10 to an irreducible σ-minimal (g, K)-module.
5.5. Definition. Let F : R → S be a homomorphism between rings, and let
M be an R-module. We shall say that M factors through the homomorphism F if
the action of R on M factors through F .
23
5.6. Lemma. Let M be an irreducible R(g, σ)-module. If M factors through the

generalized Harish-Chandra homomorphism
GHCσ : R(g, σ) −→ U(h),
then the irreducible (g, K)-module associated to M contains only K-types τ with
τ ≥ σ. In particular, it is σ-minimal and moreover σ is its unique minimal K-type.
Proof. Let τ be a dominant integral weight and let pτ ∈ R(K) be the corre-
sponding isotypical projection. Define a two-sided ideal Jτ ⊆ R(g, σ) by
Jτ = R(g, σ) ∩ R(g, K)pτ R(g, K).
Let M be an irreducible R(g, σ)-module and let Z be the corresponding irreducible

(g, K)-module. Then
Zτ = 0 ⇔ Jτ · M = 0.
Let φ ∈ h∗ be dominant and integral. The irreducible module associated to the

character
GHC eval. at σ + φ
(5.1) R(g, σ) −−−−→
σ
U(h) −−−−−−−−−→ C
is Wσ+φ,φ , and since Wσ+φ,φ contains only K-types τ ≥ σ, we find that the char-
acter (5.1) maps Jτ to zero unless τ ≥ σ. This being so for all φ, it follows that
the GHCσ maps Jτ to zero unless τ ≥ σ.
Vogan proved the converse, which is much deeper:
5.7. Theorem (Vogan [Vog79]). Every minimal irreducible R(g, σ)-module

factors through the generalized Harish-Chandra homomorphism GHCσ .
Compare [Vog79, Sections 3 and 4]. I shall present Vogan’s proof, which is a
little involved, partly because it is central to my argument and partly because the
complex semisimple case I am considering is so much simpler than the real case
considered by Vogan that it might be a helpful introduction to the general case.
Theorem 5.7 is a consequence of Proposition 4.10, Lemma 4.13 and the following
result:
5.8. Theorem (Vogan). If Z is a σ-minimal (g, K)-module, then the map
δ : H 0 (n, Z)σ ⊗ ∧d n∗1 −→ H d (n ⊕ n, Z)σ−2ρ
is injective.
5.9. Remark. Of course I checked this by hand for Z = Wσ+φ,φ in the previous
section.
Theorem 5.8 is proved by partially calculating H d (n ⊕ n, Z)σ−2ρ using the

Leray-Serre spectral sequence. The Chevalley-Eilenberg complex that computes
160
24 NIGEL HIGSON
H ∗ (n ⊕ n, Z) decomposes into a double complex
(5.2) .. ..
.O .O
∧0 n∗ ⊗ ∧2 n∗1 ⊗ Z / ∧1 n∗ ⊗ ∧2 n∗ ⊗ Z / ···
O O 1
∧0 n∗ ⊗ ∧1 n∗1 ⊗ Z / ∧1 n∗ ⊗ ∧1 n∗ ⊗ Z / ···
O O 1
∧0 n∗ ⊗ ∧0 n∗1 ⊗ Z / ∧1 n∗ ⊗ ∧0 n∗ ⊗ Z / ···
1
as a result of the decomposition (4.11). The rows compute the cohomology spaces
H p (n, ∧q n∗2 ⊗ Z); the exact form of the columns is not important. There are in any
case finitely many rows and columns. We are interested in the top row, so let us
redraw (5.2) as follows:
(5.3) ∧0 n∗ ⊗ ∧d n∗1 ⊗ Z / ∧1 n∗ ⊗ ∧d n∗ ⊗ Z / ···
O O 1
∧0 n∗ ⊗ ∧d−1 n∗1 ⊗ Z / ∧1 n∗ ⊗ ∧d−1 n∗ ⊗ Z / ···

O O 1
∧0 n∗ ⊗ ∧d−2 n∗1 ⊗ Z / ∧1 n∗ ⊗ ∧d−2 n∗ ⊗ Z / ···

O O 1
.. ..
. .
The (σ − 2ρ)-eigenspace
(5.4) H 0 (n, ∧d n∗ ⊗ Z)σ−2ρ = H 0 (n, Z)σ ⊗ ∧d n∗
appears in the top left position of (5.3) after taking the cohomology of the rows.
The map of (5.4) into H d (n ⊕ n, Z) that is implicit from (5.3) is precisely the map
that we are trying to show is injective.
Consider the row-filtration spectral sequence of the double complex (5.2), or
more precisely the (σ − 2ρ)-part of double complex (5.2). This is
E1p,q = H p (n, ∧q n∗1 ⊗ Z)σ−2ρ ⇒ H p+q (n1 ⊕ n2 , Z)σ−2ρ .
The map
H 0 (n, ∧d n∗ ⊗ Z)σ−2ρ −→ H d (n ⊕ n, Z)σ−2ρ
that Theorem 5.8 asserts to be injective factors as the composition
H 0 (n, ∧d n∗1 ⊗Z)σ−2ρ = E10,d → E20,d → · · · → Ed0,d → H d (n⊕n, Z)σ−2ρ .
The last map in the sequence is injective (the double complex has only d rows, so
Ed = E∞ ), while the kernel of the map
0,d
Er0,d −→ Er+1
25
is the image of the level-r differential

dr : Err−1,d−r −→ Er0,d .
This we shall prove to be zero for the simple reason that Err−1,d−r = 0 (and this
will suffice to prove Theorem 5.8):
5.10. Lemma. Let Z be a σ-minimal (g, K)-module. If p + q < d, then
H p (n, ∧q n∗1 ⊗ Z)σ−2ρ = 0.
In other words E1p,q = 0 if p + q < d.
Since the spaces Erp,q for r ≥ 1 are subquotients of E1p,q , the lemma implies
that they too are zero in the range p + q < d, as required.
The lemma is proved by decomposing Z as a direct sum of irreducible irre-
ducible g-representations Vμ and applying Kostant’s theorem to each summand.
The main step is as follows:
5.11. Lemma. Let Vμ be the irreducible finite-dimensional g-module with highest
weight μ and let β be a sum of q-distinct positive roots. Assume that
H p (n, C−β ⊗ Vμ )σ−2ρ = 0.
If p + q < d, then μ < σ.
Proof. Let p, q and β be as above. I need to show that if
(5.5) H p (n, Vμ )σ+β−2ρ = 0
then μ < σ. According to Kostant’s theorem, (5.5) implies that
w(μ + ρ) − ρ = σ + β − 2ρ
for some w ∈ W with |w| = p. Rewrite this as

(5.6) μ + w−1 (2ρ − β) + ρ − w−1 (ρ) = σ + w−1 (σ) − σ .
The second term on the right-hand side is negative, so it can be dropped if (5.6) is
converted into an inequality. Applying Lemma 4.18 with γ = 2ρ − β gives

(5.7) μ < μ + w−1 (2ρ − β) + ρ − w−1 (ρ) ≤ σ,
as required.
Proof of Lemma 5.10. The hypothesis is that Z decomposes as a direct sum

of finite-dimensional irreducible representations Vτ in such a way that there is no
τ < σ. So Lemma 5.11 shows that
H p (n, C−β ⊗ Z)σ−2ρ = 0
for any β that is a sum of q distinct positive roots. But ∧q n∗1 is assembled as an
n-representation from a sequence of extensions
0 −→ Xk−1 −→ Xk −→ C−β −→ 0
for varying β, where X0 is some C−β too. So the lemma can be proved using the
long exact sequence in cohomology and induction.
162
26 NIGEL HIGSON
6. Cartan Motion Group

Let G be a Lie group and let K be a Lie subgroup. The contraction of G to K
is the semidirect product Lie group
Gc = K (g/k)
associated to the adjoint action of K on the quotient vector space g/k. The Lie
algebra of Gc is of course the semidirect product
gc = k (g/k).
Its enveloping algebra may be obtained from the enveloping algebra of g by an
associated graded construction. Assign to U(gC ) the increasing algebra filtration
for which elements have the maximal order subject to
X ∈ kC ⇒ order(X) = 0
Y ∈ gC ⇒ order(X) ≤ 1.
Then there is a unique algebra isomorphism
∼
=
(6.1) ass. gr. U(gC ) −→ U(gcC )
that is the identity on U(kC ) in degree zero on the left-hand side and is the projection
from gC onto gC /kC in degree one.
Assume now that K is compact. There is a similar filtration on the Hecke
algebra R(g, K) (in which all of R(K) is assigned order zero) and a similar algebra
isomorphism
∼
=
(6.2) ass. gr. R(g, K) −→ R(gc , K).
In terms of the presentation (2.5) of the Hecke algebra it is the tensor product
of (6.1) with the identity on R(K). The isomorphism restricts to spherical Hecke
algebras, giving
∼
=
ass. gr. R(g, σ) −→ R(gc , σ).
For instance in the complex semisimple case that I shall restrict to from now on,
where K is a maximal compact subgroup and Gc is the Cartan motion group asso-
ciated to G, this amounts to
K ∼
=
Sym(g) ⊗ End(Vσ )op −→ R(gc , σ).
6.1. Remark. On the left-hand side above is Kirillov’s classical family algebra
from [Kir00].
The generalized Harish-Chandra homomorphism
is filtration-preserving for the standard filtration on the range, as is clear from the
formula (3.11). This makes possible the following definition.
6.2. Definition. Denote by
GHCσ,c : R(gc , σ) −→ Sym(h)
the associated graded counterpart of the generalized Harish-Chandra homomor-
phism GHCσ from Definition 3.9.
27
The steps taken in Sections 4 and 5 to analyze GHCσ can be repeated for GHCσ,c ,
mostly by taking the associated graded versions of the constructions there. Here
are some details.
The complexified Lie algebra of Gc is isomorphic to the semidirect product
g0 g, where g0 denotes the vector space underlying the complex Lie algebra g,
endowed with the trivial bracket. The enveloping algebra is of course
U(g0 g) = U(g0 ) U(g).
Define a homomorphism
K
U(g0 ) U(g) −→ U(h0 ) ⊗ U(h)
by restricting to K-invariants the projection

K
U(g0 ) U(g) −→ U(h0 )⊗U(n−0 ) ⊗ U(h)⊗U(n− )
essentially as in (4.7). The associated graded counterpart of the map in Defini-

tion 4.9 is obtained from this by composing with the map
I ⊗ Shift2ρ : U(h0 ) ⊗ U(h) −→ U(h0 ) ⊗ U(h).
The proof of Vogan’s Casselman-Osborne theorem carries over to gc . The assoc-

iated-graded counterpart of the diagram (4.13) certainly commutes since the orig-
inal diagram does.6 The associated-graded counterpart of the map Π in Defini-
tion 4.14, which is given by
(H0 , H) → H0 + (σ − 2ρ)(H)I
is of course still an isomorphism. Finally Lemma 5.6 implies its associated-graded

counterpart, and the spectral sequence proof of Theorem 5.8 carries over without
essential change to the associated graded case.
We arrive at the following result about the Cartan motion group Gc , which
exactly mirrors what happens for the group G itself.
6.3. Theorem. Let Gc be the Cartan motion group associated to a connected

complex semisimple group G with maximal compact subgroup K.
(a) Every irreducible (gc , K)-module has a unique minimal K-type.
(b) The irreducible (gc , K)-modules with minimal K-type σ correspond, via Propo-
sition 2.10, to the irreducible R(gc , K) modules that factor through the gener-
alized Harish-Chandra homomorphism GHCσ,c .
6.4. Remark. The approach just given to the representation theory of Gc

through Lie algebra cohomology would not be efficient except for the fact that the
cohomological methods simultaneously deal with the more complicated case of G
itself. Rader [Rad88] takes a more direct approach using a generalization of the
Chevalley restriction theorem that we shall invoke in the next section. He analyzes
not the image of the generalized Harish-Chandra homomorphism, as we shall below,
but the complete algebra R(gc , σ).
6Actually the commutativity of the associated graded diagram is easier.

164
28 NIGEL HIGSON
7. Range of the Harish-Chandra Homomorphism

For brevity denote by
(7.1) Shiftρ : Sym(h) −→ U(h)
the composition of the canonical (symmetrization) isomorphism from Sym(h) to
U(h) with the shift automorphism of U(h) given by H → H + ρ(H)I. It intertwines
the natural W -action on Sym(h) with the ρ-shifted action on U(h). I shall prove
the following result:
7.1. Theorem. The ranges of the Harish-Chandra homomorphisms
GHCσ,c : R(gc , σ) −→ Sym(h)
and
R(g, σ) −→ U(h)
GHCσ :
are the algebras of Wσ -invariants in Sym(h) and U(h) under the natural and ρ-
shifted actions, respectively. In particular the ranges identify with one another
under the shift isomorphism (7.1).
Since the map GHCσ,c is the associated graded of GHCσ , it will suffice to show
that the range of the latter is U(h)Wσ . Note that by Theorem 3.10 we already know
that the range of GHCσ is contained within U(h)Wσ .
By Proposition 4.16, the range of GHCσ is the same as the range of the com-
position
K GHCσ / Π / U(h).
(7.2) U(g) ⊗ U(g) U(h) ⊗ U(h) σ
Consider the related composition

K / Sym(h) ⊗ Sym(h) / Sym(h),
(7.3) Sym(g) ⊗ Sym(g)
in which the left-hand map is the tensor product of two copies of the projection
Sym(g) −→ Sym(h)
associated to the decomposition g ∼ = n− ⊕ h ⊕ n+ and the right-hand map is induced
from
(H1 , H2 ) → H1 − H2 + σ(H2 )I.
The range of (7.3) lies within the Wσ -invariants in Sym(h).
Consider also the diagram
K / U(h)
(7.4) U(g) ⊗ U(g) O
O
Shiftρ
K / Sym(h)
Sym(g) ⊗ Sym(g)
in which the horizontal maps are (7.2) and (7.3), the left vertical map is symmetriza-
tion in each factor, and the right vertical map is the shift isomorphism (7.1). The
diagram does not commute, but all the algebras in it are filtered in the standard
way, all the maps are filtration preserving, and the diagram commutes to leading
filtration order. So to prove that the range of (7.2) is all of U(h)Wσ , and hence
prove Theorem 7.1, it suffices to prove the following result:
29
7.2. Theorem. The range of the homomorphism

K
Sym(g) ⊗ Sym(g) −→ Sym(h)
in (7.3) and (7.4) is equal to the algebra of Wσ -invariants in Sym(h) for the natural
action of W .
This is a modest extension of the Chevalley restriction theorem, and I shall
prove it by modestly extending the standard proof of that theorem.
First it is helpful to change variables. Taking into account the K-equivariant
automorphism of Sym(g) ⊗ Sym(g) given by the map (X, Y ) → (X + Y, Y ), it
suffices to prove the surjectivity of the map
K / [Sym(h) ⊗ Sym(h)]W / Sym(h)Wσ ,
(7.5) Sym(g) ⊗ Sym(g)
in which the left-hand map is as before but the right-hand map is induced from
(H1 , H2 ) → H1 + σ(H2 )I.
Equip g with an invariant and nondegenerate quadratic form and fix dual bases
{Xj } and {X j } for g with respect to the form. If πφ is the finite-dimensional
representation of g with highest weight φ, then for every n the quantity

(7.6) Trace πφ (X j1 · · · X jn ) Xj1 · · · Xjn ⊗ I
is a K-invariant element in Sym(g) ⊗ Sym(g) (the sum is over all possible indices).
Under (7.5) the element (7.6) maps to

(7.7) Trace πφ (H j1 · · · H jn ) Hj1 · · · Hjn ,
where {Hj } and {H j } are now dual bases for h. Computing the trace with respect
to a basis of weight vectors for Vφ gives

n
Hj1 · · · Hjn Trace πφ (H j1 · · · H jn ) = Hμ ,
where the right-hand sum is over the weights μ of Vφ and Hμ ∈ h is the vector
corresponding to μ ∈ h∗ under the quadratic form. Taking linear combinations of
these elements as the dominant integral weight φ varies gives the elements

n
(7.8) Hw(φ) .
w∈W
The elements Hφn span the degree n component of Sym(h), and so the elements
(7.8) span the degree n component of Sym(h)W . As a result, the range of (7.5) at
least contains the algebra of W -invariants. For all this see [KV95, Sec. IV.7].
To prove surjectivity onto the Wσ -invariants, not just the W -invariants, con-
sider the following (K-equivariant) derivative operator on Sym(g) ⊗ Sym(g):

(7.9) D(S ⊗ T ) = ∂ηi S ⊗ Xi T.

Here {ηi } is the basis of g∗ dual to {Xi }, and ∂ηi is the directional derivative in the
ηi -direction (think of S as a polynomial on g∗ ). Under the map (7.5) the operator
D corresponds to the action of the directional derivative

(7.10) ∂μ S = σ(Hi )∂ηi S

on Sym(h) (the sum is over a basis for h and dual basis for h∗ ).
166
30 NIGEL HIGSON
Applying (7.9) and (7.10) to the elements of (7.6) and (7.7) respectively, but
with n + k in place of n, we find that the element

n+k
∂σk Hw(φ)
w∈W
lies in the range of (7.5). But now

n σ(Hw(φ) ) · Hw(φ)
n+k n+k
∂σk Hw(φ) = k n
w∈W
and for a Zariski-dense set of φ the coefficients

σ(Hw(φ) )k = w(φ), σk
associated to Weyl group elements w in distinct Wσ -cosets are distinct. It follows
that the elements
n
Hw(φ)
w∈Wσ
lie in the range, and these span the degree n component of Sym(h)Wσ , as required.
This completes the proof of the surjectivity of (7.5), and hence of Theorem 7.2.
8. Mackey Analogy
It is time to put everything together. Let G be a connected complex semisimple
group and let K be a maximal compact subgroup of G. Form the Hecke algebra
R(g, K) and consider the problem of classifying its irreducible modules.
Linearly order7 the irreducible representations of K in such a way that if τ
precedes σ in the usual weight ordering, then τ also precedes σ in the linear ordering.
Let
Pn = pσ1 + · · · + pσn ∈ R(K)
be the sum of the first n isotypical projections. If we set
Rn = Pn R(g, K)Pn ,
then the algebra R(g, K) is the increasing union of the algebras Rn .
The irreducible modules of Rn correspond to the irreducible modules Z of
R(g, K) for which Pn Z = 0. That is, the irreducible modules of Rn are in bijec-
tion with the irreducible (g, K)-modules that contain at least one of the K-types
σ1 , . . . , σn (see the proof of Proposition 2.10).
Consider now the ideal
Jn = Rn Pn−1 Rn
and the quotient algebra
Qn = Rn /Jn .
Notice that

Jn = Rn Pn−1 Pn−1 Rn and Rn−1 = Pn−1 Rn Rn Pn−1 ,
and this defines a sort of Morita equivalence between the ideal Jn and the unital
algebra Rn−1 . So Rn is assembled by extensions and Morita equivalences from the
quotient algebras Qn .
7This linear ordering step is taken for simplicity only—it makes the subsequent constructions
a bit easier to follow. For the more complicated case of real groups it would be necessary to stay
with a partial ordering.
31
The irreducible modules of Qn are of course precisely the irreducible modules

of Rn in which Pn−1 acts as the zero operator. They correspond to the irreducible
(g, K)-modules that include the K-type σn but none of the K-types σ1 , . . . , σn−1 .
According to Lemma 5.6 and Theorem 5.7 these are precisely the (g, K)-modules
with minimal K-type σn .
Notice that Qn = pσn Qn pσn and so by Lemma 2.11
∼ Sn ⊗ End(Vσ ),
Qn = n
where Sn is the K-fixed part of Qn . So Rn is assembled by extensions and (still

more) Morita equivalences from the quotient algebras Sn .
The algebra Sn is the quotient of the spherical Hecke algebra R(g, σn ) by the
intersection of R(g, σn ) with the ideal in R(g, K) generated by Pn−1 . The general-
ized Harish-Chandra homomorphism GHCσ factors as a composition of surjections
R(g, σn ) / Sn / U(h)Wσ .
By Vogan’s theorem, every irreducible Sn -module factors through the second map.
In the C ∗ -algebra context considered in [Hig08] the counterpart of the second map
was therefore an isomorphism, since it induced an isomorphism on dual spaces. But
here the map will typically have a kernel, namely the Jacobson radical of Sn .
In any case, the admissible dual of G (that is, the space of irreducible (g, K)-
modules up to equivalence) decomposes as a disjoint union of the duals of the
algebras Qn or Sn (they are the same), and these consituent pieces may be con-
cretely computed as the ranges of the generalized Harish-Chandra homomorphisms
GHCσn , giving the answer h∗ /Wσn , as we have seen.
The entire argument applies simultaneously to G and to its Cartan motion
group Gc . In fact the two can be wrapped into one8 by forming the usual Rees
algebra associated to the filtration on R(g, K). Recall this is the algebra of poly-
nomials
a 0 + a1 z + · · · + an z n
with coefficients aj ∈ R(g, K) such that the order of aj is less than or equal to j.
Evaluation at a nonzero value λ ∈ C (that is, dividing by the ideal generated by
z −λ) yields R(g, K), whereas evaluation at zero yields R(gc , K). This deformation
is assembled from the analogous deformations associated to the algebras Qn , and
these in turn are trivial, up to their Jacobson radicals.
9. Concluding Remarks
There are two main parts of the argument just presented: (a) establishing the
connection between the generalized Harish-Chandra homomorphism GHCσ and σ-
minimal K-type representations; and (b) computing the range of the generalized
Harish-Chandra homomorphism. While they are both accessible to more or less
standard representation-theoretic techniques, it would be interesting to understand
them from a more geometric point of view. To my mind the connection, pointed out
by Connes, between the Mackey analogy and K-theory via the Connes-Kasparov
conjecture adds emphasis to this.
The contraction Gc of a Lie group G to a Lie subgroup K has the following
property: if G acts on a manifold V and the subgroup K globally fixes a submanifold
8Admittedly this is for show more than anything else, at least as things stand at present.
168
32 NIGEL HIGSON
M , then the contraction naturally acts on the normal bundle of M in V (via

transformations that are fiberwise affine, not linear).
In the present context of complex semisimple groups it is natural to consider
the flag variety for gC , which is a product of two copies of the flag variety X ∼
= G/B
for g (here B is a Borel subgroup of G). The group K acts diagonally on X × X
and of course it fixes the diagonal globally. So the Cartan motion group acts on the
normal bundle (which is of course the tangent bundle of X). It is therefore natural
to attempt to analyze deformation from Gc to G in terms of the deformation to the
normal cone construction associated to the embedding of the diagonal into X × X
(in index theory this is also called the tangent groupoid). It is easy to guess that
part (a) above ought to be reinterpreted as a description of how minimal K-type
representations (thought of say as D-modules) restrict to the diagonal and then
push forward again to X × X, while part (b) ought to be some sort of rigidity
principle for the infinitesimal neighborhood of the diagonal in X × X.
Another interesting view on part (b) comes from the work of Alekseev and
Meinrenken [AM05], who gave a remarkable proof, using equivariant cohomology,
of the Duflo isomorphism theorem for quadratic Lie algebras (meaning Lie algebras
with an invariant, nondegenerate quadratic form). Consider the diagram
[Sym(g) ⊗ End(Vσ )op ]K / [Sym(g) ⊗ End(Vσ )op ]K
GHCσ,c GHCσ
∼
Sym(h)
= / U(h)
Shiftρ
relating the generalized Harish-Chandra homomorphisms GHCσ and GHCσ,c . The

dotted arrow hasn’t been defined, but for σ = 0 the Duflo-Kirillov isomorphism can
be inserted there and the diagram commutes. It obviously follows that the images
of GHCσ and GHCσ,c are equal.
The simple, cohomological proof of this given by Alekseev and Meinrenken
offers hope that a generalization to σ = 0 may be found which would directly relate
the two generalized Harish-Chandra homomorphisms.
It should be noted, however, that rather than the dotted arrow being an algebra
homomorphism, the best one can hope for in general is a vector space isomorphism.
Moreover the explicit Duflo-Kirillov formula in the case σ = 0 leads quickly to
an explicit formula for spherical matrix coefficient functions on G, and since the
corresponding functions for σ = 0 look more complicated there are limits to how
simple and explicit the dotted arrow could be in general.
Finally, a comment about the case of real groups. Vogan treated these in
[Vog79]. The real case is much more complicated, but there is no reason in principle
why the Mackey analogy could not be developed for real groups along the lines given
for complex groups here. (I thank David for patiently explaining his work during
numerous discussions with me.) In the real case it is possible for a single irreducible
(g, K)-module to have several minimal K-types, and because of this it is probably
most natural to adjust the definition of the Harish-Chandra homomorphism so as
to treat an entire associate class of minimal K-types at once. The range will then
be a noncommutative (twisted) crossed product algebra involving the R-groups of
Vogan, Knapp and Zuckermann.
33
References
[AM05] A. Alekseev and E. Meinrenken. Lie theory and the Chern-Weil homomorphism. Ann.
Sci. École Norm. Sup. (4), 38(2):303–338, 2005.
[BCH94] P. Baum, A. Connes, and N. Higson. Classifying space for proper actions and K-theory
of group C ∗ -algebras. In C ∗ -algebras: 1943–1993 (San Antonio, TX, 1993), volume 167
of Contemp. Math., pages 240–291. Amer. Math. Soc., Providence, RI, 1994.
[Ber62] F. A. Berezin. Laplace operators on semisimple Lie groups. Amer. Math. Soc. Transl.
(2), 21:239–339, 1962.
[BG80] J. N. Bernstein and S. I. Gel fand. Tensor products of finite- and infinite-dimensional
representations of semisimple Lie algebras. Compositio Math., 41(2):245–285, 1980.
[CE56] H. Cartan and S. Eilenberg. Homological algebra. Princeton University Press, Princeton,
N. J., 1956.
[CO75] W. Casselman and M. S. Osborne. The n-cohomology of representations with an infini-
tesimal character. Compositio Math., 31(2):219–227, 1975.
[Dix96] J. Dixmier. Enveloping algebras, volume 11 of Graduate Studies in Mathematics. Ameri-
can Mathematical Society, Providence, RI, 1996. Revised reprint of the 1977 translation.
[Duf75] M. Duflo. Représentations irréductibles des groupes semi-simples complexes. In Analyse
harmonique sur les groupes de Lie (Sém., Nancy-Strasbourg, 1973–75), pages 26–88.
Lecture Notes in Math., Vol. 497. Springer, Berlin, 1975.
[Enr81] T. J. Enright. Lectures on representations of complex semisimple Lie groups, volume 66
of Tata Institute of Fundamental Research Lectures on Mathematics and Physics. Pub-
lished for the Tata Institute of Fundamental Research, Bombay, 1981. Lecture notes by
Vyjayanthi Sundar.
[Fla79] D. Flath. Decomposition of representations into tensor products. In Automorphic forms,
representations and L-functions (Proc. Sympos. Pure Math., Oregon State Univ., Cor-
vallis, Ore., 1977), Part 1, Proc. Sympos. Pure Math., XXXIII, pages 179–183. Amer.
Math. Soc., Providence, R.I., 1979.
[Geo09] C.Y. George. The Mackey analogy for SL(n, R). PhD thesis, The Pennsylvania State
University, 2009.
[God52] R. Godement. A theory of spherical functions. I. Trans. Amer. Math. Soc., 73:496–556,
1952.
[HC51] Harish-Chandra. On some applications of the universal enveloping algebra of a semisim-
ple Lie algebra. Trans. Amer. Math. Soc., 70:28–96, 1951.
[Hig08] N. Higson. The Mackey analogy and K-theory. In Group Representations, Ergodic The-
ory, and Mathematical Physics: A Tribute to George W. Mackey, volume 449 of Con-
temp. Math., pages 149–172. Amer. Math. Soc., Providence, RI, 2008.
[IW53] E. Inonu and E. P. Wigner. On the contraction of groups and their representations. Proc.
Nat. Acad. Sci. U. S. A., 39:510–524, 1953.
[Jac64] N. Jacobson. Structure of rings. American Mathematical Society Colloquium Publica-
tions, Vol. 37. Revised edition. American Mathematical Society, Providence, R.I., 1964.
[Kir00] A. A. Kirillov. Family algebras. Electron. Res. Announc. Amer. Math. Soc., 6:7–20
(electronic), 2000.
[Kir01] A. A. Kirillov. Introduction to family algebras. Mosc. Math. J., 1(1):49–63, 2001.
[Kos61] B. Kostant. Lie algebra cohomology and the generalized Borel-Weil theorem. Ann. of
Math. (2), 74:329–387, 1961.
[KV95] A. W. Knapp and D. A. Vogan, Jr. Cohomological induction and unitary representations,
volume 45 of Princeton Mathematical Series. Princeton University Press, Princeton, NJ,
1995.
[Mac75] G. W. Mackey. On the analogy between semisimple Lie groups and certain related semi-
direct product groups. In Lie groups and their representations (Proc. Summer School,
Bolyai János Math. Soc., Budapest, 1971), pages 339–363. Halsted, New York, 1975.
[PP83] M. G. Penington and R. J. Plymen. The Dirac operator and the principal series for
complex semisimple Lie groups. J. Funct. Anal., 53(3):269–286, 1983.
[Rad88] C. Rader. Spherical functions on Cartan motion groups. Trans. Amer. Math. Soc.,
310(1):1–45, 1988.
170
34 NIGEL HIGSON
[Ser01] J.-P. Serre. Complex semisimple Lie algebras. Springer Monographs in Mathematics.
Springer-Verlag, Berlin, 2001. Translated from the French by G. A. Jones, Reprint of
the 1987 edition.
[Sku10] J.R. Skukalek. On the Higson-Mackey analogy, group C ∗ -algebras and K-theory. PhD
thesis, The Pennsylvania State University, 2010.
[Vog79] D. A. Vogan, Jr. The algebraic structure of the representation of semisimple Lie groups.
I. Ann. of Math. (2), 109(1):1–60, 1979.
[Vog81] D. A. Vogan, Jr. Representations of real reductive Lie groups, volume 15 of Progress in
Mathematics. Birkhäuser Boston, Mass., 1981.
[War72] G. Warner. Harmonic analysis on semi-simple Lie groups. I. Springer-Verlag, New York,
1972. Die Grundlehren der mathematischen Wissenschaften, Band 188.
[Zhe74] D. P. Zhelobenko. Garmonicheskii analiz na poluprostykh kompleksnykh gruppakh Li.
Izdat. “Nauka”, Moscow, 1974. Sovremennye Problemy Matematiki. [Current Problems
in Mathematics].
Pennsylvania State University, University Park, PA 16802, USA

E-mail address: higson@math.psu.edu
Contemporary
Mathematics
Volume 546, 2011
A survey on Hopf-cyclic cohomology and Connes-Moscovici

characteristic map
Atabey Kaygun
To Henri with admiration and gratitude.
Abstract. In 1998 Alain Connes and Henri Moscovici invented a cohomology theory for
Hopf algebras and a characteristic map associated with the cohomology theory in order
to solve a specific technical problem in transverse index theory. In the following decade,
the cohomology theory they invented developed on its own under the name Hopf-cyclic
cohomology. But the history of Hopf-cyclic cohomology and the characteristic map they
invented remained intricately linked. In this survey article, we give an account of the
development of the characteristic map and Hopf-cyclic cohomology.
Introduction
One can claim that following its inception in the seminal article [10] by Alain Connes
and Henri Moscovici, Hopf cyclic cohomology and the characteristic map associated with
it have since become a standard cohomological tool in noncommutative geometry. The
particular cohomology theory they defined has proved itself to be a robust analogue of
the Gelfand-Fuchs cohomology complete with an appropriate analogue of the classical
characteristic map.
In this survey article, we are going to trace the origins of the theory and investigate the
developments in the subject in an attempt to give a brief but focused account of the past
and the current research. Here is the plan of this article: In the first section, we will detail
the main aspects of [10] focusing mainly on the construction of Hopf-cyclic cohomology
and the characteristic map as defined by Connes and Moscovici in op. cit. In the second
section we detail the busy aftermath of op. cit. until the introduction of SAYD modules into
the theory by Hajac, Khalkhali, Rangipour and Sommerhäuser in their ground-breaking
work [23]. The third section is a study of [23] and [24], and their generalizations. In
the fourth section we take a quick detour to the dual Hopf-cyclic cohomology as defined
by Khalkhali and Rangipour in [34] which is needed for the cup product interpretation
of the Connes-Moscovici characteristic map. In the same section, we also list the main
2010 Mathematics Subject Classification. Primary 16E40; Secondary 16T05, 19D53, 55B34.
2011
c 0000
c (copyright
American Mathematical holder)
Society
1
171
2172 ATABEY KAYGUN
calculations made in Hopf-cyclic cohomology (and its dual) for a number of relevant and
important Hopf algebras. In the fifth section, starting from [37] we investigate all of the
major cup product interpretations of Connes-Moscovici characteristic map. In the same
section, we also explain how these different interpretations were seemingly incompatible
and how this problem was finally resolved by the author in [30].
1. The beginnings
If we were to summarize succinctly why Hopf-cyclic cohomology and the character-
istic map associated with it were invented, we can simply say for computing the index of a
transversally elliptic operator on a foliation. For a foliated manifold (V, F) and a transver-
sally elliptic operator D on V we have the index pairing yielding
Index(D) = ch(D), ch(E)
via the Chern-Connes character for any E ∈ K(V/F) [4, 6]. In [9] Connes and Moscovici
showed that ch(D) reduces to a finite sum of expressions of the form

(1.1) a0 [D, a1 ](k1 ) · · · [D, am ](km ) |D|−(m+2k1 +···+2km )

where [D, a](k) denotes the k-th iterated commutator of D2 with [D, a] and is a Diximier
trace or a Wodzicki residue. See also [8]. It may seem that the formula is computable by
virtue of being local, but in [10] Connes and Moscovici observe that
. . . although the general index formula easily reduces to the local form
of the Atiyah-Singer index theorem when D is say a Dirac operator on a
manifold, the actual explicit computation of all the terms (1.1) involved
in the cocycle ch(D) is a rather formidable task. As an instance of this
let us mention that even in the case of codimension one foliations, the
printed form of the explicit computation of the cocycle takes around
one hundred pages. Each step in the computation is straightforward
but the explicit computation for higher values of n is clearly impossible
without a new organizing principle which allows [one] to bypass them.
The organizing principle for codimension n foliations, it turns out, is a particular Hopf
algebra denoted by Hn . Abstractly, the setup involves a unital algebra A on which Hn acts
compatibly as
(1.2) h(a · b) = h(1) (a) · h(2) (b)
and an algebra map δ : Hn → C which satisfies
(1.3) h = δ(h(1) S (h(3) ))S 2 (h(2) )
and finally a trace map τ : A → C which satisfies
(1.4) τ(h(a)) = δ(h)τ(a)
for any a, b ∈ A and h ∈ Hn . Then one can account for all of the terms (1.1) in the local
index formula by considering terms of the form
(1.5) τ(a0 · h1 (a1 ) · · · hm (am ))
A SURVEY ON HOPF-CYCLIC COHOMOLOGY AND CONNES-MOSCOVICI CHARACTERISTIC MAP 1733
for h1 , . . . , hm ∈ Hn in this setup. Then we consider the standard cyclic module of an

algebra A and rewrite (1.5) as a map
(1.6) τ : Hn⊗m −→ HomC (A⊗m+1 , C)
by using the compatible action (1.2) and the invariance property (1.4) for every m ≥ 0.
⊗m
This yields a cocyclic structure on the graded module m≥0 Hn such that τ becomes a
morphism of cocyclic modules. Explicitly, the cocyclic structure on the graded module
Hn := ⊕m≥0 Hn⊗m is defined by
⎧
⎪
⎪
⎪(1 ⊗ h1 ⊗ · · · ⊗ hm ) if i = 0
⎪
⎪
⎨
∂i (h1 ⊗ · · · ⊗ hm ) = ⎪
⎪
(1) (2)
(h1 ⊗ · · · ⊗ hi ⊗ hi ⊗ · · · ⊗ hm ) if 0 < i ≤ m
⎪
⎪
⎪
⎩(h ⊗ · · · ⊗ h ⊗ 1) if i = m + 1
1 m
σ j (h1 ⊗ · · · ⊗ hm ) =(hi+1 )(h1 ⊗ · · · ⊗ hi ⊗ hi+2 ⊗ · · · ⊗ hm )

τm (h1 ⊗ · · · ⊗ hm ) =S (h(m+1)
1 )h2 ⊗ · · · ⊗ S (h(3) (2) (1)
1 )hm ⊗ S (h1 )δ(h1 )
The effect of the morphism (1.6) in periodic cyclic cohomology
HP∗ (τ) : HP∗ (Hn ) → HP∗ (A)

is the Connes-Moscovici characteristic map.
We must warn the reader that the short description we gave above does little justice to
the intricacies of the subject since the technical details of the subject as developed in [10]
are rather complicated, but still illuminating.
In [10] Connes and Moscovici also investigate in detail the Hopf algebra structure of
the class of Hopf algebras Hn and prove that the periodic cyclic cohomology of H1 is
essentially the Gelfand-Fuchs cohomology of the Lie algebra of formal vector fields on
R. They show that the primary characteristic classes of codimension-1 foliations, namely
the transverse fundamental class and the Godbillion-Vey class, do lie in the image of their
characteristic map. More importantly, they show that H1 has only two non-trivial periodic
classes based on their identification with the Gelfand-Fuchs cohomology, and they are sent
to the transverse fundamental class and the Godbillion-Vey class under the characteristic
map. This fact indicates, at least cohomologically, that H1 is a universal algebraic object
classifying codimension-1 foliations.
At this point, let us give an explicit description of H1 in terms of generators and
relations before we describe the characteristic classes Connes and Moscovici expressed in
terms of Hopf-cyclic cocycles. As a noncommutative algebra H1 is countably generated
by symbols X, Y and δm for m ≥ 1. The relations are
(1.7) [Y, X] = X, [Y, δm ] = mδm , [X, δm ] = δm+1 , [δm , δ ] = 0
for every , m ≥ 1. The coproduct on the generators are given by
Δ(Y) = (1 ⊗ Y) + (Y ⊗ 1), Δ(δ1 ) =(1 ⊗ δ1 ) + (δ1 ⊗ 1),
Δ(X) = (1 ⊗ X) + (X ⊗ 1) + (δ1 ⊗ Y)
One can write explicit formulas for Δ(δm ) using the fact that Δ is multiplicative and the
commutator relation [X, δm ] = Xδm − δm X = δm+1 for every m ≥ 1. In order to complete
4174 ATABEY KAYGUN
the setup we need a character δ : H1 → C, which is defined on the generators by
(1.8) δ(Y) = 1, δ(X) = 0, δ(δm ) = 0 for every m ≥ 1
Now, we are ready to determine the primary characteristic classes: the transverse funda-
mental class is represented by the cyclic 2-cocycle
(1.9) (X ⊗ Y) − (Y ⊗ X) − (δ1 Y ⊗ Y)
and the Godbillion-Vey class by the 1-cocycle δ1 .

The Hopf algebra H1 and its higher dimensional analogues Hn are deformations of the
Lie algebra of the group of affine transformations of Rn for n ≥ 1. We note that this class
of Hopf algebras is different than previously defined deformations of Lie algebras/groups
such as quantum groups. The full generalization of deformations of this type for arbitrary
primitive Lie pseudo-groups is given by Moscovici and Rangipour in [42], with applica-
tions in foliations with certain transverse symmetries in [43].
2. Early works
Besides the original framework Connes has developed for cyclic cohomology [5]
which uses cyclic invariant Hochschild cocycles, there are other computational paradigms
such as the Cuntz-Quillen framework of X-complexes [18]. An early development in the
subject came from Crainic in the preprint [16] which was later published as [17] where
he reframed Connes and Moscovici’s cohomology theory within the Cuntz-Quillen frame-
work for arbitrary Hopf algebras provided that the invariant character δ satisfies (1.3). He
also found an analogue of Bott’s characteristic map [2] k : H ∗ (WOq ) → H ∗ (V/F) of a fo-
liated manifold (V, F) of codimension q by developing a noncommutative analogue W(H)
of the Weil algebra WOq for an arbitrary Hopf algebra H. This approach was later gen-
eralized by Sharygin [46], and Nikonov and Sharygin [44] utilizing the full generality of
coefficients in stable anti-Yetter-Drinfeld modules [24].
The similarly highly geometric approach also forms the basis of Gorokhovsky’s re-
markable preprint [19] which was later published as [20]. In this paper Gorokhovsky ex-
tends to scope of the Hopf-cyclic theory to differential graded Hopf algebras, but his main
goal was to answer the following simple question. Can one obtain the secondary charac-
teristic classes of foliations from non-periodic Hopf-cyclic cohomology similar to the way
Connes and Moscovici obtained the Godbillion-Vey class and the transverse fundamen-
tal class from periodic Hopf-cyclic cohomology? In that paper, Gorokhovsky provides an
explicit affirmative answer for a certain class of foliated manifolds. Later Kaminker and
Xiang in [27] considered these secondary classes for Riemannian-foliated flat bundles by
using an extension of the cohomology theory for Hopf algebroids similar to [12] and [36].
On the other hand, the first attempt at computing non-periodic cohomology classes
of the Connes-Moscovici Hopf algebra H1 the writer is aware of is Antal’s unpublished
thesis [1]. There Antal shows that the first non-periodic Hopf-cyclic cohomology of H1 is
generated by the Godbillion-Vey class δ1 and the class δ1 called the Schwarzian which is
defined as
1
(2.1) δ1 = δ2 − δ21
2
The arduous task of computing all of the non-periodic classes of the Connes-Moscovici
Hopf algebra H1 was completed by Moscovici and Rangipour in [42] based on methods
they developed in [41].
3. Stable anti-Yetter-Drinfeld modules

In [11] Connes and Moscovici defined our next important object: modular pairs in
involution. A modular pair in involution in a Hopf algebra H is a pair (σ, δ), where σ ∈ H
is a group-like element and δ : H → C is a character such that
(3.1) δ(σ) = 1 and δ(h(1) S (h(3) ))S 2 (h(2) ) = σhσ−1
They show that the cocyclic module they defined for the class of Hopf algebras Hn can
be defined for an arbitrary Hopf algebra H if the Hopf algebra H has a modular pair in
involution. The technical condition (3.1) is needed to make sure that the cyclic identity

τmm = id is satisfied for any m ≥ 1 in the cocyclic module H they similarly defined.
But one of the most exciting developments in the theory of Hopf-cyclic cohomology
came from Hajac, Khalkhali, Rangipour and Sommerhäuser [24, 23]. They made the im-
portant discovery that one has to consider H as a H-module coalgebra over itself when
we consider Hopf-cyclic cohomology of H. The second fundamental discovery they made
is that a modular pair in involution (σ, δ) in a Hopf algebra H is really a 1-dimensional
H-module/comodule σ Cδ = C with comodule structure λ : σ Cδ → H ⊗ σ Cδ defined by
λ(1) = σ ⊗ 1 and with module structure ρ : H ⊗ σ Cδ → σ Cδ defined by h · 1 = δ(h)1
for any h ∈ H. Then the Hopf-cocyclic module of H is obtained from the ordinary co-
cyclic module of H viewed as a coalgebra which is twisted by the H-module/comodule
σ
Cδ . In [23] the authors determine what specific conditions are required for a higher di-
mensional H-module/comodule M to yield cocyclic modules for arbitrary Hopf algebras.
These conditions are
(3.2) m(−1) m(0) =m and
(hm) (−1)
⊗ (hm) (0)
=h m(−1) S −1 (h(3) ) = h(2) m(0)
(1)
for any m ∈ M and h ∈ H. Such modules are called stable anti-Yetter-Drinfeld (SAYD)
modules [24].
Following [23], in [28] Kaygun showed that if one trivializes the diagonal action of the
underlying Hopf algebra H on the cocyclic module of H viewed as a module coalgebra
over itself twisted by a SAYD module M then one obtains the cocyclic module of H as
defined in [23]. If the coefficient module is not SAYD, the quotient need not be a cocyclic
module. But one can always trivialize the diagonal action and force the quotient to be a
cocyclic module at the same time. The immediate consequence of this approach was to
extension of the theory to bialgebras and arbitrary coefficient modules [28].
4. The dual theory and cyclic duality

The discovery of Hopf-cyclic cohomology by Connes and Moscovici came also with
the identification of the cohomology with Gelfand-Fuchs cohomology of formal vector
fields over R for H1 [10]. There are two other important non-trivial computations of
Hopf-cyclic cohomology: for the quantum group Uq (sl2 ) by Kusterman, Rognes and Tuset
6176 ATABEY KAYGUN
in [40] and for the quantum group A(S Lq (2)) by Hadfield and Krähmer in [21, 22]. Then
in [17] Crainic showed that the Hopf-cyclic cohomology of the group ring C[G] with co-
efficients in σ Cδ for a modular pair in involution (σ, δ) is Cotor∗H (C1 , Cσ ), which is trivial.
Here we use Cα to denote the comodule 1 → α ⊗ 1 or 1 → 1 ⊗ α, whichever is appro-
priate for any α ∈ G. This result seems to contradict the appropriate analogue of the Van
Est isomorphism [50] when we consider the identification by Connes and Moscovici [10]
of Hopf-cyclic cohomology of a universal enveloping algebra U(g) with the Lie algebra
cohomology of g. The resolution came from Khalkhali and Rangipour in [34] where they
showed that for a Hopf algebra H together with a SAYD module M, there is a non-trivial
cyclic module which yields the correct cohomology for group rings, namely the group co-
homology. Next they proved in [38] that the cyclic module they defined in [34] is the cyclic
dual of the cocyclic module of [23]. This fact indicates that there is a deep connection
between the Van Est and cyclic duality isomorphisms in the context of noncommutative
geometry.
The dual cyclic theory with coefficients in objects similar to modular pairs in involu-
tion appeared earlier in the thesis of Taillefer [49] which is published as [47] and [48]. And
also there is Jara and Ştefan’s [25] which is published as [26]. Most notably, the closest def-
inition given to that of Hopf-cyclic cohomology and its dual theory with SAYD coefficients
is Khalkhali and Rangipour’s invariant cyclic cohomology [35]. However, the connections
of these different variants of the dual theory with the Connes-Moscovici Hopf-cyclic co-
homology became apparent after the publication of [23], and Khalkhali and Rangipour’s
cyclic duality isomorphism in Hopf-cyclic cohomology [38].
5. The characteristic map and the cup products

Connes-Moscovici characteristic map (1.6) does not immediately appear to be given
by a product. However, considering the explicit definition in (1.5) one can see that it
involves a pairing of a Hopf-cyclic cocycle of a Hopf algebra H and a cyclic cycle of an H-
module algebra A together with an invariant trace τ. An invariant trace is a 0-dimensional
Hopf-cyclic cocycle very much like an ordinary trace is a 0-dimensional cyclic cocycle.
So, one should think of (1.5) as a pairing
p
HPH1 (H1 , 1 Cδ ) ⊗ HP0H1 (A, 1 Cδ ) −→ HP p (A)
Connes used a similar argument to write the periodicity operator S : HC p (A) → HC p+2 (A)
as a Yoneda product [3]. We have to note that invariant traces on A represent classes in
HP0H1 (A, 1 Cδ ), the dual Hopf-cyclic cohomology of the H1 -module algebra A, and was
not available until Khalkhali and Rangipour defined their dual cohomology in [34]. Us-
ing the Hopf invariant version of the closed graded traces used by Connes [7, Chapter 3],
in [37] Khalkhali and Rangipour developed the natural generalization of the cup product
in Hopf-cyclic cohomology and extended the Connes-Moscovici characteristic map as a
pairing of the form
(5.1) p
HPH (H, M) ⊗ HPqH (A, M) −→ HP p+q (A)
and its non-periodic version
p q
(5.2) HCH (H, M) ⊗ HCH (A, M) −→ HC p+q (A)
for an arbitrary Hopf algebra H, an H-module algebra A and an arbitrary SAYD module
M. Later, building on their methods Nikonov and Sharygin in [44] rewrote the Connes-
Moscovici characteristic map within the Cuntz-Quillen framework using X-complexes.
Since the characteristic map is an instance of a cup product, one can ask how it relates
to other kinds of cohomological products. In order to answer this question, one needs to
express the Hopf-cyclic cohomology as a derived functor. The first attempt in this direction
was given by Khalkhali and Kaygun in [31] which is published as [32]. We should also
mention [45] where Rangipour gives an interpretation of the cup product in terms of cyclic
cocycles similar to [29] instead of Hopf-invariant and cyclic invariant Hochschild cocycles
[37]. But the full task of expressing the Hopf-cyclic cohomology and the characteristic map
in terms of derived functors and cohomological cup products was completed by Kaygun
in [29] and [30].
Now, we have at least three different setups extending the Connes-Moscovici charac-
teristic map, and no apparent way of relating these seemingly different constructions.
(1) The original Connes-Moscovici characteristic map as defined in [10] which is

generalized by Khalkhali and Rangipour in [37]. In this setup the characteris-
tic map is based on the Hopf-invariant version of cyclic invariant Hochschild
cocycles and closed graded traces defined by Connes [7, Chapter 3].
(2) The extension of the Connes-Moscovici characteristic map as defined by Crainic
in [17] and later generalized by Nikonov and Sharygin in [44]. This setup is
based also on invariant closed graded traces but uses Cuntz and Quillen’s X-
complexes.
(3) The extension of the Connes-Moscovici characteristic map as defined by Kaygun
in [29] and [30]. This setup uses derived functor interpretation of Hopf-cyclic
cohomology and Yoneda-type products.
This discrepancy is resolved by Kaygun in [30] by showing that all of these different setups
produce isomorphic characteristic maps. The author accomplished this difficult task by
employing the abstract yet powerful machinery of derived categories and derived functors
on double abelian categories, and their universality properties applied to the various cup
products defined in Hopf-cyclic cohomology.
6. In place of a conclusion
Since the invention of the theory by Connes and Moscovici, Hopf-cyclic cohomology
has emerged as the true replacement of equivariant cohomology of noncommutative spaces
in the last decade. The characteristic map they defined has found deep applications in trans-
verse index theory [9, 10, 12], and quite unexpectedly, in number theory [13, 14, 15]. See
Marcolli’s excellent review on the latter subject at AMS Math Reviews MR2074984. In
this survey article we restricted ourselves to the Connes-Moscovici characteristic map and
some aspects of Hopf-cyclic cohomology even though there were other exciting develop-
ments which we fail to mention here. For a different point of view on the subject we refer
the reader to [39] and [33].
8178 ATABEY KAYGUN
Acknowledgments
The author would like to express his great admiration and respect for Henri Moscovici,
and would like to thank him for his encouragement and support over the years. The au-
thor would also like to thank the organizers of the conference Noncommutative Geometric
Methods in Global Analysis for the opportunity to contribute to this volume honoring Henri
Moscovici on his 65th birthday.
References
1. T. Antal, Cyclic cohomology and Hopf algebras, Ph.D. thesis, The Ohio State University, 2004.
2. R. Bott, Lectures on characteristic classes and foliations, Lecture Notes in Mathematics, vol. 279, pp. 1–94,
Springer Verlag, 1972.
3. A. Connes, Cohomologie cyclique et foncteurs Extn , C. R. Acad. Sci. Paris Sér. I Math. 296 (1983), no. 23,
953–958.
4. , Introduction to noncommutative differential geometry, Lecture Notes in Mathematics, vol. 1111,
Springer Verlag, 1984.
5. , Noncommutative differential geometry, Inst. Hautes Études Sci. Publ. Math. (1985), no. 62, 257–
360.
6. , Cyclic cohomology and the transverse fundamental class of a foliation, Geometric methods in
operator algebras (Kyoto, 1983), Pitman Res. Notes Math. Ser., vol. 123, Longman Sci. Tech., Harlow, 1986,
pp. 52–144.
7. , Noncommutative geometry, Academic Press Inc., 1994 (Also available online at author’s web page
http://www.alainconnes.org/docs/book94bigpdf.pdf).
8. , Cyclic cohomology, noncommutative geometry and quantum group symmetries, Noncommutative
geometry, Lecture Notes in Math., vol. 1831, pp. 1–71.
9. A. Connes and H. Moscovici, Local index theorem in noncommutative geometry, Geom. Func. Anal. 5
(1995), 174–243.
10. , Hopf algebras, cyclic cohomology and the transverse index theorem, Comm. Math. Phys. 198
(1998), 199–246.
11. , Cyclic cohomology and Hopf algebras, Lett. Math. Phys. 48 (1999), no. 1, 97–108, Moshé Flato
(1937–1998).
12. , Differentiable cyclic cohomology and Hopf algebraic structures in transverse geometry, Essays on
geometry and related topics, Vol. 1, 2, Monogr. Enseign. Math., vol. 38, 2001, pp. 217–255.
13. , Modular Hecke algebras and their Hopf symmetry, Mosc. Math. J. 4 (2004), no. 1, 67–109, 310.
14. , Rankin-Cohen brackets and the Hopf algebra of transverse geometry, Mosc. Math. J. 4 (2004),
no. 1, 111–130, 311.
15. , Transgressions of the Godbillon-Vey class and Rademacher functions, Noncommutative geometry
and number theory, Aspects Math., E37, Vieweg, Wiesbaden, 2006, pp. 79–107.
16. M. Crainic, Cyclic cohomology of Hopf algebras, and a non-commutative Chern-Weil theory,
arXiv:math/9812113v3 [math.QA], 1998.
17. , Cyclic cohomology of Hopf algebras, J. Pure Appl. Algebra 166 (2002), no. 1-2, 29–66.
18. J. Cuntz and D. Quillen, Cyclic homology and nonsingularity, J. Amer. Math. Soc. 8 (1995), no. 2, 373–442.
19. A. Gorokhovsky, Secondary characteristic classes and cyclic cohomology of Hopf algebra,
arXiv:math/0002126v1 [math.OA], 2000.
20. , Secondary characteristic classes and cyclic cohomology of Hopf algebras, Topology 41 (2002),
no. 5, 993–1016.
21. T. Hadfield and U. Krähmer, Twisted homology of quantum SL(2), K-Theory 34 (2005), no. 4, 327–360.
22. , Twisted homology of quantum SL(2)-part II, Journal of K-Theory 6 (2010), no. 1, 69–98.
23. P. M. Hajac, M. Khalkhali, B. Rangipour, and Y. Sommerhäuser, Hopf–cyclic homology and cohomology
with coefficients, C. R. Math. Acad. Sci. Paris 338 (2004), no. 9, 667–672.
24. , Stable anti-Yetter–Drinfeld modules, C. R. Math Acad. Sci. Paris 338 (2004), no. 8, 587–590.
25. P. Jara and D. Ştefan, Cyclic homology of Hopf Galois extensions and Hopf algebras,
arXiv:math/0307099v1 [math.KT], 2003.
26. , Hopf-cyclic homology and relative cyclic homology of Hopf-Galois extensions, Proc. London Math.
Soc. (3) 93 (2006), no. 1, 138–174.
27. J. Kaminker and X. Tang, Hopf algebroids and secondary characteristic classes, J. Noncommut. Geom. 3
(2009), no. 1, 1–25.
28. A. Kaygun, Bialgebra cyclic homology with coefficients, K–Theory 34 (2005), no. 2, 151–194.
29. , Products in Hopf-cyclic cohomology, Homology, Homotopy and Applications 10 (2008), no. 2,
115–133.
30. , Uniqueness of pairings in Hopf-cyclic cohomology, Journal of K-theory 6 (2010), no. 1, 1–29.
31. A. Kaygun and M. Khalkhali, Bivariant Hopf cyclic cohomology, arXiv:math/0606341 [math.KT],
2006.
32. , Bivariant Hopf cyclic cohomology, Communications in Algebra 38 (2010), 2513–2537.
33. M. Khalkhali, Basic noncommutative geometry, EMS Series of Lectures in Mathematics, European Mathe-
matical Society (EMS), Zürich, 2009.
34. M. Khalkhali and B. Rangipour, A new cyclic module for Hopf algebras, K–Theory 27 (2002), no. 2, 111–
131.
35. , Invariant cyclic cohomology, K–Theory 28 (2003), no. 2, 183–205.
36. , Para-Hopf algebroids and their cyclic cohomology, Letters in Mathematical Physics (2004), 259–
272.
37. , Cup products in Hopf-cyclic cohomology, C. R. Math. Acad. Sci. Paris 340 (2005), no. 1, 9–14.
38. , A note on cyclic duality and Hopf algebras, Comm. Algebra 33 (2005), 763–773.
39. , Introduction to Hopf-cyclic cohomology, Noncommutative geometry and number theory, Aspects
Math., E37, Vieweg, Wiesbaden, 2006, pp. 155–178.
40. J. Kustermans, J. Rognes, and L. Tuset, The Connes-Moscovici approach to cyclic cohomology for compact
quantum groups, K-Theory 26 (2002), no. 2, 101–137.
41. H. Moscovici and B. Rangipour, Cyclic cohomology of Hopf algebras of transverse symmetries in codimen-
sion 1, Adv. Math. 210 (2007), no. 1, 323–374.
42. , Hopf algebras of primitive Lie pseudogroups and Hopf cyclic cohomology, Adv. Math. 220 (2009),
no. 3, 706–790.
43. , Hopf cyclic cohomology and transverse characteristic classes, arXiv:1009.0955v1 [math.QA],
2010.
44. I. M. Nikonov and G. I. Sharygin, On the Hopf-type cyclic cohomology with coefficients, C ∗ -algebras and
elliptic theory, Trends Math., Birkhäuser, Basel, 2006, pp. 203–212.
45. B. Rangipour, Cup products in Hopf cyclic cohomology via cyclic modules, Homology, Homotopy and Ap-
plications 10 (2008), no. 2, 273–286.
46. G. I. Sharygin, Hopf-type cyclic cohomology via the Karoubi operator, Noncommutative geometry and quan-
tum groups (Warsaw, 2001), Banach Center Publ., vol. 61, Polish Acad. Sci., Warsaw, 2003, pp. 199–217.
47. R. Taillefer, Cohomology theories of Hopf bimodules and cup-product, C. R. Acad. Sci. Paris Sér. I Math.
332 (2001), no. 3, 189–194.
48. , Cyclic homology of Hopf algebras, K–Theory 24 (2001), no. 1, 69–85.
49. , Theories homologiques des algebres de Hopf, Ph.D. thesis, Universite Montpellier II, 2001.
50. W. T. van Est, Group cohomology and Lie algebra cohomology in Lie groups. I, II, Nederl. Akad. Wetensch.
Proc. Ser. A. 56 = Indagationes Math. 15 (1953).
Department of Mathematics and Computer Science, Bahcesehir University, Besiktas 34359 Istanbul,
TURKEY
E-mail address: atabey.kaygun@bahcesehir.edu.tr
Volume 546, 2011
A Super Version of the Connes-Moscovici Hopf Algebra
Masoud Khalkhali and Arash Pourkia
Dedicated, with much appreciation, to Henri Moscovici.
Abstract. We define a super version of the Connes-Moscovici Hopf algebra,

H1 . For that, we consider the supergroup Gs = Dif f + (R1,1 ) of orientation
preserving diffeomorphisms of the superline R1,1 and define two (super) sub-
groups Gs1 and Gs2 of Gs where Gs1 is the supergroup of affine transformations.
The super Hopf algebra Hs1 is defined as a certain bicrossproduct super Hopf
algebra of the super Hopf algebras attached to Gs1 and Gs2 . We also give an
explicit description of Hs1 in terms of generators and relations.
1. Introduction
In [2], Connes and Moscovici, among many other things, defined a Hopf alge-
bra H(n), for any n ≥ 1, and computed the periodic Hopf cyclic cohomology of
H(n). Our main focus in this paper is H(1), to be denoted from now on by H1 ,
and its super analogue. It is by now clear that the Connes-Moscovici Hopf algebra
H1 is a fundamental object of noncommutative geometry. An important feature
of H1 , and in fact its raison d’être, is that it acts as quantum symmetries of vari-
ous algebras of interest in noncommutative geometry, like the algebra of leaves of
codimension one foliations and the algebra of modular forms modulo the action of
Hecke correspondences [2, 3, 4, 5].
Our starting point, in fact the motivation to develop the super analogue of H1 ,
was to extend the results of [5] to cover the Rankin-Cohen brackets on super mod-
ular forms and super pseudodifferential operators as they are described in Section 7
of [1]. In [5] it is shown that the Rankin-Cohen brackets on modular forms [1] can
be derived via the action of H1 on the modular Hecke algebras. In fact, more gen-
erally, it is shown how to obtain such brackets on any associative algebra endowed
with an action of the Hopf algebra H1 , such that the derivation corresponding to
the Schwarzian derivative is inner. To carry out this program in the SUSY case, as
a first step one needs a super analogue of H1 .
2000 Mathematics Subject Classification. Primary 58B34; Secondary 16T05.

Key words and phrases. Connes-Moscovici Hopf algebra, bicrossproduct Hopf algebras, su-
pergroups, super Hopf algebras.
2011
c 0000
c Mathematical
holder)
1
181
182
2 MASOUD KHALKHALI AND ARASH POURKIA
The Connes-Moscovici Hopf algebra H1 is isomorphic to a certain bicrossproduct

Hopf algebra F (G2 )U (g1 ) [2, 3, 8, 11, 12]. The actions and coactions involved in
this bicrossproduct can be derived and understood by looking at the factorization of
the group of orientation preserving diffeomorphisms of the real line, G = Dif f + (R),
into two subgroups G1 and G2 . Here G1 is the group of affine transformations and
G2 is the subgroup of those diffeomorphisms φ with φ(0) = 0 and φ̇(0) = 1, where
d
φ̇(x) = dx (φ(x)).
In this paper our goal is to define a super version of H1 , which we will denote
by H1s . For that we define a super version of the group G = Dif f + (R), namely
the supergroup Gs = Dif f + (R1,1 ) of orientation preserving diffeomorphisms of the
superline R1,1 . We define two (super) subgroups Gs1 and Gs2 of Gs , where Gs1 is
the group of affine transformations. We show that the factorization Gs = Gs1 Gs2
holds. We will use this factorization to define a bicrossproduct super Hopf algebra
F (Gs2 )U (gs1 ), analogous to the non-super case. We will call this bicrossproduct
super Hopf algebra the super version of H1 and denote it by H1s .
One difficulty in working with super Hopf algebras is that they are not ‘honest’
Hopf algebras. In fact their multiplication map is not a morphism of coalgebras. It
is so only up to sign, and this issue of signs can be quite confusing and demands
a lot of care. Throughout this paper, for notations to be more consistent with the
non-super case, we use the following conventions. To denote the comultiplication
Δ : B → B⊗B of a bialgebra B we use Sweedler’s notation (summation understood)
Δn (b) = b(1) ⊗ b(2) ⊗ · · · ⊗ b(n+1) , for any n ≥ 1. Also for a coaction ∇ : A → B ⊗ A
of B on A we write ∇(a) = a(1) ⊗ a(2) .
One of the recent important developments regarding the Connes-Moscovici
Hopf algebras H(n) is in [12], in which the authors generalize the Connes-Moscovici
work, [2], to assign a Hopf algebra to any infinite primitive Lie pseudogroup. They
also introduce an elaborate machinery, based on the bicrossproduct realization of
those Hopf algebras, to compute their periodic and non-periodic Hopf cyclic coho-
mology.
We would like to thank Bahram Rangipour for useful discussions and for sug-
gesting a proof to show that F (Gs2 ) is a super Hopf algebra. We would also like
to thank Arthur Greenspoon for carefully reading the manuscript and many useful
suggestions. We are honored to dedicate this paper, with much appreciation, to
Henri Moscovici on the occasion of his 65th birthday.
2. The Connes-Moscovici Hopf algebra H1

In this section we recall the definition of the standard (non-super) Connes-
Moscovici Hopf algebra H1 and its description in terms of a bicrossproduct Hopf
algebra [2, 3]. The following definition gives a description of H1 by generators and
relations.
Definition 2.1. [2, 3] The Connes-Moscovici Hopf algebra H1 is generated
by elements X, Y , δn , n ≥ 1 with relations:
[Y, X] = X, [X, δn ] = δn+1 , [Y, δn ] = nδn , [δm , δn ] = 0, ∀m, n
Δ(X) = X ⊗ 1 + 1 ⊗ X + Y ⊗ δ1 , Δ(Y ) = Y ⊗ 1 + 1 ⊗ Y, Δ(δ1 ) = δ1 ⊗ 1 + 1 ⊗ δ1 ,
ε(X) = ε(Y ) = ε(δn ) = 0, ∀n,
(2.1) S(X) = Y δ1 − X, S(Y ) = −Y, S(δ1 ) = −δ1 .
A SUPER VERSION OF THE CONNES-MOSCOVICI HOPF ALGEBRA 183
3
Remark 2.2. The above definition follows the right-handed notation as in [8],
in the sense that in the definition of Δ(X), the term δ1 ⊗ Y in [2, 3] is replaced by
Y ⊗ δ1 . Or, alternatively, if we denote the original Connes-Moscovici Hopf algebra
by H1CM , then one can say we are working in H1 cop CM , [12].
Lemma 2.3. [8, 9] let A and H be two Hopf algebras such that A is a left
H-module algebra, and H is a right A-comodule coalgebra. Let furthermore these
structures satisfy the follow compatibility conditions:
Δ(h a) = h(1) (1) a(1) ⊗ h(1) (2) (h(2) a(2) ),
∇r (gh) = g(1) (1) h(1) ⊗ g(1) (2) (g(2) h(2) ),
h(2) (1) ⊗ (h(1) a)h(2) (2) = h(1) (1) ⊗ h(1) (2) (h(2) a),
ε(h a) = ε(h)ε(a), ∇r (1) = 1 ⊗ 1,
for any a, b in A and g, h in H, where we have denoted the actions by h a and
the coactions by ∇r (h) = h(1) ⊗ h(2) . Then the vector space A ⊗ H can be equipped
with a Hopf algebra structure as follows:
(a ⊗ h)(b ⊗ g) = a(h(1) b) ⊗ h(2) g,
Δ(a ⊗ h) = a(1) ⊗ h(1) (1) ⊗ a(2) h(1) (2) ⊗ h(2) ,

ε(a ⊗ h) = ε(a)ε(h),
S(a ⊗ h) = (1 ⊗ S(h(1) ))(S(ah(2) ) ⊗ 1),
for any a, b in A and g, h in H. It is called the left-right bicrossproduct Hopf algebra
and is denoted by AH.
in [2, 3] a Hopf subalgebra of H1 is defined as the unital commutative subalge-
bra of H1 generated by {δn , n ≥ 1}. This Hopf algebra, which we denote by F (G2 ),
is isomorphic to the so-called comeasuring Hopf algebra of the real line, generated
by {an , n ≥ 1} with a1 = 1 and with the following relations [7, 8]:

n
Δ(an ) = ( a i1 · · · a ik ) ⊗ a k ,
k=1 i1 +···+ik =n
ε(an ) = δn,1 ,
c
(2n − c1 )!c1 ! ac11 ac22 · · · an+1
n+1
S(an+1 ) = (−1)n−c1 ,
(n + 1)! c1 !c2 ! · · · cn+1 !
(c1 ,...,cn+1 )∈Λ
where

n+1
n+1
Λ = {(c1 , . . . , cn+1 ) | cj = n, jcj = 2n}.
j=1 j=1
If instead of generators an we work with n!an we get the relations for the so called
Faà di Bruno Hopf algebra which is isomorphic to F (G2 ). One can also define
another Hopf algebra, denoted by U (g1 ), as follows. Let g1 be the Lie algebra
generated by two elements X and Y as in H1 (i.e., [Y, X] = X), and let U (g1 )
denote the universal enveloping algebra of g1 .
184
Lemma 2.4. [8] F (G2 ) is a left U (g1 )-module algebra via the actions
X an = (n + 1)an+1 − 2a2 an , Y an = (n − 1)an ,
and U (g1 ) is a right F (G2 )-comodule coalgebra via the coactions
∇r (X) = X ⊗ 1 + Y ⊗ 2a2 , ∇r (Y ) = Y ⊗ 1.
The actions and coactions are compatible in the sense of Lemma (2.3).
Theorem 2.5. [2, 3, 8, 11, 12] H1 is isomorphic to the bicrossproduct Hopf
algebra F (G2 )U (g1 ).
A good way to understand the actions, coactions and even the notations in-
troduced above is to look at the factorization of the group Dif f + (R) [2, 3]. Let
us recall that given any group G with two subgroups G1 and G2 , we say we have
a group factorization G = G1 G2 if for any g ∈ G there is a unique decomposition
g = ab where a ∈ G1 and b ∈ G2 . Given any group factorization G = G1 G2 , one
has always a left action of G2 on G1 and a right action of G1 on G2 defined in the
following way. First one defines two maps π1 : G → G1 , π2 : G → G2 by π1 (g) = a
and π2 (g) = b, for any g = ab in G, with a ∈ G1 and b ∈ G2 . Next, one can define
the aforementioned actions by g2 g1 = π1 (g2 g1 ) and g2 g1 = π2 (g2 g1 ) for any
g1 in G1 and g2 in G2 .
Let
G = Dif f + (R) = {Φ ∈ Dif f (R) | Φ̇(x) > 0, ∀x ∈ R},
d
where Φ̇(x) = dx (Φ(x)), be the group of orientation preserving homeomorphisms
of the real line and
G1 = {ψ = (a, b) ∈ G | ψ(x) = ax + b, a, b ∈ R, a > 0},
be the affine subgroup of G. The following representation of G1 as a subgroup of
GL(2) is very useful:

a b
(2.2) G1 = (a, b) = ∈ GL(2) | a > 0 .
0 1
Let also
G2 = {φ ∈ G | φ(0) = 0, φ̇(0) = 1}.
+
The factorization Dif f (R) = G1 G2 is as follows. For any Φ in G we have Φ = ψφ
with ψ ∈ G1 and φ ∈ G2 , where
Φ(x) − Φ(0)
(2.3) ψ = (Φ̇(0), Φ(0)), φ(x) = , ∀x ∈ R.
Φ̇(0)
The actions and coactions in lemma (2.4) are induced from the factorization of
the group G = Dif f + (R) as follows:
φ(ax + b) − φ(b)
φ ψ = (aφ̇(b), φ(b)), (φ ψ)(x) = ,
aφ̇(b)
for any ψ = (a, b) in G1 and φ in G2 .
Next, using the matrix representation of G1 in GL(2) as in (2.2), we get a basis
for its Lie algebra g1 = Lie(G1 ). It turns out that g1 is generated by two elements

0 1 1 0
X= , Y = ,
0 0 0 0
5
with the relation [Y, X] = X as in Definition (2.1). On the other hand, in the Hopf
algebra F (G2 ) defined by relations (2), the generators an can be considered as the
following functions on G2 :
1
an (φ) = φ(n) (0), ∀φ ∈ G2 .
n!
The actions defined in Lemma (2.4) can be realized in the following way. The
exponentials etX and etY , as elements of the affine group G1 , are given by
etX = (1, t), etY = (et , 0).
The action X an in lemma (2.4) can be identified as:
d
(X an )(φ) = | an (φ etX )
dt t=0
d φ(x + t) − φ(t)
= |t=0 an ( )
dt φ̇(t)
d φ(n) (t)
= |t=0 ( )
dt n!φ̇(t)
d φ(n) (0) + tφ(n+1) (0)
= |t=0 ( )
dt n!(φ̇(0) + tφ̈(0))
d
= |t=0 (1/n!)(φ(n) (0) − tφ̈(0)φ(n) (0) + tφ(n+1) (0))
dt
= (1/n!)(−φ̈(0)φ(n) (0) + φ(n+1) (0))
= [(n + 1)an+1 − 2a2 an ](φ),
for any φ in G2 . This implies
X an = (n + 1)an+1 − 2a2 an ,
as in Lemma (2.4). A similar computation will give the action Y an as in Lemma
(2.4). The coactions defined in Lemma (2.4) can also be realized using the factor-
ization G = G1 G2 [8].
Note that the above realization of actions and coactions is not necessary for the
proof of Lemma (2.4). It rather gives a good intuition about where those formulas
for actions and coactions come from.
3. The supergroup Gs = Dif f + (R1,1 ) and its factorization

In this section, by replacing R by the supermanifold R1,1 , we define a super
version of the group G = Dif f + (R), namely the supergroup Gs = Dif f + (R1,1 ).
Analogous to the non-super case, we consider two sub supergroups Gs1 and Gs2 of
Gs , where Gs1 is the affine part of Gs . We establish the factorization Gs = Gs1 Gs2 .
This factorization will result in a left action of Gs2 on Gs1 and a right action of Gs1
on Gs2 .
For the general theory of supermanifolds we refer to [6, 13, 14]. The super-
manifold R1,1 , the superline, is a super ringed space S = (R1 , OS ), where OS is a
sheaf of supercommutative R-algebras over R defined, for each U ⊂ R open, by
OS (U ) = C ∞ (U ) ⊗ Λ• (R) = C ∞ (U )[θ],
where θ, the generator of the exterior algebra Λ• (R), is called the odd generator.
We also normally denote the even indeterminate by x. Supermanifolds form a
186
category where a morphism f : S1 → S2 is a morphism of the underlying super

ringed spaces [6, 13, 14].
A super Lie group is a group object in the category of supermanifolds. Al-
ternatively, a super Lie group can be defined as a representable functor from
the category of supermanifolds to the category of groups. A typical example is
GL(p, q) = GL(Rp,q ), the super general linear group of automorphisms of Rp,q .
This supergroup has a matrix representation as follows. It is formed by matrices

A B
,
C D
where A, D are, respectively, p × p and q × q invertible matrices consisting of even
elements, and C, D are, respectively, p × q and q × p matrices consisting of odd
elements. As a special case, GL(2, 1) is formed by matrices
⎛ ⎞
a b x
⎝ c d y ⎠,
z w e
where a, b, c, d, e are even elements, ad − bc = 0, e = 0, and x, y, z, w are odd
elements.
Definition 3.1. The supergroup of orientation preserving diffeomorphisms of
the super real line, R1,1 , is defined as follows:
Gs = Dif f + (R1,1 ) = {Φ(x, θ) = (A(x) + B(x)θ , C(x) + D(x)θ)},
such that A(x), D(x) are even, A(x), D(x) ∈ Dif f (R), Ȧ(x) > 0, Ḋ(x) > 0, and
B(x), C(x) are odd.
Definition 3.2. The affine part of Gs , denoted by Gs1 and also denoted by
Af f (R1,1 ), is defined by:
Gs1 = {ψ(x, θ) = (ax + bθ + e , cx + dθ + f ) ∈ Gs },
such that a, e, d are even, a, d > 0, and b, c, f are odd. An element ψ(x, θ) =
(ax + bθ + e , cx + dθ + f ) of Gs1 can also be represented in the following way:

a b x e
ψ(x, θ) = + .
c d θ f
Therefore there exists the following representation of Gs1 in GL(2, 1), which will be
very useful for our purpose:
⎧ ⎛ ⎞ ⎫
⎨ a e b ⎬
Gs1 = ψ = ⎝ 0 1 0 ⎠ | a, e, d are even, b, c, f are odd, and a, d > 0 .
⎩ ⎭
c f d
Definition 3.3. A second super subgroup Gs2 of Gs is defined by:
Gs2 = {φ = (A(x) + B(x)θ, C(x) + D(x)θ) ∈ Gs | φ(0, 0) = 0, Jφ(0, 0) = 1}.
In other words
(3.1) Gs2 = {φ = (A(x) + B(x)θ, C(x) + D(x)θ) ∈ Gs },
where A(0) = B(0) = C(0) = Ċ(0) = 0, and Ȧ(0) = D(0) = 1.
7
In order to proceed to the factorization Gs = Gs1 Gs2 , for any Φ ∈ Dif f + (R1,1 )
we define

Ȧ(0) B(0) x A(0)
(3.2) π1 (Φ) = (JΦ(0, 0), Φ(0, 0)) = + ∈ Gs1 ,
Ċ(0) D(0) θ C(0)
and
π2 (Φ) = (JΦ(0, 0))−1 (Φ(x, θ) − Φ(0, 0))
(3.3) −1
Ȧ(0) B(0)
= (Φ(x, θ) − Φ(0, 0)) ∈ Gs2 .
Ċ(0) D(0)
Here we have used the definition

∂Φ1
∂x − ∂Φ
∂θ
1
Ȧ(x) + Ḃ(x)θ B(x)
JΦ(x, θ) := ∂Φ2 ∂Φ2 = ,
∂x ∂θ Ċ(x) + Ḋ(x)θ D(x)
where Φ1 = A(x) + B(x)θ and Φ2 = C(x) + D(x)θ are, respectively, the even and
odd components of Φ ∈ Dif f + (R1,1 ). The operator ∂x ∂
is even and the operator
∂
∂θ is odd. Also, the formula for the inverse supermatrix is
−1
a b 1 d + bc
a −b
= .
c d da −c a + cbd
Now if we let ψ(x, θ) = π1 (Φ) and φ(x, θ) = π2 (Φ), it is clear that for any
Φ ∈ Dif f + (R1,1 )
Φ = ψφ,
which proves the factorization
Gs = Gs1 Gs2 .
Therefore, we have the following two natural actions. The left action of Gs2 on GS1 ,
Gs2 × Gs1 → Gs1 , defined by φ ψ = π1 (φψ), and the right action of Gs1 on Gs2 ,
Gs2 × Gs1 → Gs2 , defined by φ ψ = π2 (φψ).
4. The super Hopf algebras U (gs1 ) and F (Gs2 )

In this section we assign to supergroups Gs1 and Gs2 , the super Hopf algebras
U (gs1 )
and F (Gs2 ), respectively. The super Hopf algebra U (gs1 ) is just the universal
enveloping algebra of the super Lie algebra gs1 = Lie(Gs1 ). As for F (Gs2 ), we
define the coordinate functions an , bn , cn and dn on Gs2 . Then F (Gs2 ) would be the
corresponding Faà di Bruno or rather comeasuring super Hopf algebra generated,
as a supercommutative superalgebra, by an , bn , cn and dn , for which we will define
the Hopf algebra structure.
4.1. The super Hopf algebra U (gs1 ). The super Lie algebra gs1 = Lie(Gs1 )
is generated by three even generators:
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
0 1 0 1 0 0 0 0 0
X = ⎝ 0 0 0 ⎠, Y = ⎝ 0 0 0 ⎠, Z = ⎝ 0 0 0 ⎠,
0 0 0 0 0 0 0 0 1
and three odd generators
188
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
0 0 1 0 0 0 0 0 0
U = ⎝ 0 0 0 ⎠, V = ⎝ 0 0 0 ⎠, W =⎝ 0 0 0 ⎠.
0 0 0 −1 0 0 0 −1 0
It is easy to check that the following super bracket relations hold:
[Y, X] = X, [X, Z] = 0, [Y, Z] = 0,
[X, U ] = 0, [X, V ] = −W, [X, W ] = 0,
[Y, U ] = U, [Y, V ] = −V, [Y, W ] = 0,
[Z, U ] = −U, [Z, V ] = V, [Z, W ] = W,
[U, V ] = −(Y + Z), [U, W ] = −X, [V, W ] = 0.
(4.1) [X, X] = [Y, Y ] = [Z, Z] = [U, U ] = [V, V ] = [W, W ] = 0.
The super Hopf algebra U (gs1 ) is the universal enveloping algebra of gs1 .
4.2. The super Hopf algebra F (Gs2 ). The super Hopf algebra F (Gs2 ), as a
supercommutative superalgebra, is the super polynomial algebra R[an , bn , cn , dn ],
generated by two sets of even generators an , dn , n ≥ 0, a0 = 0, a1 = d0 = 1,
and two sets of odd generators bn , cn , n ≥ 0, b0 = c0 = c1 = 0, where for any
φ(x, θ) = (A(x) + B(x)θ , C(x) + D(x)θ) in Gs2 we have:
an (φ) = (1/n!)A(n) (0),
bn (φ) = (1/n!)B (n) (0),
cn (φ) = (1/n!)C (n) (0),
(4.2) dn (φ) = (1/n!)D(n) (0).
To define the coproduct on F (Gs2 ), analogous to the non-super case, we use the
following formalism:
(4.3) mΔ(an )(φ ⊗ φ ) = an (φ ◦ φ ), φ, φ ∈ Gs2 ,
and similar formulas for bn , cn and dn . Note that using this formalism provides us
with the coassociativity property of Δ.
We need to study the composition of two elements of Gs2 . Let φ(x, θ) = (A(x)+
B(x)θ , C(x) + D(x)θ) and φ (x, θ) = (A (x) + B (x)θ , C (x) + D (x)θ) be two
elements of Gs2 . Let f = A (x) + B (x)θ and g = C (x) + D (x)θ. It is easy to check
that the composition is given by
φ ◦ φ (x, θ) = (A(f ) + B(f )g , C(f ) + D(f )g),
where
A(f ) + B(f )g = [A(A (x) + B(A (x))C (x)] +
[Ȧ(A (x))B (x) + B(A (x))D (x) − Ḃ(A (x))B (x)C (x)]θ,
and
C(f ) + D(f )g = [C(A (x)) + D(A (x))C (x)] +
[Ċ(A (x))B (x) + D(A (x))D (x) − Ḋ(A (x))B (x)C (x)]θ.
Now we proceed to comultiplications, starting with Δ(an ). It is not hard to
show that
a1 (φ ◦ φ ) = 1 = m(1 ⊗ 1)(φ ⊗ φ ),
9
a2 (φ ◦ φ ) = a2 + a2 = m(1 ⊗ a2 + a2 ⊗ 1)(φ ⊗ φ ),

a3 (φ ◦ φ ) = a3 + 2a2 a2 + a3 + b1 c2 = m(1 ⊗ a3 + 2a2 ⊗ a2 + a3 ⊗ 1 + b1 ⊗ c2 )(φ ⊗ φ ).
Therefore by (4.3) we have:
Δ(a1 ) = Δ(1) = 1 ⊗ 1,
Δ(a2 ) = 1 ⊗ a2 + a2 ⊗ 1,
Δ(a3 ) = 1 ⊗ a3 + a3 ⊗ 1 + 2a2 ⊗ a2 + b1 ⊗ c2 ,
and more generally for n ≥ 1,

n
Δ(an ) = ak ⊗ a l1 a l2 · · · a lk +
k=1 l1 +l2 +···+lk =n

n
i
bk ⊗ ( al1 al2 · · · alk )cn−i .
i=1 k=1 l1 +l2 +···+lk =i
With a similar method we have, for Δ(bn ),

Δ(b1 ) = 1 ⊗ b1 + b1 ⊗ 1,
Δ(b2 ) = 1 ⊗ b2 + b2 ⊗ 1 + 2a2 ⊗ b1 + b1 ⊗ d1 + b1 ⊗ a2 ,
Δ(b3 ) = 1 ⊗ b3 + b3 ⊗ 1 + 2a2 ⊗ b2 + 2a2 ⊗ a2 b1 + 3a3 ⊗ b1 + b1 ⊗ d2

+ b1 ⊗ a2 d1 + b2 ⊗ d1 + b1 ⊗ a3 + 2b2 ⊗ a2 − b1 ⊗ b1 c2 ,
and in general

n
i
Δ(bn ) = 1 ⊗ bn + (k + 1)ak+1 ⊗ ( al1 al2 · · · alk )bn−i
i=1 k=1 l1 +l2 +···+lk =i

n
i
+ bk ⊗ ( al1 al2 · · · alk )dn−i
i=1 k=1 l1 +l2 +···+lk =i
⎛ ⎞
n
i
j
− ⎝b1 ⊗ bi cn−i + (k + 1)bk+1 ⊗ ( al1 al2 · · · alk )bi−j cn−i ⎠ .
i=1 j=1 k=1 l1 +l2 +···+lk =j
For Δ(cn ), n ≥ 1, we have:

n
Δ(cn ) = 1 ⊗ cn + ck ⊗ a l1 a l2 · · · a lk +
k=1 l1 +l2 +···+lk =n

n
i
dk ⊗ ( al1 al2 · · · alk )cn−i .
i=1 k=1 l1 +l2 +···+lk =i
In particular for n = 2, 3:
Δ(c2 ) = 1 ⊗ c2 + c2 ⊗ 1,
Δ(c3 ) = 1 ⊗ c3 + c3 ⊗ 1 + 2c2 ⊗ a2 + d1 ⊗ c2 .
190
For Δ(dn ), n ≥ 1, we have:

n
i
Δ(dn ) = 1 ⊗ dn + (k + 1)ck+1 ⊗ ( al1 al2 · · · alk )bn−i
i=1 k=1 l1 +l2 +···+lk =i

n
i
+ dk ⊗ ( al1 al2 · · · alk )dn−i
i=1 k=1 l1 +l2 +···+lk =i
⎛ ⎞
n
i
j
− ⎝d1 ⊗ bi cn−i + (k + 1)dk+1 ⊗ ( al1 al2 · · · alk )bi−j cn−i ⎠ .
i=1 j=1 k=1 l1 +l2 +···+lk =j
In particular for n = 1, 2, 3:
Δ(d1 ) = 1 ⊗ d1 + d1 ⊗ 1,
Δ(d2 ) = 1 ⊗ d2 + d2 ⊗ 1 + 2c2 ⊗ b1 + d1 ⊗ d1 + d1 ⊗ a2 ,
Δ(d3 ) = 1 ⊗ d3 + d3 ⊗ 1 + 2c2 ⊗ b2 + 2c2 ⊗ a2 b1 + 3c3 ⊗ b1 + d1 ⊗ d2

+ d1 ⊗ a2 d1 + d2 ⊗ d1 + d1 ⊗ a3 + 2d2 ⊗ a2 − d1 ⊗ b1 c2 .
We extend Δ linearly to F (G2 ) via the relation
Δ(ab) := Δ(a)Δ(b), ∀a, b ∈ F (G2 ),
where multiplication on the right hand side operates in the super sense. We also
set ε(1) = 1, and define ε to be equal to zero on all other generators. This defines
a super bialgebra structure on F (G2 ).
To prove that F (Gs2 ) is a super Hopf algebra, the only missing data is the
antipode map. Analogous to the non-super case, it suffices to define an anti-algebra
and coalgebra map S : F (Gs2 ) → F (Gs2 ) satisfying
S(a)(φ) = a(φ−1 ), φ ∈ Gs2 , a = an , bn , cn , dn .
We shall do this in an inductive fashion at the end of the next section, where we
define the actions and coactions between F (Gs2 ) and U (gs1 ). The reason behind this
is that we want to use the fact that the antipode, if it exists, should interact, in a
nice way, with those actions and coactions, in the sense of relation (5.8) below [10].
5. Actions and coactions

In this section we prove that F (Gs2 ) is a super left U (gs1 )-module algebra and
U (gs1 )is a super
⎛ right F (G
s
⎞2 )-comodule coalgebra.
1 0 0
For Y = ⎝ 0 0 0 ⎠ ∈ gs1 we have
0 0 0
⎛ t ⎞
e 0 0
etY = ⎝ 0 1 0 ⎠ = (et x, θ) ∈ Gs1 , (t even).
0 0 1
The action Y an can be realized by
d
(Y an )(φ) = | an (φ etY ),
dt t=0
11
for any φ = (A(x) + B(x)θ, C(x) + D(x)θ) ∈ Gs2 . Thus, by computing φ etY =
π2 (φetY ), we obtain
an (φ etY ) = (1/n!)e(n−1)t A(n) (0),
and
d
(Y an )(φ) = | an (φ etY ) = (n − 1)an (φ).
dt t=0
This implies that
Y an = (n − 1)an .
In the same way we have
Y bn = (n − 1)bn ,
Y cn = ncn ,
Y dn = ndn .
Using the same method we can derive all other actions. In fact we have the
following lemma:
Lemma 5.1. Let us define the actions of X, Y, Z, U, V, W on an , bn , cn , dn by
the following relations:
X an = (n + 1)an+1 − 2an a2 − b1 cn , X bn = (n + 1)bn+1 − 2bn a2 − b1 dn ,
X cn = −2c2 an + (n + 1)cn+1 − cn d1 , X dn = −2c2 bn + (n + 1)dn+1 − dn d1 ,
Y an = (n − 1)an , Y bn = (n − 1)bn , Y cn = ncn , Y dn = ndn ,
Z an = 0, Z bn = bn , Z cn = −cn , Z dn = 0,
U an = cn , U bn = −(n + 1)an+1 + dn , U cn = 0, U dn = (n + 1)cn+1 ,
V an = bn−1 , V bn = 0, V cn = an − dn−1 , V dn = +bn ,
W an = −b1 an + bn , W bn = −b1 bn ,
(5.1) W cn = d1 an − dn , W dn = d1 bn ,
and extend those actions to an action of U (gs1 ) on F (Gs2 ) such that
(5.2) (gh) a = g (h a),
(5.3) h (ab) := (−1)|a||h(2) | (h(1) a)(h(2) b),

for a, b in F (Gs2 ) and g, h in U (gs1 ). Then F (Gs2 ) is a super left U (gs1 )-module
algebra.
Proof. It is enough to show that this action is consistent with the bracket
relations (4.1). We check this just for some of the bracket relations. The rest would
be the same:
(5.2) (5.1)
(Y X) an = Y (X an ) = Y ((n + 1)an+1 − 2an a2 − b1 cn )
= (n + 1)(Y an+1 ) − 2Y (an a2 ) − Y (b1 cn )
(5.3)
= n(n + 1)an+1 − 2(an (Y a2 ) + (Y an )a2 ) − (b1 (Y cn ) + (Y b1 )cn )
= (n2 + n)an+1 − 2nan a2 − nb1 cn ,
(XY ) an = X (Y an ) = X ((n − 1)an )
= (n − 1)((n + 1)an+1 − 2an a2 − b1 cn )
= (n2 − 1)an+1 − 2(n − 1)an a2 − (n − 1)b1 cn .
192
Therefore,
(5.1)
[Y, X] an = (Y X − XY ) an = (n + 1)an+1 − 2an a2 − b1 cn = X an ,
which is consistent with the relation [Y, X] = X of (4.1). Next we check a bracket
involving odd generators:
(U W ) dn = U (W dn ) = U (d1 bn )
= (d1 (U bn ) + (U d1 )bn ) = d1 (−(n + 1)an+1 + dn ) + 2c2 bn
= −(n + 1)d1 an+1 + d1 dn + 2c2 bn ,
(W U ) dn = W (U dn ) = W ((n + 1)cn+1 ) = (n + 1)d1 an+1 − (n + 1)dn+1 .
Therefore,
[U, W ] dn = (U W + W U ) dn = 2c2 bn − (n + 1)dn+1 + d1 dn = (−X) dn ,
which agrees with the relation [U, W ] = −X.

By using almost the same method we can find the coactions and prove the
following lemma. Let us denote the right coaction of F (Gs2 ) on U (gs1 ), by ∇r :
U (gs1 ) → U (gs1 ) ⊗ F (Gs2 ).
Lemma 5.2. Let us define the coactions of F (Gs2 ) on generators X, Y, Z, U, V, W
of U (gs1 ) by
∇r (X) = 2Y ⊗ a2 + X ⊗ 1 + Z ⊗ d1 + U ⊗ b1 + 2V ⊗ c2 , ∇r (Y ) = Y ⊗ 1,
∇r (Z) = Z ⊗ 1, ∇r (U ) = U ⊗ 1, ∇r (V ) = V ⊗ 1,
(5.4) ∇r (W ) = Y ⊗ b1 + V ⊗ d1 + W ⊗ 1,
and extend them to U (gs1 ) such that
∇r (gh) = (−1)|h |(|g(1) (2) |+|g(2) |)
(1)
(5.5) g(1) (1) h(1) ⊗ g(1) (2) (g(2) h(2) ),
for all g and h in U (gs1 ). Then U (gs1 ) is a super right F (Gs2 )-comodule coalgebra.
Proof. It is straightforward to check the coaction property, (id ⊗ Δ)∇r (h) =
(∇r ⊗ id)∇r (h), for all h in U (gs1 ), in other words:
h(1) ⊗ h(2) (1) ⊗ h(2) (2) = h(1)(1) ⊗ h(1)(2) ⊗ h(2) .
We prove that this coaction is consistent with the bracket relations (4.1). We verify
this only for one of the purely odd cases. The rest are similar. By formula (5.5) we
have
∇r (V W ) = (−1)|W |(|V(1) (2) |+|V(2) |)

(1)
V(1) (1) W (1) ⊗ V(1) (2) (V(2) W (2) )
= [(−1)|W ||V |)
(1)
W (1) ⊗ (V W (2) )]
+ [(−1)|W ||V (2) |
(1)
V (1) W (1) ⊗ V (2) W (2) ]
= [(−1)|W |
(1)
W (1) ⊗ (V W (2) )] + [V (1) W (1) ⊗ W (2) ]
= [Y ⊗ (V b1 ) − V ⊗ (V d1 ) − W ⊗ (V 1)]
+ [V Y ⊗ b1 + V 2 ⊗ d1 + V W ⊗ 1]
= [−V ⊗ b1 ] + [V Y ⊗ b1 + V 2 ⊗ d1 + V W ⊗ 1,
13
and
∇r (W V ) = (−1)|V |(|W(1) (2) |+|W(2) |)
(1)
W(1) (1) V (1) ⊗ W(1) (2) (W(2) V (2) )
= [(−1)|V ||W |)
(1)
V (1) ⊗ (W V (2) )]
+ [(−1)|V ||W (2) |
(1)
W (1) V (1) ⊗ W (2) V (2) ]
= [−V ⊗ (0)] + [(−1)|W |
(2)
W (1) V ⊗ W (2) ]
= −Y V ⊗ b1 + V 2 ⊗ d1 + W V ⊗ 1.
Thus,
∇r ([V, W ]) = ∇r (V W + W V ) = ∇r (V W ) + ∇r (W V )
= [V, Y ] ⊗ b1 + [V, V ] ⊗ d1 + [V, W ] ⊗ 1 − v ⊗ b1 = 0,
which agrees with [V, W ] = 0 of relations (4.1).
We also leave it to the reader to check that U (gs1 ) is a right F (Gs2 )-comodule
coalgebra, i.e., for all h in U (gs1 ),
h(1) (1) ⊗ h(1) (2) ⊗ h(2) = (−1)|h(2) ||h(1) (2) |

(1)
h(1) (1) ⊗ h(2) (1) ⊗ h(1) (2) h(2) (2) .

Antipode for F (Gs2 ). Now we prove that F (Gs2 ) is a super Hopf algebra, by
defining an anti-algebra and coalgebra map S : F (Gs2 ) → F (Gs2 ) satisfying the
antipode property or the following identity:
(5.6) S(a)(φ) = a(φ−1 ), φ ∈ Gs2 , a = an , bn , cn , dn .
We do this, inductively, by defining S on generators an , bn , cn , dn , and then extend
it linearly to F (Gs2 ) via the following relations:
S(ab) := (−1)|a||b| S(b)S(a),
Δ(S(a)) := (−1)|a(1) ||a(2) | S(a(2) ) ⊗ S(a(1) ).

Let us define, first, S(1) = 1, and S(a) = −a, for a = a2 , b1 , c2 , d1 . It is obvious
that the antipode property holds for these elements. Now suppose relation (5.6) is
true for all elements ai , bi , ci , di , i ≤ n. The actions of X defined in relations (5.1)
in Lemma 5.1 allow us to write the higher degree elements, an+1 , bn+1 , cn+1 , dn+1 ,
in terms of some other elements of lower degree, ai , bi , ci , di , i ≤ n. Therefore, to
finish the process it is enough to prove that
(5.7) S(X a)(φ) = (X a)(φ−1 ), φ ∈ Gs2 , a = ai , bi , ci , di , i ≤ n.
To prove this identity we need the following three lemmas.
Lemma 5.3. [10] The antipode S : F (Gs2 ) → F (Gs2 ), if it exists, should satisfy
the following relation:
(5.8) S(g a) = (g (1) S(a))S(g (2) ), g ∈ U (gs1 ), a ∈ F (Gs2 )
Lemma 5.4. For any φ = (A(x) + B(x)θ, C(x) + D(x)θ) in Gs2 , one can prove
d (1) d
(5.9) |t=0 etX X (2) (φ) = |t=0 φ etX .
dt dt
194
Proof. From the coactions defined in Lemma 5.2 we have:

d
LHS = |t=0 etY 2a2 (φ) + etX 1(φ) + etZ d1 (φ) − etU b1 (φ) − etV 2c2 (φ)
dt
d
= |t=0 ((et X, θ)2a2 + (X + t, θ) + (X, et θ)d1
dt
− (X + tθ, θ)b1 − (X, −tX + θ)2c2 )
= (X, 0)2a2 + (1, 0) + (0, θ)d1 − (θ, 0)b1 − (0, −X)2c2
= (2a2 X + 1 + b1 θ, 2c2 X + d1 θ)
From the discussion in Section 3 and formula (3.2), we have

d d Ȧ(t) B(t) x A(t)
LHS = |t=0 (π1 (φ(X + t, θ))) = |t=0 +
dt dt Ċ(t) D(t) θ C(t)
d
= |t=0 Ȧ(t)X + A(t) + B(t)θ, Ċ(t)X + C(t) + D(t)θ
dt

= Ä(0)X + Ȧ(0) + Ḃ(0)θ, C̈(0)X + Ċ(0) + Ḋ(0)θ
= (2a2 X + 1 + b1 θ, 2c2 X + d1 θ)
= RHS

Lemma 5.5. If G = G1 G2 is a factorisation of the group G, then, for any ϕ1 ,

ϕ2 in G2 and ψ in G1 , one has
(5.10) (ϕ1 ϕ2 ) ψ = (ϕ1 (ϕ2 ψ)) (ϕ2 ψ)
Now we prove the identity (5.7), S(X a)(φ) = (X a)(φ−1 ).
Proof.
Lemma 5.3
S(X a)(φ) = (X (1) S(a))(φ) S(X (2))(φ)
d (1)

= |t=0 S(a)(φ etX ) S(X (2) )(φ)
dt

Induction d
|t=0 a((φ etX )−1 ) X (2) (φ−1 )
(1)
=
dt

Lemma 5.5 d
|t=0 a(φ−1 (φ etX )) X (2) (φ−1 )
(1)
=
dt
Lemma 5.4 d
= |t=0 a(φ−1 etX )
dt
= (X a)(φ−1 )

6. Compatibilities and the super Hopf algebra H1s

The following proposition is the super analogue of Lemma (2.3). It gives the
compatibility conditions to construct a bicrossproduct super Hopf algebra. To
complete the construction of the bicrossproduct super Hopf algebra F (Gs2 )U (gs1 )
one needs to check these compatibility conditions between the actions and coactions
introduced in Lemmas (5.1) and (5.2) in the last section.
15
Proposition 6.1. Let A and H be two super Hopf algebras such that A is a
left H-module algebra, and H is a right A-comodule coalgebra. Let furthermore
these structures satisfy the follow compatibility conditions:
Δ(h a) = (−1)|a(1) |(|h(1) |+|h(2) |)

(2)
(6.1) h(1) (1) a(1) ⊗ h(1) (2) (h(2) a(2) ),
ε(h a) = ε(h)ε(a),
∇r (gh) = (−1)|h |(|g(1) (2) |+|g(2) |)

(1)
(6.2) g(1) (1) h(1) ⊗ g(1) (2) (g(2) h(2) ),
∇r (1) = 1 ⊗ 1,
(6.3)
(−1)|h(1) ||h(2) |+|a||h(2) |
(1) (2)
h(2) (1) ⊗ (h(1) a)h(2) (2) = h(1) (1) ⊗ h(1) (2) (h(2) a),
for any a, b in A and g, h in H, where we have denoted the actions by, h a and
the coactions by ∇r (h) = h(0) ⊗ h(1) . Then the super vector space A ⊗ H can be
equipped with a super Hopf algebra structure as follows:
(6.4) (a ⊗ h)(b ⊗ g) = (−1)|h(2) ||b| a(h(1) b) ⊗ h(2) g,
Δ(a ⊗ h) = (−1)|h(1) ||a(2) |

(1)
(6.5) a(1) ⊗ h(1) (1) ⊗ a(2) h(1) (2) ⊗ h(2) ,
ε(a ⊗ h) = ε(a)ε(h),
S(a ⊗ h) = (−1)|h ||a|

(1)
(1 ⊗ S(h(1) ))(S(ah(2) ) ⊗ 1),
for any a, b in A and g, h in H. We call this super Hopf algebra the left-right
bicrossproduct super Hopf algebra AH.
Proof. We prove that Δ, defined in (6.5), is an algebra map. First we have:
(6.6)
(6.4) α0
Δ((a ⊗ h) · (b ⊗ g)) = (−1) Δ(a(h(1) b) ⊗ h(2) g)
(6.5) α0 +α1
= (−1) (a(h(1) b))(1) ⊗ (h(2) g)(1) (1) ⊗ (a(h(1) b))(2) (h(2) g)(1) (2)
⊗ (h(2) g)(2)
α0 +α1 +α2 +α3 (1)
= (−1) a(1) (h(1) b)(1) ⊗ (h(2)(1) g(1) )
(2)
⊗ a(2) (h(1) b)(2) (h(2)(1) g(1) ) ⊗ h(2)(2) g(2)
(6.1),(6.2) α0 +α1 +α2 +α3 +α4 +α5
= (−1) a(1) (h(1)(1) (1) b(1) ) ⊗ (h(2)(1)(1) (1) g(1) (1) )
⊗ a(2) h(1)(1) (2) (h(1)(2) b(2) )(h(2)(1)(1) (2) (h(2)(1)(2) g(1) (2) )) ⊗ h(2)(2) g(2)
coassociativity α0 +α1 +α2 +α3 +α4 +α5
= (−1) a(1) (h(1) (1) b(1) ) ⊗ (h(3) (1) g(1) (1) )
⊗ a(2) h(1) (2) (h(2) b(2) )(h(3) (2) (h(4) g(1) (2) )) ⊗ h(5) g(2) ,
196
where
α0 = |h(2) ||b| = (|h(2)(1)(1) | + ||h(2)(1)(2) | + ||h(2)(2) |)(|b(1) | + |b(2) |)
= (|h(3) | + ||h(4) | + ||h(5) |)(|b(1) | + |b(2) |)
(1)
α1 = |(h(2) g)(1) ||(a(h(1) b)(2) |
= |(h(2)(1)(1) (1)
g(1) (1) )||a(2) h(1)(1) (2) (h(1)(2) b(2) )|
= (|h(2)(1)(1) (1) | + |g(1) (1) |)(|a(2) | + |h(1)(1) (2) | + |h(1)(2) | + |b(2) |)
= (|h(3) (1) | + |g(1) (1) |)(|a(2) | + |h(1) (2) | + |h(2) | + |b(2) |)
= |g(1) (1) ||a(2) | + |g(1) (1) ||h(1) (2) | + |g(1) (1) ||h(2) | + |g(1) (1) ||b(2) |
+ |h(3) (1) ||a(2) | + |h(3) (1) ||h(1) (2) | + |h(3) (1) ||h(2) | + |h(3) (1) ||b(2) |
α2 = |a(2) ||(h(1) b)(1) | = |a(2) ||(h(1)(1) (1) b(1) )|
= |a(2) |(|h(1)(1) (1) | + |b(1) |)
= |a(2) |(|h(1) (1) | + |b(1) |) = |a(2) ||h(1) (1) | + |a(2) ||b(1) |
α3 = |h(2)(2) ||g(1) | = |h(5) ||g(1) |
α4 = |b(1) |(|(h(1)(2) | + |h(1)(1) (2) |)
= |b(1) |(|(h(2) | + |h(1) (2) |) = |b(1) ||(h(2) | + |b(1) ||h(1) (2) |
α5 = |g(1) (1) |(|h(2)(1)(2) | + |h(2)(1)(1) (2) |)
= |g(1) (1) |(|h(4) | + |h(3) (2) |) = |g(1) (1) ||h(4) | + |g(1) (1) ||h(3) (2) |
On the other hand we have

Δ(a ⊗ h) · Δ(b ⊗ g)
(6.5) β0
= (−1) (a(1) ⊗ h(1) (1) ⊗ a(2) h(1) (2) ⊗ h(2) )·
(b(1) ⊗ g(1) (1) ⊗ b(2) g(1) (2) ⊗ g(2) )
β0 +β1
= (−1) (a(1) ⊗ h(1) (1) )·
(b(1) ⊗ g(1) (1) ) ⊗ (a(2) h(1) (2) ⊗ h(2) ) · (b(2) g(1) (2) ⊗ g(2) )
(6.4) β0 +β1 +β2
= (−1) a(1) (h(1) (1) (1) b(1) ) ⊗ h(1) (1) (2) g(1) (1)
⊗ a(2) h(1) (2) (h(2)(1) b(2) g(1) (2) ) ⊗ h(2)(2) g(2)
where β0 = |h(1) (1) ||a(2) | + |g(1) (1) ||b(2) |, β1 = |(b(1) ⊗ g(1) (1) )||(a(2) h(1) (2) ⊗ h(2) )|,
and β2 = |b(1) ||h(1) (1) (2) | + |b(2) g(1) (2) ||h(2)(2) |.
Continuing this computation, using the standard sign rules for superalgebras,
coalgebras and Hopf algebras and coassociativity of ΔH , we obtain
(6.7)
β0 +β1 +β2 +β3 −β4 −β5 −β6 +β7 −β8 +β9 −β10 +β11 −β12 +β13 +β14 +β15
Δ(a ⊗ h) · Δ(b ⊗ g) = (−1)
(a(1) h(1) (1) b(1) ⊗ h(3) (1) g(1) (1)
⊗ a(2) h(1) (2) (h(2) b(2) )h(3) (2) (h(4) g(1) (2) ) ⊗ h(5) g(2) )
17
where at the end

β0 + β1 + β2 + β3 − β4 − β5 − β6 + β7 − β8 + β9 − β10 + β11 − β12 + β13 + β14 + β15
= (|h(3) | + ||h(4) | + ||h(5) |)(|b(1) | + |b(2) |)
+ |g(1) (1) ||a(2) | + |g(1) (1) ||h(1) (2) | + |g(1) (1) ||h(2) | + |g(1) (1) ||b(2) |
+ |h(3) (1) ||a(2) | + |h(3) (1) ||h(1) (2) | + |h(3) (1) ||h(2) | + |h(3) (1) ||b(2) |
+ |a(2) ||h(1) (1) | + |a(2) ||b(1) |
+ |h(5) ||g(1) | + |b(1) ||(h(2) | + |b(1) ||h(1) (2) | + |g(1) (1) ||h(4) | + |g(1) (1) ||h(3) (2) |
Therefore,
(6.8) α0 + α1 + α2 + α3 + α4 + α5 =
β0 + β1 + β2 + β3 − β4 − β5 − β6 + β7 − β8 + β9 − β10 + β11 − β12 + β13 + β14 + β15 .
By (6.6), (6.7) and (6.8) we have
Δ((a ⊗ h) · (b ⊗ g)) = Δ(a ⊗ h) · Δ(b ⊗ g).
The rest of the proof can be done by a similar method.
Remark 6.2. The above lemma is actually true in any symmetric monoidal
category. The proof involves braiding diagrams.
We skip the lengthy, but computational, proof of the following theorem which
is the main result of this paper.
Theorem 6.3. The actions and coactions defined in Lemmas (5.1) and (5.2)
satisfy the conditions (6.1)-(6.3) in Proposition (6.1). Therefore we have a bi-
crossproduct super Hopf algebra F (Gs2 )U (gs1 ) with the following structures:
(a ⊗ h)(b ⊗ g) = (−1)|h(2) ||b| a(h(1) b) ⊗ h(2) g,
Δ(a ⊗ h) = (−1)|h(1) ||a(2) |

(1)
a(1) ⊗ h(1) (1) ⊗ a(2) h(1) (2) ⊗ h(2) ,
ε(a ⊗ h) = ε(a)ε(h),
S(a ⊗ h) = (−1)|h ||a|
(1)
(1 ⊗ S(h(1) ))(S(ah(2) ) ⊗ 1),
for any a, b in F (Gs2 ) and g, h in U (gs1 ).
The bicrossproduct super Hopf algebra H1s := F (Gs2 )U (gs1 ) is the super ver-
sion of the Connes-Moscovici Hopf algebra H1 .
References
[1] P. Beazley Cohen, Y. Manin, and D. Zagier, Automorphic pseudodifferential operators, Alge-
braic aspects of integrable systems, 17–47, Progr. Nonlinear Differential Equations Appl., 26,
Birkhäuser Boston, Boston, MA, 1997.
[2] A. Connes and H. Moscovici, Hopf algebras, Cyclic cohomology and the transverse index
theorem, Comm. Math. Phys. 198 (1998), no. 1, 199–246.
[3] A. Connes, and H. Moscovici, Cyclic cohomology and Hopf algebra symmetry, Lett. Math.
Phys. 52 (2000), no. 1, 1–28.
[4] A. Connes, and H. Moscovici, Modular Hecke algebras and their Hopf symmetry, Mosc. Math.
J. 4 (2004), no. 1, 67–109, 310.
[5] A. Connes, and H. Moscovici, Rankin-Cohen brackets and the Hopf algebra of transverse
geometry, Mosc. Math. J. 4 (2004), no. 1, 111–130, 311.
198
[6] P. Deligne, and J. W. Morgan, Notes on supersymmetry (following Joseph Bernstein), Quan-
tum fields and strings: a course for mathematicians 1, 41-97, American Mathematical Society,
Providence, RI, 1999.
[7] H. Figueroa, and J. Gracia-Bonda, Combinatorial Hopf algebras in quantum field theory, I,
Rev. Math. Phys. 17 (2005), no. 8, 881–976.
[8] T. Hadfield and S. Majid, Bicrossproduct Approach to the Connes-Moscovici Hopf Algebra,
J. Algebra 312 (2007), 228-256.
[9] S. Majid, Foundations of quantum group theory, Cambridge University Press, Cambridge,
1995.
[10] H. Moscovici and B. Rangipour, Hopf cyclic cohomology and transverse characteristic classes,
To appear.
[11] H. Moscovici and B. Rangipour, Cyclic cohomology of Hopf algebras of transverse symmetries
in codimension 1, Adv. Math. 210 (2007), no. 1, 323–374.
[12] H. Moscovici and B. Rangipour, Hopf algebras of primitive Lie pseudogroups and Hopf cyclic
cohomology, Adv. Math. 220 (2009), no. 3, 706–790.
[13] A. Rogers, Supermanifolds. Theory and applications, World Scientific Publishing Co. Pte.
Ltd., Hackensack, NJ, 2007.
[14] V. S. Varadarajan, Supersymmetry for mathematicians: an introduction, Courant Lecture
Notes in Mathematics, 11. New York University, Courant Institute of Mathematical Sciences,
New York; American Mathematical Society, Providence, RI, 2004.
Department of Mathematics, University of Western Ontario, London, Ontario,

Canada, N6A 5B7
E-mail address: masoud@uwo.ca
Department of Mathematics, University of Western Ontario, London, Ontario,

Canada, N6A 5B7
E-mail address: apourkia@uwo.ca
Volume 546, 2011
Analytic Torsion of Z2 -graded Elliptic Complexes
Varghese Mathai and Siye Wu
Abstract. We define analytic torsion of Z2 -graded elliptic complexes as an

element in the graded determinant line of the cohomology of the complex, gen-
eralizing most of the variants of Ray-Singer analytic torsion in the literature. It
applies to a myriad of new examples, including flat superconnection complexes,
twisted analytic and twisted holomorphic torsions, etc. The definition uses
pseudo-differential operators and residue traces. We also study properties of
analytic torsion for Z2 -graded elliptic complexes, including the behavior under
variation of the metric. For compact odd-dimensional manifolds, the analytic
torsion is independent of the metric, whereas for even-dimensional manifolds,
a relative version of the analytic torsion is independent of the metric. Finally,
the relation to topological field theories is studied.
Introduction
In [13], we investigated the analytic torsion for the twisted de Rham complex
(Ω • (X, E), dE + H ∧ ·), where E is a vector bundle with a flat connection dE and H
is a closed differential form of odd degree on a closed compact oriented manifold X.
The novel feature of our construction was the necessary use of pseudo-differential
operators and residue traces in defining the torsion. When X is odd dimensional,
we showed that it was independent of the choice of metric. In this paper, we
generalize this construction, by defining analytic torsion for an arbitrary Z2 -graded
elliptic complex as an element in the graded determinant line of the cohomology
of the complex. The definition again uses pseudo-differential operators and residue
traces. We also study properties of analytic torsion for Z2 -graded elliptic complexes,
including its behavior under variation of the metric. For compact odd-dimensional
manifolds, the analytic torsion is independent of the metric, whereas for even-
dimensional manifolds, only a relative version of the analytic torsion is independent
of the metric.
We specialize this construction to several new situations where the analytic
torsion can be defined. This includes the case of flat superconnection complexes
and the analytic torsion of the twisted Dolbeault complex (Ω 0,• (X, E), ∂¯E + H ∧ ·),
2010 Mathematics Subject Classification. Primary 58J52; Secondary 58J10, 35K08, 57R56.
Key words and phrases. Analytic torsion, elliptic complex, superconnection, twisted holo-
morphic torsion, Calabi-Yau manifolds, topological field theories.
V.M. and S.W. are supported in part by the Australian Research Council. S.W. is supported
in part by the Research Grants Council of Hong Kong. We thank the referee for useful comments.
1
199
200
2 VARGHESE MATHAI AND SIYE WU
¯
where E is a holomorphic vector bundle and H is a ∂-closed differential form of type
(0, odd) on a closed connected complex manifold X. When H is zero, this was first
studied by Ray and Singer in [17]. Although the definition depends on a choice
of Hermitian metric, we deduce from our general theory that a relative version of
torsion, defined as a ratio of the twisted holomorphic torsions, is independent of the
metric. (Of course they do depend on the complex structure.) Twisted holomorphic
torsion is defined in several natural situations including for Calabi-Yau manifolds,
or whenever there is a holomorphic gerbe.
Finally, we explain how twisted analytic torsion appears in topological field
theory with a twisted abelian Chern-Simons action functional.
For a more detailed literature review on analytic torsion and its variants, we
refer to the introduction in [13].
We briefly summarize the contents of the paper. §1 is on the definition of
Z2 -graded elliptic complexes. §2 provides the definition of the analytic torsion
of a Z2 -graded elliptic complex as an element in the graded determinant line of
the cohomology of the complex. §3 contains functorial properties of the analytic
torsion. §4 establishes the invariance of the analytic torsion under deformation of
metrics in the odd-dimensional case. §5 shows the invariance of the relative analytic
torsion under deformation of metrics in the even-dimensional case. §6 contains the
definition and properties of analytic torsion of flat superconnections. §7 contains
the definition and properties of the analytic torsion of twisted Dolbeault complexes.
§8 relates the twisted analytic torsion to topological field theories.
1. Z2 -graded elliptic complexes

Let X be a smooth closed manifold of dimension n and E = E0̄ ⊕ E1̄ a Z2 -
graded vector bundle over X. (We use k̄ to denote the integer k modulo 2.)
Suppose D : Γ (X, E) → Γ (X, E) is an elliptic differential operator which is odd
with respect to the grading in E and satisfies D2 = 0. Then D is of the form
D = D D1̄ on Γ (X, E) = Γ (X, E0̄ ) ⊕ Γ (X, E1̄ ), where D0̄ : Γ (X, E0̄ ) → Γ (X, E1̄ )
0̄
and D1̄ : Γ (X, E1̄ ) → Γ (X, E0̄ ) are differential operators such that D1̄ D0̄ = 0 and
D0̄ D1̄ = 0. Furthermore,
D D D
· · · → Γ (X, E0̄ ) −→
0̄
Γ (X, E1̄ ) −→
1̄
Γ (X, E0̄ ) −→
0̄
Γ (X, E1̄ ) → · · ·
is a Z2 -graded elliptic complex, which we denote by (E, D) for short. Its cohomology
groups are
H 0̄ (X, E, D) = ker D0̄ / im D1̄ , H 1̄ (X, E, D) = ker D1̄ / im D0̄ .

It follows from the Hodge theorem for elliptic complexes, as will be explained
shortly, that they are finite dimensional. We call
b0̄ (X, E, D) = dim H 0̄ (X, E, D), b1̄ (X, E, D) = dim H 1̄ (X, E, D)

the Betti numbers of the Z2 -graded elliptic complex. Its index or Euler character-
istic is χ(X, E, D) = b0̄ (X, E, D) − b1̄ (X, E, D).
h We choose a 0̄Riemannian metric g on X and an Hermitian form of type h =
0̄
h on E = E ⊕E 1̄
. Then there is a scalar product ·, · on Γ (X, E). The Lapla-
1̄
cian L = D† D + DD† on Γ (X, E) = Γ (X, E0̄ ) ⊕ Γ (X, E1̄ ) is, in graded components,
ANALYTIC TORSION OF Z2 -GRADED ELLIPTIC COMPLEXES 201
3
L
L= 0̄
L1̄ , where
L0̄ = D0̄† D0̄ + D1̄ D1̄† , L1̄ = D1̄† D1̄ + D0̄ D0̄† .
They are self-adjoint elliptic operators with positive-definite leading symbols. By
the Hodge theorem for elliptic complexes, one has
H 0̄ (X, E, D) ∼
= ker L0̄ , H 1̄ (X, E, D) ∼
= ker L1̄ .
By ellipticity, these
spaces are finite dimensional, and hence b0̄ , b1̄ are finite.
K0̄ (t,x,y)
Let K(t, x, y) = K1̄ (t,x,y) , where t > 0, x, y ∈ X, be the kernel of
−tL
e−tL0̄
e = e−tL1̄ . Suppose the order of L (or that of L0̄ and L1̄ ) is d > 1.
By Lemma 1.7.4 of [7], when restricted to the diagonal, the heat kernel has the
asymptotic expansion
∞
2l−n
K(t, x, x) ∼ t d al (x),
l=0

al,0̄ (x)
where al (x) = al,1̄ (x) can be computed locally as a combinatorial expres-
sion in the jets of the symbols. We set a n2 (x) = 0 if n is odd. Denote by a n2 =
a n a n (x)
a n ,1̄ the operator acting on Γ (X, E) pointwise by a n
,0̄ ,0̄
2
2
(x) = 2
a n ,1̄ (x) .
2 2
Then the index density of the elliptic complex is str a n2 (x) and the index is χ(X, E) =
Str(a n2 ). Here and subsequently, str is the pointwise supertrace whereas Str is the
supertrace taken on the space of sections.
2. Definition of the analytic torsion

We generalize the construction in §2 of [13]. Recall that the zeta-function of a
semi-positive definite self-adjoint operator A (whenever it is defined) is
ζ(s, A) = Tr A−s ,
where Tr stands for the trace restricted to the subspace orthogonal to ker(A). If
ζ(s, A) can be extended meromorphically in s so that it is holomorphic at s = 0,
then the zeta-function regularized determinant of A is

Det A = e−ζ (0,A)
.
If A is an elliptic differential operator of order d on a compact manifold X of
dimension n, then ζ(s, A) is holomorphic when Re(s) > n/d and can be extended
meromorphically to the entire complex plane with possible simple poles at { n−l m ,l =
0, 1, 2, . . . } only [20]. Moreover, the extended function is holomorphic at s = 0 and
therefore the determinant Det A is defined for such an operator.
We return to the setting of the Z2 -graded elliptic complex (E, D) in §1. As in
†
D0̄ D0̄
[13], we consider the partial Laplacian D† D = †
D D
.
1̄ 1̄
Proposition 2.1. For k = 0, 1, ζ(s, Dk̄† Dk̄ ) is holomorphic in the half-plane

for Re(s) > n/d and extends meromorphically to C with possible simple poles at
d , l = 0, 1, 2, . . . } and possible double poles at the negative integers only, and is
{ n−l
holomorphic at s = 0.
202
P
Proof. Let P = 0̄
P1̄ be the projection onto the closure of im D† = im D0̄† ⊕
im D1̄† . As DD† and L are equal and invertible on (the closure of) im D, we have
P = D† (DD† )−1 D = D† L−1 D,
which implies that P (and hence P0̄ , P1̄ ) is a pseudo-differential operator of order
0. Moreover, for k = 0, 1,
ζ(s, Dk̄† Dk̄ ) = Tr(Pk̄ L−s
k̄
).
By general theory [10, 9], ζ(s, Dk̄† Dk̄ ) is holomorphic in the half-plane Re(s) >
n/d and extends meromorphically to C with possible simple poles at { n−l d , l =
0, 1, 2, . . . } and possible double poles at the negative integers only. The Laurent
series of ζ(s, Dk̄† Dk̄ ) at s = 0 is
∞
c−1 (Pk̄ , Lk̄ )
Tr(Pk̄ L−s
k̄
)= + c0 (Pk̄ , Lk̄ ) + cl (Pk̄ , Lk̄ ) sl .
s
l=1
1
Here c−1 (Pk̄ , Lk̄ ) = res(Pk̄ ), where res(Pk̄ ) is known as the non-commutative
d
residue of Pk̄ [21, 11]. Since Pk̄ is a projection, res(Pk̄ ) = 0 [21, 3, 8]. Therefore
ζ(s, Dk̄† Dk̄ ) is regular at s = 0.
The scalar product on Γ (X, Ek̄ ) restricts to one on the null space of the Lapla-
cian, ker(Lk̄ ) ∼
bk̄
= H k̄ (X, E, D). For k = 0, 1, let {νk̄,i }i=1 be an oriented orthonormal
basis of H k̄ (X, E, D) and let ηk̄ = νk̄,1 ∧ · · · ∧ νk̄,bk̄ , the unit volume element. Then
η0̄ ⊗ η1̄−1 ∈ det H • (X, E, D).
Definition 2.2. The analytic torsion of the Z2 -graded elliptic complex (E, D)
is
τ (X, E, D) = (Det D0̄† D0̄ )1/2 (Det D1̄† D1̄ )−1/2 η0̄ ⊗ η1̄−1 ∈ det H • (X, E, D).
3. Functorial properties of the analytic torsion

We summarize some properties of the analytic torsion of Z2 -graded elliptic
complexes, generalizing those of the Ray-Singer torsion [16] and of the torsion of
the twisted de Rham complex [13]. We omit the proofs as they are similar.
Suppose X is a compact, closed, oriented Riemannian manifold and E1 , E2 are
two Z2 -graded Hermitian vector bundles over X. Then E1 ⊕ E2 is also a Z2 -graded
vector bundle with (E1 ⊕ E2 )k̄ = Ek̄1 ⊕ Ek̄2 for k = 0, 1. If (E1 , D1 ) and (E2 , D2 ) are
two Z2 -graded elliptic complexes on X, then so is the direct sum (E1 ⊕ E2 , D1 ⊕ D2 )
defined in the obvious way. We have the following
Proposition 3.1. Under the natural identification of determinant lines,
τ (X, E1 ⊕ E2 , D1 ⊕ D2 ) = τ (X, E1 , D1 ) ⊗ τ (X, E2 , D2 ).
Now suppose p : X → X is a covering of compact, closed, oriented manifolds
(with finite index). Choose a Riemannian metric on X , which pulls back to one on
X. Let E → X be a Z2 -graded Hermitian
vector bundle. Then the vector bundle
p∗ E → X defined by (p∗ E)x = x∈p−1 (x ) Ex (for x ∈ X ) is also Z2 -graded and
has an induced Hermitian form. There is a natural isometry Γ (X, E) ∼ = Γ (X , p∗ E).
If D is a differential operator on Γ (X, E), the operator p∗ D on Γ (X , p∗ E) given
5
by the above isomorphism is a differential operator on X . If (E, D) is a Z2 -graded

elliptic complex, then so is (p∗ E, p∗ D). We have
τ (X, E, D) = τ (X , p∗ D).
Finally, suppose Xi (i = 1, 2) are closed, oriented Riemannian manifolds and
Ei → Xi (i = 1, 2) are Z2 -graded Hermitian vector bundles. Denote by πi : X1 ×
X2 → Xi (i = 1, 2) the projections. Set E1 E2 = π1∗ E1 ⊗ π2∗ E2 ; it is also a Z2 -
graded vector bundle with (E1 E2 )0̄ = π1∗ E0̄1 ⊗π2∗ E0̄2 ⊕π1∗ E1̄1 ⊗π2∗ E1̄2 and (E1 E2 )1̄ =
π1∗ E0̄1 ⊗ π2∗ E1̄2 ⊕ π1∗ E1̄1 ⊗ π2∗ E0̄2 . If (E1 , D1 ) and (E2 , D2 ) are two Z2 -graded elliptic
complexes, then so is (E1 E2 , D1 D2 ), where the operator D1 D2 acts on
Γ (X1 × X2 , E1 E2 ) according to
(D1 D2 )(π1∗ s1 ⊗ π2∗ s2 ) = π1∗ (D1 s1 ) ⊗ π2∗ s2 + (−1)|s1 | π1∗ s1 ⊗ π2∗ (D2 s2 )
for any si ∈ Γ (Xi , Ei ), i = 1, 2. We have
τ (X1 ×X2 , E1 E2 , D1 D2 ) = τ (X1 , E1 , D1 )⊗χ(X2 ,E2 ,D2 ) ⊗τ (X2 , E2 , D2 )⊗χ(X1 ,E1 ,D1 ) .
4. Invariance of the torsion under deformation of metrics: the

odd-dimensional case
We note that the operator D, or D0̄ and D1̄ , in the Z2 -graded elliptic complex
(E, D) are not dependent on the metric. However, the corresponding partial Lapla-
cians D† D or D0̄† D0̄ , D1̄† D1̄ do depend on the metric, and therefore a priori, so does
the analytic torsion τ (X, E, D). In this section, we study the dependence of the
analytic torsion on the metrics and prove that for closed, oriented odd-dimensional
manifolds X, the torsion τ (X, E, D) is independent of the choice of metric.
Suppose we change the metric g on X and the Hermitian form h on E to gu
and hu , respectively, along a path parameterized by u ∈ R. Although the torsion
τ (X, E, D) is an element of the determinant line det H • (X, E), its variation
∂ ∂
log τ (X, E, D) = τ (X, E, D)−1 τ (X, E, D)
∂u ∂u
makes sense as a function of u. For any u, the Hermitian structure ·, ·u on Γ (X, E)
is related to the undeformed one by
· , · u = Γu (·), ·

for some invertible operator Γu = Γ0̄ Γ1̄ . Let αu = Γu−1 ∂u
∂
Γu . We have the
following
Theorem 4.1. Under the above deformation of g and h, we have
∂
log τ (X, E, D) = Str α a n2 .
∂u
In particular, the above is zero if dim X = n is odd. In this case, the analytic
torsion τ (X, E, D) is independent of the choice of metric.
204
Proof. We generalize the proof of Lemma 3.1 in [13]. The adjoint of D with
respect to ·, ·u is Du† = Γu−1 D† Γu . Its variation is given by
∂ †
D = −[αu , Du† ].
∂u u
In graded components, this is
∂D0̄ ∂D1̄
= −α1̄ D0̄ + D0 α0̄ , = −α0̄ D1̄ + D1 α1̄ ,
∂u ∂u
where α0̄ = Γ0̄−1 ∂u∂
Γ0̄ , α1̄ = Γ1̄−1 ∂u
∂
Γ1̄ . Following [16, 13], we set
∞
†
f (s, u) = ts−1 Str(e−tD D P ) dt
0

= Γ(s) ζ(s, D0̄† D0̄ ) − ζ(s, D1̄† D1̄ ) ,
omitting the dependence on u on the right-hand side. Then, as P ∂P
∂u = 0,
∞
∂f † † ∂P
= ts−1 Str t[α, D† ]De−tD D + P e−tD D dt
∂u ∂u
∞
0
† † † ∂P
= ts−1 Str tα e−tD D D† D + e−tDD DD† + e−tD D P dt
∂u
0 ∞

= ts Str αe−tL L dt
0
∞
∂
=− ts Str α(e−tL − Q) dt
∂t
0∞

=s ts−1 Str α(e−tL − Q) dt.
0
Here Q denotes the orthogonal projection onto the nullspace of L. By the asymp-
1
totic expansion of Str(αe−tL ) as t ↓ 0, 0 ts−1 Str(αe−tL ) dt has a possible first
order pole at s = 0 with residue Str(αa n2 ). On the other hand, because of the
∞
exponential decay of Str α(e−tL − Q) for large t, 1 ts−1 Str α(e−tL − Q) dt is
an entire function in s. So
∂f

= − Str α(a n2 − Q)
∂u s=0
is finite and hence
∂
ζ(0, D0̄† D0̄ ) − ζ(0, D1̄† D1̄ ) = 0.
∂u
Since
Det D† D 1
log
0̄ 0̄
†
= − lim f (s, u) − ζ(0, D0̄† D0̄ ) − ζ(0, D1̄† D1̄ ) ,
Det D D1̄ s→0 s
1̄
we get
∂ Det D† D
0̄ 0̄
log †
= Str α(a n2 − Q) .
∂u Det D D1̄
1̄
Finally, along the path of deformation, the volume elements η0̄ , η1̄ can be chosen
so that (cf. Lemma 3.3 of [13])
∂ 1
(η ⊗ η1̄−1 ) = − Str(αQ) η0̄ ⊗ η1̄−1 .
∂u 0̄ 2
The results then follow.
7
When the elliptic complex is the (Z-graded) de Rham complex of differential

forms with values in a flat vector bundle, the variation of the torsion can be inte-
grated to an anomaly formula [2].
5. Invariance of relative torsion under deformation of metrics: the

even-dimensional case
When n = dim X is even, the torsion does depend on the metrics g on X and
h on E (Theorem 4.1). However, we will prove that the relative analytic torsion
defined below is independent of the choice of metric.
We first explain extension by flat bundles. Let Π = π1 (X) be the fundamental
group of X and let ρ : Π → GL(m, C) be a representation of Π . Then ρ determines
a flat bundle Fρ over X given by

Fρ = X × Cm / ∼, (xγ, v) ∼ (x, ρ(γ)v),
where X is the universal cover of X. Smooth sections of Fρ are smooth maps

s : X → Cm that are Π -equivariant, i.e., s ◦ γ = ρ(γ)s for all γ ∈ Π . We want to
extend D to an action on the sections of Eρ = E ⊗ Fρ . Since D is a differential
operator, it lifts to the universal cover X as a Π -periodic operator D : Γ (X, →
E)

Γ (X, E), where E is the pull-back of E to X. By tensoring with the identity operator
on Cm , we can extend it to D : Γ (X,
E ⊗ Cm ) → Γ (X,
E ⊗ Cm ). Since for any Π -
equivariant section s ∈ Γ (X, ⊗ C ),
E m
◦ γ = D(s
(Ds) ◦ γ) = D(ρ(γ)s)

= ρ(γ)(Ds),
the operator D descends to a differential operator Dρ : Γ (X, Eρ ) → Γ (X, Eρ ). If
(E, D) is a Z2 -graded elliptic complex, then so is (Eρ , Dρ ).
Now suppose X is a closed, compact, oriented Riemannian manifold and E is
a Z2 -graded Hermitian vector bundle. Let ρ1 , ρ2 be unitary representations of Π
of the same dimension m. Then the flat bundles Fρi are Hermitian bundles and so
are Eρi = E ⊗ Fρi (i = 1, 2). Furthermore, if (E, D) is a Z2 -graded elliptic complex
as in §1, then so are (Eρi , Dρi ) for i = 1, 2.
Definition 5.1. The relative analytic torsion is the quotient
τ (X, Eρ1 , Dρ1 ) ⊗ τ (X, Eρ2 , Dρ2 )−1 ∈ det H • (X, Eρ1 , Dρ1 ) ⊗ det H • (X, Eρ2 , Dρ2 )−1 .
To show its independence of the metric, let Kρi (t, x, y) denote, for i = 1, 2,
the heat kernel of the Laplacians Lρi = Dρ†i Dρi + Dρi Dρ†i . Since the Hermitian
bundles Eρ1 and Eρ2 , together with the differential operators Dρ1 and Dρ2 are
locally identical, the difference in the two heat kernels, when restricted to the
diagonal, is exponentially small for small t. More precisely, we have
Proposition 5.2. In the notation above, there are positive constants C, C
such that as t → 0, one has for all x ∈ X,
|Kρ1 (t, x, x) − Kρ2 (t, x, x)| ≤ Ct−n/d exp[−C t− d−1 ],
d
where d is the order of the Laplacians.

Proof. Let K(t, x, y) denote the heat kernel of the Laplacian L
=D
†D
+D
D †

on X. Then, by the Selberg principle, one has for x, y ∈ X,

Kρj (t, x̄, ȳ) = x, yγ)ρj (γ),
K(t,
γ∈Π
206
where x̄ ∈ X stands for the projection of x ∈ X. It follows that

Kρ1 (t, x̄, ȳ) − Kρ2 (t, x̄, ȳ) = x, yγ)(ρ1 (γ) − ρ2 (γ)).
K(t,
γ∈Π \{1}
Since ρi (i = 1, 2) are unitary representations, one has

|Kρ1 (t, x̄, ȳ) − Kρ2 (t, x̄, ȳ)| ≤ x, yγ)|.
2|K(t,
γ∈Π \{1}
is [5]
The off-diagonal Gaussian estimate for the heat kernel on X
d
x, y)| ≤ C1 t−n/d exp − C2 d(x, y) d−1 ,
|K(t,
t
Therefore
where d(x, y) is the Riemannian distance between x, y ∈ X.
d(x, xγ) d−1
d
|Kρ1 (t, x̄, x̄) − Kρ2 (t, x̄, x̄)| ≤ 2C1 t−n/d exp − C2 .
t
γ∈Π \{1}
By Milnor’s theorem [14], there is a positive constant C3 such that d(x, xγ) ≥
C3 (γ), where denotes a word metric on Π . Moreover, the number of elements in
the sphere Sl of radius l in Π satisfies #Sl ≤ C4 eC5 l for some positive constants
C4 , C5 . Therefore
d(x, xγ) d−1
d
exp − C2
t
γ∈Π \{1}
d
≤ exp − C ((γ)/t) d−1
γ∈Π \{1}
∞
d
≤ exp − C (l/t) d−1 C4 eC5 l
l=1
∞

≤ C4 exp[−C t− d−1 ] exp − C (l d−1 − 1) + C5 l
d d
l=1
for all t such that 0 < t ≤ 1 for some positive constant C . Since d
d−1 > 1, the
infinite sum over l converges and hence the result.
Theorem 5.3. Let X be a closed oriented manifold of even dimension. Let

ρ1 , ρ2 be unitary representations of π1 (X) of the same dimension. Then the relative
analytic torsion τ (X, Eρ1 , Dρ1 ) ⊗ τ (X, Eρ2 , Dρ2 )−1 is independent of the choice of
metric.
Proof. By Theorem 4.1, under a one-parameter deformation of the metric,

∂
log τ (X, Eρi , Dρi ) = Str α aρni )
∂u 2
for i = 1, 2. By Proposition 5.2, we have aρn1 = aρn2 . Therefore the change in relative
2 2
torsion is zero.
9
6. Analytic torsion of flat superconnections

The concept of superconnection was initiated by Quillen, cf. [15, 12, 1]. Let
X be a smooth manifold and let F = F0̄ ⊕ F1̄ be a Z2 -graded vector bundle over
X. Then the space Ω (X, F) of F-valued differential forms has a Z2 -grading with
Ω (X, F)0̄ = Ω 0̄ (X, F0̄ ) ⊕ Ω 1̄ (X, F1̄ ), Ω (X, F)1̄ = Ω 0̄ (X, F1̄ ) ⊕ Ω 1̄ (X, F0̄ ).
A superconnection is a first-order differential operator A on Ω (X, F) that is odd

with respect to the Z2 -grading and satisfies
A(α ∧ s) = dα ∧ s + (−1)|α| α ∧ As
for any α ∈ Ω (X) and s ∈ Ω (X, F). The bundle End(F) is also Z2 -graded and
A extends to Ω (X, End(F)). The curvature of the superconnection is FA = A2 ∈
Ω (X, End(F))0̄ . It satisfies the Bianchi identity AFA = 0. A superconnection A is
of the form A = ∇ + A, where ∇ is a usual connection on F preserving the grading
and A ∈ Ω (X, End(F))1̄ . Thus the superconnections form an affine space modeled
on the vector space Ω (X, End(F))1̄ .
The superconnection is flat if FA = 0. In this case, writing A = A0̄ A1̄ , there
is a Z2 -graded elliptic complex
A A A
· · · → Ω (X, F)0̄ −→
0̄
Ω (X, F)1̄ −→
1̄
Ω (X, F)0̄ −→
0̄
Ω (X, F)1̄ → · · · ,
groups H (X, F, A), k = 0, 1. In fact, this is a

k̄
We can define the cohomology
special case of §1 with E = T X ⊗ F and D = A. If X is a closed, compact,
oriented Riemannian manifold and F is an Hermitian vector bundle, then we can
define the analytic torsion of a flat superconnection as τ (X, F, A) = τ (X, E, D) ∈
det H • (X, F, A) with the above choice of (E, D). The functorial properties (§3) and
invariance under metric deformations (§4, 5) hold in this case.
We consider a special case when F = F0̄ and F1̄ = 0. Then Ω (X, F)k̄ =
Ω (X, F) for k = 0, 1. A superconnection is of the form ∇ + A, where ∇ is a
k̄
usual connection on F and A ∈ Ω 1̄ (X, End(F)). Suppose A is of degree 3 or

higher. Then the superconnection is flat if and only if ∇ is flat and ∇A + A2 = 0.
When A is of the form A = H idF for some H ∈ Ω 1̄ (X), the above condition on
A becomes dH = 0 and the Z2 -graded elliptic complex is the twisted de Rham
complex (Ω (X), d + H ∧ ·). Its analytic torsion τ (X, F, H) was studied in [13].
Among other properties, the latter is also invariant under the deformation of H by
an exact form when X is odd dimensional; the rest of the section will be devoted
to generalizing this property to the analytic torsion of flat superconnections.
We return to the general case of a flat superconnection A over a graded vector
bundle F. Suppose G ∈ Ω (X, End(F))0̄ is invertible pointwise. Then A = G−1 AG
is another flat superconnection on F; we say that A is gauge equivalent to A. There
is an isomorphism of cohomology groups H • (X, F, A) ∼ = H • (X, F, A ), and hence of
the corresponding determinant lines, induced by G. Now suppose A is deformed to
Av along a path parameterized by v so that each Av is gauge equivalent to A via
Gv . Let
∂Gv
βv = G−1
v ∈ Ω (X, End(F))0̄ .
∂v
208
Theorem 6.1. Under deformation of A by gauge equivalence and the natural

identification of determinant lines, we have
∂
log τ (X, F, A) = Str β a n2 .
∂v
If dim X = n is odd, then the above is zero. In this case, the analytic torsion
τ (X, F, A) is invariant under gauge equivalence.
Proof. Under the deformation, we have
∂A ∂A†
= [β, A], = −[β † , A† ].
∂v ∂v
The component of β in Ω 1̄ (X, End(F)1̄ ) does not contribute to the trace or super-
trace, whereas that in Ω 0̄ (X, End(F)0̄ ) is even in the degree of differential forms.
Following the proofs of Lemmas 3.5 and 3.7 of [13], we get the desired variation
formula upon a suitable choice of volume elements and identification of determinant
lines under the deformation; the rest follows easily.
If dim X is even, a relative version of analytic torsion (cf. §5) is invariant under
gauge equivalence.
7. Analytic torsion of twisted Dolbeault complexes

Let X be a connected, closed, compact complex manifold and let F be a holo-
morphic vector bundle over X. Denote by Ω p,q (X, F) the space of smooth differ-
ential (p, q)-forms on X with values in F. A holomorphic connection on F can act
on Ω p,q (X, F) and splits uniquely as ∂F + ∂¯F , where
∂F : Ω p,q (X, F) → Ω p+1,q (X, F), ∂¯F : Ω p,q (X, F) → Ω p,q+1 (X, F)
satisfying ∂¯F
2
= 0. This yields the Dolbeault complex of differential forms with
values in F.
Let Ω p,0̄ (X, F), Ω p,1̄ (X, F) be the space of differential forms that is of de-
gree p in the holomorphic part and of even, odd degree, respectively, in the anti-
holomorphic part. Consider a differential form H ∈ Ω 0,1̄ (X) that is ∂-closed, ¯ i.e.,
¯ = 0. Let ∂¯F,H = ∂¯F + H ∧ · . We call H a holomorphic flux form and ∂¯F,H
∂H
the Dolbeault operator twisted by H. Setting ∂¯k̄ = ∂¯F,H acting on Ω p,k̄ (X, E) for
k = 0, 1, we have ∂¯1̄ ∂¯0̄ = ∂¯0̄ ∂¯1̄ = 0 and a Z2 -graded elliptic complex, which we call
the twisted Dolbeault complex
∂¯ ∂¯ ∂¯
· · · → Ω p,0̄ (X, F) −→
0̄
Ω p,1̄ (X, F) −→
1̄
Ω p,0̄ (X, F) −→
0̄
Ω p,1̄ (X, F) → · · · .
We define the twisted Dolbeault cohomology groups as
H p,0̄ (X, F, H) = ker ∂¯0̄ / im ∂¯1̄ , H p,1̄ (X, F, H) = ker ∂¯1̄ / im ∂¯0̄ .
As in [18, 13], if the degree of H is 3 or higher, there is a spectral sequence whose
E2 -terms are H p,• (X, F) converging to H p,• (X, F, H). If H and H differ by a ∂-
¯
∼
exact form, then there are natural isomorphisms H (X, F, H ) = H (X, F, H).
p,• p,•
The above construction is the holomorphic counterpart of the twisted de Rham

complex studied in [18, 13]. Holomorphic flux forms arise naturally in a number
of prominent situations. Suppose that X is a Calabi-Yau manifold of odd complex
dimension n. Then the canonical bundle of X is trivial, i.e., there is a nowhere zero
section Ω which satisfies ∂Ω = 0. Here H = Ω ∈ Ω0,n (X) is ∂-closed. ¯ Another
example comes from holomorphic gerbes (or holomorphic sheaves of groupoids).
11
The 3-curvature of a holomorphic curving on a holomorphic gerbe on a complex

manifold X is a closed holomorphic 3-form Ω on X (cf. [4], 5.3.17 part (4)). Again,
¯
H = Ω ∈ Ω 0,3 (X) is ∂-closed.
The twisted Dolbeault
p 1,0 complex • is0,1also∗ a special Z2 -graded elliptic complex
(E, D) with E = (T X)∗ ⊗ (T X) ⊗ F and D = ∂¯F,H . Suppose X is
closed and compact. Given a Riemannian metric on X and a Hermitian form on
F, we have the analytic torsion of the twisted Dolbeault complex (cf. §2)
τp (X, F, H) = τ (X, E, D) ∈ det H p,• (X, F, H)
with the above choice of (E, D). It isatisfies the functorial properties in §3. Since
X is always of even (real) dimension, only a relative version of the analytic torsion
for the twisted Dolbeault complex is independent of the metric. We conclude from
Theorem 5.3 the following
Corollary 7.1. Let F be a holomorphic vector bundle over a compact complex

¯
manifold X. Suppose H ∈ Ω 0,1̄ (X) is ∂-closed. For two flat bundles on X given
by the representations ρ1 , ρ2 of π1 (X) of the same dimension, the relative twisted
holomorphic torsion τ (X, Fρ1 , H) ⊗ τ (X, Fρ2 , H)−1 is invariant under any defor-
¯
mation of H by an ∂-exact form, up to natural identification of the determinant
lines.
For a non-trivial example of twisted holomorphic torsion, consider the compact

Calabi-Yau manifold T × M , where T is a compact complex torus of dimension
1 and M is a K3 surface. Let L =√Lu,v be a flat line bundle over T defined by
the character χu,v (m, n) = exp(2π −1(mu + nv)), 0 ≤ u, v ≤ 1, m, n ∈ Z. If
(m, n) = (0, 0), then the Dolbeault cohomology H • (T, L) is trivial. Recall that the
non-trivial holomorphic torsion of (T, L) as calculated by [17] is

√

2 θ1 (u − τ v, τ )
τ0 (T, L) =
eπ −1v τ
,

η(τ )
where τ (with Im τ > 0) is the complex modulus of T . Here the theta function is
defined as
√ ∞
√
π −1(w+τ /6) −1(|k|τ −
k w)
θ1 (w, τ ) = −η(τ )e (1 − e2π ),
k=−∞

where k = sign k + 12 and η(τ ) is the Dedekind eta function. We still denote by
L the pull-back of L to T × M . The Dolbeault cohomology groups of (T × M, L)
are trivial, and so are the H-twisted ones. Since χ(OT (L)) = 0 and χ(OM ) = 2, we
have [17]
τ0 (T × M, L) = τ0 (T, L)⊗2 .
Let H = ᾱ ∧ λ̄, where α is a holomorphic 1-form on T and λ a holomorphic 2-form
on M . By perturbation theory [6], one has,
τ0 (T × M, L, H) = eo(|H|) τ0 (T × M, L) = eo(|H|) τ0 (T, L)⊗2 ,
where o(|H|) → 0 as H → 0. Therefore τ0 (T × M, L, H) is non-trivial whenever

|H| is sufficiently small.
210
8. Relation to topological field theories

In [19], a topological field theory of antisymmetric tensor fields was constructed
and the partition function was shown to be equal to the Ray-Singer analytic torsion.
The metric independence of the torsion is evidence that the quantized theory is
topologically invariant. In this section, we extend the relation to twisted analytic
torsion by constructing topological field theories that contain a coupling with the
flux form.
Suppose X is a compact, oriented manifold of dimension n and H is a flux
form, a closed differential form of odd degree. For k = 0 or 1, we define a theory
whose action is
Sk̄ [B, C] = B ∧ dH C,
X
where B ∈ Ω n−k (X), C ∈ Ω k̄ (X) are the dynamical fields. Since the operator
dH = d + H ∧ · is not compatible with the Z-grading, the forms B, C cannot be
chosen to have fixed degrees. Instead, the degrees of B, C have the same parity
when dim X is odd and opposite parity when dim X is even. The classical equations
of motion are
dH C = 0, d−H B = 0.
The action S[B, C] and the equations of motion are invariant under a set of gauge
transformations
C → C + dH C (1) , B → B + d−H B (1) ,
where B (1) , C (1) can be any forms whose degrees have opposite parity to B, C,
respectively. The phase space is the space of solutions to the equation of motion
modulo the gauge transformations. In this case, it is H n−k (X, −H) ⊕ H k̄ (X, H),
expressed in terms of the de Rham cohomology groups twisted by the fluxes ±H.
To quantize the theory, we consider the partition function

Zk̄ (X, H) = DBDC e−Sk̄ [B,C] .
We need to introduce a Riemannian metric on X which determines the “measures”

DB, DC. The integration of the transverse parts of B, C yields the determinant
Det (d†H dH )−1/2 (defined in §2 of [13]); that of the zero modes contributes volume
elements on the cohomology groups. The longitudinal modes of B, C are treated by
adding Faddeev-Popov ghost fields which contribute to determinant factors in the
numerator, and there are secondary and higher ghosts since B (1) , C (1) themselves
contain redundancies.
We consider a special case when dim X = 2l + 1 is odd and H is a top-degree
form (cf. §5.1 of [13]). If B, C ∈ Ω 1̄ (X), then B ∧ dH C = BdC, and the theory is
equivalent to an untwisted theory. We now assume that B, C ∈ Ω 0̄ (X). Then the
bosonic determinant from the integration of the transverse modes is
† † −1/2
d0 d0 + H H H † d2l l−1
Det (Det d†2i d2i )−1/2 ,
d†2l H d†2l d2l i=1
where di is d on Ω i (X) for 0 ≤ i ≤ 2l + 1. The crucial feature in this case is that

H does not appear in the gauge transformations
B → B + dB (1) , C → C + dC (1) .
13
Moreover, we can choose B (1) , C (1) ∈ Ω 1̄ (X) to be of degree 2l − 1 or less. Further

redundancies in B (1) , C (1) are described by a hierarchy of gauge transformations
B (i) → B (i) + dB (i+1) , C (i) → C (i) + dC (i+1) ,
where B (i) , C (i) ∈ Ω ī (X) are of degree 2l − i or less, for 1 ≤ i ≤ 2l − 1. The
Faddeev-Popov procedure yields the determinant factors
2l
(−1)i+1
Det (d†2l−i d2l−i ) Det (d†2l−i−2 d2l−i−2 ) · · ·
i=0

l
l−1
= Det (d†2i d2i )−l/2 Det (d†2i+1 d2i+1 )(l+1)/2 .
i=0 i=0
Taking into account the contribution of the zero modes, the partition function is
Z0̄ (X, H) = τ (X, H)−1 ⊗ τ (X)⊗(−l) ∈ det H • (X, H)−1 ⊗ det H • (X)⊗(−l) .
Here τ (X) ∈ det H • (X) is the classical Ray-Singer torsion [16]. The independence
of the partition on the metric indicates that the quantum theory is also metric
independent although it is necessary to use a metric in the definition.
It would be interesting to construct topological field theories when the flux
form H is of an arbitrary degree, when the manifold has a boundary [22], and
those related to the analytic torsion of other Z2 -graded elliptic complexes such as
the twisted Dolbeault complex.
References
[1] N. Berline, E. Getzler and M. Vergne, Heat kernels and Dirac operators, Grund. Math.
Wissen., 298, Springer-Verlag, Berlin, 1992.
[2] J.-M. Bismut and W. Zhang, Métriques de Reidemeister et métriques de Ray-Singer sur le
déterminant de la cohomologie d’un fibré plat: une extension d’un résultat de Cheeger et
Müller, C. R. Acad. Sci. Paris Sér. I Math. 313 (1991) 775-782; An extension of a theorem
by Cheeger and Müller, with an appendix by F. Laudenbach, Astérisque 205 (1992) 1-235.
[3] J. Brüning and M. Lesch, On the η-invariant of certain nonlocal boundary value problems,
Duke Math. J. 96 (1999) 425-468, [arXiv:dg-ga/9609001].
[4] J.-L. Brylinski, Loop spaces, characteristic classes and geometric quantization, Progress in
Mathematics, 107, Birkhäuser Boston, Inc., Boston, MA (1993).
[5] J. Brüning and T. Sunada, On the spectrum of periodic elliptic operators, Nagoya Math. J.
126 (1992) 159-171.
[6] M. Farber, Singularities of the analytic torsion, J. Diff. Geom. 41 (1995) 528-572.
[7] P.B. Gilkey, Invariance theory, the heat equation, and the Atiyah-Singer index theorem, Math.
Lecture Series, vol. 11, Publish or Perish, Inc., Wilmington, DE, 1984; 2nd ed., (Studies Adv.
Math.), CRC Press, Boca Raton, FL, 1995.
[8] G. Grubb, Spectral boundary conditions for generalizations of Laplace and Dirac operators,
Commun. Math. Phys. 240 (2003) 243-280, [arXiv:math/0302286].
[9] G. Grubb, A resolvent approach to traces and zeta Laurent expansions, in: B. Booß-Bavnbek,
G. Grubb and K.P. Wojciechowski (Eds.), Spectral geometry of manifolds with boundary and
decomposition of manifolds, (Roskilde Univ., 2003), Contemp. Math., vol. 366, Amer. Math.
Soc., Providence, RI, 2005, pp. 67-93, corrected in [arXiv:math/0311081v4].
[10] G. Grubb and R.T. Seeley, Weakly parametric pseudodifferential operators and Atiyah-
Patodi-Singer boundary problems, Invent. Math. 121 (1995) 481-529.
[11] V. Guillemin, A new proof of Weyl’s formula on the asymptotic distribution of eigenvalues,
Adv. Math. 55 (1985) 131-160.
[12] V. Mathai and D. Quillen, Superconnections, Thom classes, and equivariant differential forms,
Topology 25 (1986) 85-110.
212
[13] V. Mathai and S. Wu, Analytic torsion for twisted de Rham complexes, preprint (2008),
arXiv:0810.4204v3[math.DG].
[14] J. Milnor, A note on curvature and fundamental group, J. Diff. Geom. 2 (1968) 1-7.
[15] D. Quillen, Superconnections and the Chern character, Topology 24 (1985) 89-95.
[16] D.B. Ray and I.M. Singer, R-torsion and the Laplacian on Riemannian manifolds, Adv. Math.
7 (1971) 145-210; Analytic torsion, in: D.C. Spencer (Ed.), Partial differential equations
(Univ. California, Berkeley, Calif., 1971), Proc. Sympos. Pure Math., vol. XXIII, Amer.
Math. Soc., Providence, R.I., 1973, pp. 167-181.
[17] D.B. Ray and I.M. Singer, Analytic torsion for complex manifolds, Ann. Math. (2) 98 (1973)
154-177.
[18] R. Rohm and E. Witten, The antisymmetric tensor field in superstring theory, Ann. Phys.
170 (1986) 454-489.
[19] A.S. Schwarz, The partition function of degenerate quadratic functional and Ray-Singer in-
variants, Lett. Math. Phys. 2 (1977/78) 247-252; The partition function of a degenerate
functional, Commun. Math. Phys. 67 (1979) 1-16.
[20] R.T. Seeley, Complex powers of an elliptic operator, in: A.P. Calderón (Ed.), Singular Inte-
grals (Chicago, IL, 1966), Proc. Symp. Pure Math., vol. X, Amer. Math. Soc., Providence,
RI, 1967, pp. 288-307.
[21] M. Wodzicki, Local invariants of spectral asymmetry, Invent. Math. 75 (1984) 143-177.
[22] S. Wu, Topological quantum field theories on manifolds with a boundary, Commun. Math.
Phys. 136 (1991) 157-168.
Department of Mathematics, University of Adelaide, Adelaide 5005, Australia

E-mail address: mathai.varghese@adelaide.edu.au
Department of Mathematics, University of Colorado, Boulder, CO 80309, USA and

Department of Mathematics, University of Hong Kong, Pokfulam Road, Hong Kong
E-mail address: swu@math.colorado.edu swu@maths.hku.hk
Volume 546, 2011
The K-groups and the index theory of certain comparison

C ∗ -algebras
Bertrand Monthubert and Victor Nistor
Abstract. We compute the K-theory of the comparison C ∗ -algebra associ-

ated to a manifold with corners. These comparison algebras are an example of
the abstract pseudodifferential algebras introduced by Connes and Moscovici
[16]. Our calculation is obtained by showing that the comparison algebra as-
sociated to a manifold with corners is a homomorphic image of an explicit
groupoid C ∗ -algebra. We then prove an index theorem with values in the
K-theory groups of the comparison algebra.
Contents
Introduction
1. Groupoids and comparison algebras
2. The analytic index
3. Properties of the full C ∗ -analytic index
4. A topological index theorem
5. K-theory of comparison algebras
References
Introduction
The work of Henri Moscovici encompasses many areas of mathematics, most
notably Non-commutative Geometry, Group Representations, Geometry, and Ab-
stract Analysis. His work on Non-commutative geometry, mostly joint works with
Alain Connes, has lead to many breakthroughs in Index Theory and Operator Al-
gebras, as well as to applications to other areas. We are happy to dedicate this
paper to Henri Moscovici on the occasion of his 65th birthday.
The problem studied in this paper pertains to the general program of un-
derstanding index theory on singular and non-compact spaces. On such spaces,
Monthubert was partially supported by a ACI Jeunes Chercheurs. Manuscripts available from
http://www.math.univ-toulouse.fr/∼ Monthube. Nistor was partially supported by the
NSF Grant DMS 0555831. Manuscripts available from http://www.math.psu.edu/nistor/.
2011
c 0000
c Mathematical
holder)
1
213
214
2 B. MONTHUBERT AND V. NISTOR
the Fredholm property depends on more than the principal symbol, so cyclic co-
cycles are needed in order to obtain explicit index formulas. Moscovici has ob-
tained many results in this direction, including [6, 10, 15, 16, 40, 41]. See also
[7, 9, 12, 13, 24, 26, 27, 45].
One of the central concepts in a recent paper by Connes and Moscovici is that of
an abstract algebra of pseudodifferential operators [16]. These algebras generalize
similar algebras introduced earlier. In this paper, we would like to study certain
natural C ∗ -algebras associated to non-compact Riemannian manifolds, applying in
particular the point of view of the work of Connes and Moscovici mentioned above.
Let us now explain the framework of this paper. Let M0 be a complete Rie-
mannian manifold and let Δ = d∗ d be the positive Laplace operator on M0 associ-
ated to the metric. It is well known that Δ is essentially self-adjoint [18, 49] and
the references therein, and hence we can define Λ = (1 + Δ)−1/2 using functional
calculus. Let us also assume that a certain algebra D = ∪Dn of differential oper-
ators is given on M0 , where Dn denotes the space of differential operators in D of
degree at most n. Let us assume that Δ ∈ D2 and that Ln Λn defines a bounded
operator on L2 (M0 ) for any Ln ∈ Dn . Then the comparison algebra of M0 (and D)
is defined to be the C ∗ -algebra generated by the operators of the form Ln Λn for
any Ln ∈ Dn . This definition is almost the same as the one in [20, 17], where the
comparison algebra was defined as the C ∗ -algebra generated by all operators of the
form L1 Λ for any L1 ∈ D1 and all compact operators. One of our results, Theorem
4, implies that the two definitions are the same for suitable M0 . Let us denote the
comparison C ∗ -algebra of M0 by A(M0 ) (the dependence on the algebra D will be
implicit).
The comparison algebra A(M0 ) is a convenient tool to study many analytic
properties of differential operators on M0 , such as invertibility between Sobolev
spaces, spectrum, compactness, the Fredholm property, and the index [14, 20, 23,
29, 31, 50]. For instance, the principal symbol of order zero pseudodifferential
operators extends to a continuous map σ0 : A(M0 ) → C(S ∗ A) with kernel denoted
by A−1 (M0 ), where S ∗ A is a suitable compactification of the cosphere bundle of
T ∗ M0 .
In this paper, we concentrate on the index properties of elliptic operators on
a certain class of non-compact manifolds, called “manifolds with poly-cylindrical
ends.” Recall that a manifold with poly-cylindrical ends is, locally, a product of
manifolds with cylindrical ends. Our index depends only on the principal symbol,
so it takes values in the K-theory of the C ∗ -algebra A−1 (M0 ). In principle, one
therefore has to first compute the K-theory of the algebra A−1 (M0 ) and then to
compute the index map. That then leads in principle to a computation of the K-
groups of the algebra A(M0 ), up to an extension of groups, using the six-term exact
sequence in K-theory. In our case, however, the computation of the K-groups of
the algebra A−1 (M0 ) is not so simple, and in fact will be obtained also from an
index theorem in which the index map is an isomorphism (by replacing M with
one of its classifying spaces, in a sense that is explained in the last section). Index
calculations of this sort are sometimes necessary in applications, for instance in the
study of the Hartree equation [25] and in the study of boundary value problems on
polyhedral domains [32].
Let us now explain our approach. Let us first assume that the given manifold
M0 is the interior of the space of units of a Lie groupoid G. Then the Lie groupoid
COMPARISON C ∗ -ALGEBRAS 215
3
structure of G gives rise to a natural algebra of differential operators D on M0 ,

such as in the case of singular foliations [4, 14]. In general, there will be no natural
metric on M0 , and even if a metric is chosen on M0 , the associated Laplace operator
Δ ∈ D. However, if M0 is a Lie manifold [3], then a natural class of metrics exists on
M0 and Δ ∈ D for any metric in this class. Recall from [3] that M0 is a Lie manifold
if the tangent bundle T M0 extends to a bundle A → M on a compactification M
of M0 to a manifold with corners such that the space of smooth sections V := Γ(A)
of A has a natural Lie algebra structure induced by the Lie bracket of vector fields
and such that that the diffeomorphisms generated by vector fields in V preserve the
faces of M .
We shall show that the comparison algebra of a Lie manifold M0 identifies with
a subalgebra of a homomorphic image of a groupoid (pseudodifferential) algebra.
For any manifold with corners, we shall denote by VM the Lie algebra of all vector
fields tangent to all faces of M . Then VM Γ(AM ) as C ∞ (M )–modules for a
unique (up to isomorphism) vector bundle AM → M , by the Serre-Swan theorem.
If the vector bundle A → M defining a Lie manifold M satisfies A AM , then we
shall say that M0 , the interior of M , is a manifold with poly-cylindrical ends. If
moreover G has connected fibers d−1 (x) of the domain map, we prove that A(M0 )
is (isomorphic to) the norm closure of the image of Ψ0 (G). We then use this result
to compute the K-theory of the algebra A−1 (M0 ) and the index in K0 (A−1 (M0 ))
of elliptic operators in the comparison algebra of a manifold with poly-cylindrical
ends.
Let us explain our result in a little more detail. Let us assume that our algebra
D of differential operators is generated by C ∞ (M ) and V = Γ(A). Let D ∈ D;
then the principal symbol of D extends to a symbol defined on A∗ , the dual of A.
Assume that D is elliptic, in the sense that its principal symbol is invertible on A∗
outside the zero section. Then the K-theory six-term exact sequence applied to the
tangent (or adiabatic) groupoid of G defines a map
∗
a : K (A ) → K0 (A−1 (M0 )).
inda = indM 0
(1)
One of our main results is a computation of the groups K0 (A−1 (M0 )) and of the
map inda in case M0 is a manifold with poly-cylindrical ends.
Our calculation of the group K0 (A−1 (M0 )) is modelled on the proof in [38],
where full details can be found. It proceeds as follows. Consider an embedding
ι : M → X of manifolds with corners and let ι! be the push-forward map in K-
theory, which exists because M has a tubular neighborhood in X by [3], where
this is proved in the more general setting of Lie manifolds and their submanifolds.
Morita equivalence then gives rise to a morphism ι∗ : K0 (C ∗ (M )) → K0 (C ∗ (X)).
Then the main result of Theorem 7 is that the following diagram commutes:
ι
K0 (C ∗ (M )) −−−∗−→ K0 (C ∗ (X))

⏐ ⏐ X
a ⏐ ⏐inda
(2) indM
ι
K 0 (A∗M ) −−−!−→ K 0 (A∗X ).
If the manifold with corners X is such that the natural morphisms
ι∗ : K0 (C ∗ a(M )) → K0 (C ∗ (X)) and indX ∗ ∗
a : K (AX ) → K0 (C (X))
0
are isomorphisms, we are going to say that X is a classifying space for M . In that
case, we can interpret the above diagram as a topological index theorem in the
216
usual sense. We also obtain an identification of the groups K0 (C ∗ (M )) and of the

map indM a .
Let us now very briefly summarize the contents of the paper. In Section 1, we
introduce comparison algebras and we show that they are closely related to groupoid
algebras. We show that the groupoid C ∗ -algebra Ψ0 (GM ) and the comparison C ∗ -
algebra A(M0 ) are in fact isomorphic for M0 a manifold with poly-cylindrical ends
with compactification M . In Section 2, we recall the definition of the full C ∗ -
analytic index using the tangent groupoid. In the process, we establish several
technical results on tangent groupoids. Section 3 contains the main properties of
the full C ∗ -analytic index. In this section, we also introduce the morphism j∗
associated to an embedding of manifolds with corners j and we provide conditions
for j∗ and indM a to be isomorphisms. We also discuss he compatibility of the full
C ∗ -analytic index and of the shriek maps. This is then used to establish the equality
of the full C ∗ -analytic and principal symbol topological index. Some of the proofs
in this paper are only sketched. See [38] for full details.
We thank Bernd Ammann, Catarina Carvalho, Severino Toscano Melo, Sergiu
Moroianu, and Georges Skandalis for useful discussions. The second named author
would like to thank the Max Planck Institute for Mathematics Bonn, where part of
this work was completed, for hospitality and support. We also thank an anonymous
referee for carefully reading our paper.
1. Groupoids and comparison algebras

Groupoids arise in many of the above mentioned works of Henri Moscovici. In
fact, they are a mainstay of the applications of C ∗ -algebras to geometry and index
theory. See for instance [1, 2, 8, 51, 52]. For this paper, we shall need to consider
pseudodifferential operators on groupoids [39, 44].
For simplicity, we shall assume from now on that all our manifolds with corners
have embedded faces.
1.1. Pseudodifferential operators on groupoids. Throughout this paper,

we shall fix a Lie groupoid G with units M and Lie algebroid A = A(G). Here M
is allowed to have corners. To G there is associated the pseudodifferential calculus
Ψ∞ (G), whose operators of order m form a linear space denoted by Ψm (G), m ∈ R,
see [39, 44]. In short, this calculus is defined as follows. Let s : G → M be the
source map and Gx = s−1 (x). Then Ψm (G), m ∈ Z, consists of smooth families
of classical, order m pseudodifferential operators (Px ∈ Ψm (Gx )), x ∈ M , that are
right invariant with respect to multiplication by elements of G and are “uniformly
supported.” To define what uniformly supported means, let us observe that the
right invariance of the operators Px implies that their distribution kernels KPx
descend to a distribution kP ∈ I m (G, M ) [36, 44]. Then the family P = (Px ) is
called uniformly supported if, by definition, kP has compact support in G.
We then have the following result [29, 37, 44].
Theorem 1. The space Ψ∞ (G) is a filtered algebra, closed under adjoints, so
that the usual principal symbol of pseudodifferential operators defines a surjective
map
(A∗ )/Scl (A∗ ),
(m)
σG : Ψm (G) → Scl m m−1
with kernel Ψm−1 (G), for any m ∈ Z.

5
1.2. Comparison algebras. We shall denote by π the natural action of

Ψ∞ (G) on Cc∞ (M0 ) or on its completions and by a−∞ the completion of π(Ψ−∞ (G))
acting on all Sobolev spaces H −m (M0 ) → H m (M0 ). Let us define
an := π(Ψn (G)) + a−∞ .
The following result was proved in [30] (see also [3, 31]).
Theorem 2. The space a := π(Ψ(G))+a−∞ is a filtered algebra by the pseudo-
differential degree such that a0 := π(Ψ0 (G)) + a−∞ consists of bounded operators, is
closed under the adjoint, and is spectrally invariant on H 0 (M0 ) = L2 (M0 ). More-
over, a−∞ is a two-sided ideal of a and
Λ := (1 + Δ)−1/2 ∈ a−1 := π(Ψ−1 (G)) + a−∞ .

From the above theorem we obtain that the comparison algebra is a subalgebra
of the norm closure of a0 .
Theorem 3. Let M0 be a Lie manifold. Then we have that
A(M0 ) ⊂ a0 = π(Ψ0 (G)).
Moreover, A(M0 ) contains all compact operators.
Proof. Let Ln ∈ Dn . Since Λ ∈ a−1 and Dn ⊂ an := π(Ψ−1 (G)) + a−∞ , it
follows that Ln Λn ∈ a0 , and hence the result.
To show that A(M0 ) contains the subalgebra of compact operators, let us notice
first that e−tΔ ∈ A(M0 ), since it can be written as a function of Λ := (1 + Δ)−1/2 .
Since Cc∞ (M0 ) ⊂ C ∞ (M ), we have that φe−tΔ ψ ⊂ A(M0 ). Let us denote by K
the ideal of compact operators acting on L2 (M0 ). We have that an integral kernel
operator on L2 (M0 ) with kernel in Cc (M0 × M0 ) is compact, and, moreover, the
resulting map Cc (M0 × M0 ) → K is continuous. Since the distribution kernel
e−tΔ (x, y) is everywhere positive (for t > 0), we can approximate any distribution
kernel in Cc (M0 × M0 ) by linear combinations of operators of the form φe−tΔ ψ,
and hence Cc (M0 × M0 ) ⊂ A(M0 ). Therefore A(M0 ) contains all Hilbert-Schmidt
operators. Since the latter are dense in K, we finally get that K ⊂ A(M0 ).
Let M0 be a manifold with poly-cylindrical ends and G be a groupoid such that

A(G) AM , where, we recall, AM → M is a vector bundle such that VM := Γ(AM )
consists of all smooth vector fields on M that are tangent to all the faces of M . A
groupoid G with this property is said to integrate AM , and is not unique in general.
However, if the fibers of the source map s : G → M are all connected and simply-
connected, then G is unique (up to isomorphism) [33, 43] and will be denoted by
GM . For GM the vector representation π is injective [37, 31]. We shall also denote
by C ∗ (M ) = C ∗ (GM ). Recall that C ∗ (GM ) = Ψ−1 (G), [31, 37].
The algebra Ψ(GM ) was considered before by many authors, including [34, 47,
48]. For this algebra, we actually have equality in the above theorem.
Theorem 4. Let M0 be a manifold with poly-cylindrical ends and G be a Lie
groupoid with connected, simply-connected fibers d−1 (x) such that A(G) AM .
Then
A(M0 ) = a0 Ψ0 (GM ).
218
Proof. Let us recall that for manifolds with poly-cylindrical ends the vector
representation π is injective on the norm closure Ψ0 (GM ). We shall thus identify
Ψ0 (GM ) with π(Ψ0 (GM )). Since the principal symbol map acting on both A(M0 )
and on Ψ0 (GM ) has the same range, namely C(S ∗ AM ), it is enough to show that
A−1 (M0 ) = Ψ−1 (G) = C ∗ (M ).
Let us notice that we can consider families, so proving A−1 (M0 ) = C ∗ (M )
is equivalent to proving A−1 (M0 ) ⊗ C0 (X) = C ∗ (M ) ⊗ C0 (X). Moreover, the
inclusion A−1 (M0 ) ⊂ C ∗ (M ) is compatible with the natural representations of
C ∗ (M ) associated to the faces of M , as seen from their construction in [35]. It
is enough then to prove that we have isomorphisms on subquotients defined by
these representations, which are all of the form A−1 (X0 ) ⊗ C0 (X), for some lower
dimensional manifolds. The proof finally reduces to showing K ⊂ A−1 (M0 ), where
K is the algebra of compact operators. For this we use Theorem 3.
For manifolds with cylindrical ends (that is, when M has no corners of codi-
mension two or higher), this theorem was proved earlier in [19].
2. The analytic index

2.1. The adiabatic and tangent groupoids. For the definition and study
of the full C ∗ -analytic index, we shall need the adiabatic and tangent groupoids
associated to a differentiable groupoid G. We now recall their definition.
Let G be a Lie groupoid with space of units denoted by M . We now construct
both the adiabatic groupoid adG and the tangent groupoid T G associated to G, fol-
lowing [14, 28, 29, 39, 46]. The space of units of the groupoid adG is M × [0, ∞)
and the tangent groupoid T G will be defined as the restriction of adG to M × [0, 1].
As a set, the groupoid adG is the disjoint union:
ad
G = A(G) × {0} ∪ G × (0, ∞).
We shall define a groupoid structure A(G)×{0} by noticing that it is a commutative
bundle of Lie groups, this structure being induced by its vector bundle structure.
We give G × (0, ∞) the product groupoid structure. Then the groupoid structure
of adG is such that A(G) × {0} and G × (0, ∞) are subgroupoids. To give adG a dif-
ferentiable structure, it is enough to specify A(adG), since its knowledge completely
determines the differentiable structure of adG [43]. Then
(3) Γ(A(adG)) = tΓ(A(G × [0, ∞))).
To understand this construction, let us consider the product groupoid G × [0, ∞)
endowed with pointwise operations. Then we can identify a section X ∈ Γ(A(G ×
[0, ∞))) with a smooth function [0, ∞)
t → X(t) ∈ Γ(A(G)). Our construction is
such that Γ(A(adG)) = {tX(t)}, with X ∈ Γ(A(G × [0, ∞))).
It is easy to show that
Lemma 1. Let H = G × Rn . We have that C ∗ (adH) C ∗ (adG) ⊗ C0 (Rn ) and
that C ∗ (T H) C ∗ (T G) ⊗ C0 (Rn ).
2.2. The full C ∗ -analytic index. For each t ∈ [0, 1], M × {t} is a closed
invariant subset of M × [0, ∞) for the adiabatic and tangent groupoids, and hence
it defines an evaluation morphism
et : C ∗ (T G) → C ∗ (T GM ×{t} ),
7
and, in particular, an exact sequence

0 → C ∗ (T GM ×(0,1] ) → C ∗ (T G) −→ C ∗ (A(G)) → 0.
e 0
(4)
Since K∗ (C ∗ (T GM ×(0,1] )) = K∗ (C ∗ (G) ⊗ C0 ((0, 1]) = 0, the evaluation map e0
induces an isomorphism in K-theory.
The C ∗ -algebra C ∗ (A(G)) is commutative and hence we have the following
canonical isomorphism C ∗ (A(G)) C0 (A∗ (G)). This then gives the canonical iso-
morpphism K∗ (C ∗ (A(G))) K ∗ (A∗ (G)) which allows us then to define the full
C ∗ -analytic index inda as the composition map
(5) indGa = e1 ◦ e−1 ∗ ∗ ∗
0 : K (A (G)) → K∗ (C (G)),
where e1 : C ∗ (T G) → C ∗ (T GM ×{1} ) = C ∗ (G) is defined by the restriction map to

M × {1}. The definition of the full C ∗ -analytic index gives the following [38].
Proposition 1. Let G be a Lie groupoid with Lie algebroid π : A(G) →
M . Also, let N ⊂ F ⊂ M be a closed, invariant subset which is an embedded
submanifold of a face F of M . Then the full C ∗ -analytic index defines a morphism
of the six-term exact sequences associated to the pair (A∗ (G), π −1 (N )) and to the
ideal C ∗ (GN c ) ⊂ C ∗ (G), N c := M N :
K 0 (π −1 (N c )) −−−−→ K 0 (A∗ (G)) −−−−→ K 0 (π −1 (N )) −−−−→ K 1 (π −1 (N c ))
⏐ ⏐ ⏐ ⏐
⏐ ⏐ ⏐ ⏐

K0 (C ∗ (GN c )) −−−−→ K0 (C ∗ (G)) −−−−→ K0 (C ∗ (GN )) −−−−→ K1 (C ∗ (GN c ))
Proof. The six-term, periodic long exact sequence in K-theory associated to
the pair (A∗ (G), π −1 (N )) is naturally isomorphic to the six-term exact sequence in
K-theory associated to the pair C0 (A∗M N ) ⊂ C0 (A∗ (G)). The result follows from
the naturality of the six-term exact sequence in K-theory and the definition of the
full C ∗ -analytic index (6).
Recall that for M a smooth manifold with corners with embedded faces, we
have denoted A(GM ) = AM and C ∗ (M ) := C ∗ (GM ). Then the full C ∗ -analytic
index becomes the desired map
∗ ∗ ∗
(6) a : K (AM ) → K∗ (C (M )).
indM
See [21, 22] for more properties of the analytic index.
Remark 5. Assume the manifold M is smooth (so it has no corners or bound-
aries). Then GM = M × M is the product groupoid and hence we see right away
that Ψ∞ (GM ) = Ψ∞ (M ). In particular, we have that C ∗ (M ) := C ∗ (GM ) K,
the algebra of compact operators on M . Then K0 (C ∗ (M )) = Z, and hence inda
is precisely the analytic index introduced in [5]. This construction extends to the
case when M is not compact, if one uses pseudodifferential operators of order zero
that are “multiplication at infinity,” as in [11].
3. Properties of the full C ∗ -analytic index

We now proceed as in [38], in particular, we need the following proposition for
proof the index theorem of Theorem 8.
220
Proposition 2. Let X be a manifold with embedded faces such that each open
face of X is diffeomorphic to a Euclidean space. Then the full C ∗ -analytic index
∗ ∗ ∗
a : K (AX ) → K∗ (C (X)),
indX
defined in Equation (6), is an isomorphism.
Proof. The proof is by induction on the number of faces of X using Proposi-
tion 1, the six-term exact sequence in K-theory, and the Five Lemma in homological
algebra.
Remark 6. The above proposition can be regarded as a Baum–Connes iso-

morphism for manifolds with corners.
Proposition 3. Let ι : M → X be a closed embedding of manifolds with
corners. Assume that, for each open face F of X, the intersection F ∩ M is a
non-empty open face of M and that every open face of M is obtained in this way.
Then K∗ (C ∗ (M )) → K∗ (C ∗ (X)) is an isomorphism denoted by ι∗ .
Proof. Recall from [42] that two locally compact groupoids G and H are
equivalent provided there exists a topological space Ω and two continuous, surjective
open maps r : Ω → G (0) and d : Ω → H (0) together with a left (respectively right)
action of G (respectively H) on Ω with respect to r (respectively d), such that
r (respectively d) is a principal fibration of structural groupoid H (respectively
G). An important theorem of Muhly–Renault–Williams states that if G and H are
equivalent, then K∗ (C ∗ (G)) K∗ (C ∗ (H)) [42].
Our result then follows from the fact that Ω := r −1 (M ) establishes the desired
equivalence between GM and GX .
We can now prove a part of our principal symbol topological index theorem,
Theorem 8, involving an embedding ι : M → X of our manifold with corners M
into another manifold with corners X. This theorem amounts to the fact that the
diagram (2) is commutative. In order to prove the commutativity of this diagram,
we shall first consider a tubular neighborhood
k j
(7) M → U → X
of M in X, so that ι = j ◦ k. The diagram (2) is then decomposed into the
two diagrams below, and hence the proof of the commutativity of the diagram
(2) reduces to the proof of the commutativity of the two diagrams below, whose
morphisms are defined as follows: the morphism k∗ is defined by Proposition 3 and
the morphism j∗ is defined by the inclusion of algebras. The morphism ι∗ is defined
by ι∗ = j∗ ◦ k∗ . Finally, the morphism k! is the push-forward morphism.
Let us now turn our attention to the following diagram:
k j∗
K∗ (C ∗ (M )) −−−∗−→ K∗ (C ∗ (U )) −−−−→ K∗ (C ∗ (X))

⏐ ⏐ ⏐ X
a ⏐ a⏐ ⏐inda
(8) indM indU
k j!
K ∗ (A∗M ) −−−−
!
→ K ∗ (A∗U ) −−−−→ K ∗ (A∗X ).
The commutativity of the left diagram is part of the following proposition,
which is the most technical part of the proof. Its proof is obtained by integrating
a Lie algebroid obtained as a double deformation of a tangent space [38].
9
Proposition 4. Let π : U → M be a real vector bundle over a manifold with

corners M and let k : M → U be the “zero section” embedding. Then we have the
following commutative diagram
k
K∗ (C ∗ (M )) −−−∗−→ K∗ (C ∗ (U ))

⏐ ⏐
(9) indMa ⏐ a⏐
indU

K ∗ (A∗M ) −−−−→ K ∗ (A∗U )
k!
The commutativity of the second square in the Diagram 8 is an immediate

consequence of the naturality of the tangent groupoid construction.
Proposition 5. Let j : U → X be the inclusion of the open subset U . Then
the diagram below commutes:
j∗
K∗ (C ∗ (U )) −−−−→ K∗ (C ∗ (X))

⏐ ⏐ X
indU
a ⏐ ⏐inda
j∗
K ∗ (A∗U ) −−−−→ K ∗ (A∗X ).

As explained above, the previous two propositions give

ι
Theorem 7. Let M → X be a closed embedding of manifolds with corners.
Then the diagram
ι
K∗ (C ∗ (M ))) −−−∗−→ K∗ (C ∗ (X))

⏐ M ⏐
(10) ⏐inda indXa ⏐
ι
K ∗ (A∗M ) −−−!−→ K ∗ (A∗X )
is commutative.
4. A topological index theorem

Motivated by Theorem 7 and by the results of Section 3 (see Propositions 2
and 3) we introduce the following definition.
Definition 1. A strong classifying manifold XM of M is a compact manifold
with corners XM , together with a closed embedding ι : M → XM with the following
properties:
(i) each open face of XM is diffeomorphic to a Euclidean space,
(ii) F → F ∩ M induces a bijection between the open faces of XM and M .
Note that if M ⊂ XM are as in the above definition, then each face of M is the
transverse intersection of M with a face of XM .
Proposition 6. Let M be a manifold with embedded faces and ι : M → XM
be a strong classifying space of M . Then the maps ι∗ and indX
a of Theorem 7 are
isomorphisms. That is, a strong classifying space for M is a classifying space for
M.
Proof. This was proved in Propositions 2 and 3.
222
Let ι : M → XM be a strong classifying space for M . The above proposition

then allows us to define (see Diagram 10)
−1 ∗ ∗ ∗
t := ι∗ ◦ inda ◦ι! : K (AM ) → K∗ (C (M )).
indM X
If M is a smooth compact manifold (so, in particular, ∂M = ∅), then C ∗ (M ) =

K, the algebra of compact operators on L2 (M ) and hence K0 (C ∗ (M )) = Z. Any
embedding ι : M → RN will then be a classifying space for M . Moreover, for X =
Rn , the map ι−1 X ∗
∗ ◦ inda : K (T X) → Z is the inverse of j! : K (pt) → K (T R )
0 0 N
R N
−1
and hence indt = (j! ) ι! , which is the definition of the topological index from [5].
In view of this fact, we shall also call the map indM
t the topological index associated
to M . Theorem 7 then gives the following result:
Theorem 8. Let M be a manifold with corners and AM and C ∗ (M ) be the
Lie algebroid and the C ∗ -algebra associated to M . Then the principal topological
index map indM t depends only on M , that is, it is independent of the classifying
space XM , and we have
∗ ∗ ∗
t = inda : K (AM ) → K∗ (C (M )).
indM M
If M is a smooth compact manifold (no boundary), this recovers the Atiyah-

Singer index theorem on the equality of the full C ∗ -analytic and principal symbol
topological index [5].
5. K-theory of comparison algebras

The isomorphism K∗ (C ∗ (M )) K ∗ (XM ) provides us with a way of determin-
ing the groups K∗ (C ∗ (M )). In particular, we have completed the determination of
the K-theory groups of the comparison algebra A−1 (M0 ) = C ∗ (M ), if the interior
of M is a endowed with a metric making it a manifold with poly-cylindrical ends.
Theorem 9. Let M be a manifold with corners and embedded faces. We have
K∗ (C ∗ (M )) K ∗ (XM ). Moreover Kj (C ∗ (M )) ⊗ Q Qpj , where pj is the number
of faces of M of dimension ≡ j modulo 2.
The last part of the above theorem is proved by showing that the Atiyah-
Hirzebruch spectral sequence of XM collapses at E 2 . This is part of a joint work
with Etienne Fieux.
It is not difficult to construct a classifying manifold XM of M [38], that is,
a manifold such that ι∗ : K0 (C ∗ (M )) → K0 (C ∗ (X)) and indX ∗
a : K (AX ) →
0
∗
K0 (C (X)) are isomorphisms. Let us assume M is compact with embedded faces.
The space XM is obtained from an embedding X → [0, ∞)N for some large N , and
then by removing suitable hyperplanes from the boundary of [0, ∞)N such that
each face of M is the transverse intersection of M and of a face of XM .
References
[1] P. Albin and F. Rochon. Families index for manifolds with hyperbolic cusp singularities. Int.
Math. Res. Not. IMRN, (4):625–697, 2009.
[2] A. Alldridge and T. Johansen. An index theorem for Wiener-Hopf operators. Adv. Math.,
218(1):163–201, 2008.
[3] B. Ammann, R. Lauter, and V. Nistor. On the geometry of Riemannian manifolds with a Lie
structure at infinity. Int. J. Math. Math. Sci., (1-4):161–193, 2004.
[4] I. Androulidakis and G. Skandalis. The holonomy groupoid of a singular foliation. J. Reine
Angew. Math., 626:1–37, 2009.
11
[5] M. F. Atiyah and I. M. Singer. The index of elliptic operators, I. Annals of Math., 87:484–530,
1968.
[6] D. Barbasch and H. Moscovici. L2 -index and the Selberg trace formula. J. Funct. Anal.,
53(2):151–201, 1983.
[7] M.-T. Benameur and J. Heitsch. Index theory and non-commutative geometry. I. Higher
families index theory. K-Theory, 33(2):151–183, 2004.
[8] Moulay-Tahar Benameur and Paolo Piazza. Index, eta and rho invariants on foliated bundles.
Astérisque, (327):201–287 (2010), 2009.
[9] J.-P. Brasselet and A. Legrand. Teleman localization of Hochschild homology in a singular
setting. Russ. J. Math. Phys., 16(3):391–403, 2009.
[10] J. Brüning and H. Moscovici. L2 -index for certain Dirac-Schrödinger operators. Duke Math.
J., 66(2):311–336, 1992.
[11] C. Carvalho. A topological approach to the cobordism invariance of the index. K-Theory,
2006.
[12] A. Connes. Sur la théorie non commutative de l’intégration. In Algèbres d’opérateurs (Sém.,
Les Plans-sur-Bex, 1978), volume 725 of Lecture Notes in Math., pages 19–143. Springer,
Berlin, 1979.
[13] A. Connes. A survey of foliations and operator algebras. In Operator algebras and applica-
tions, Part I (Kingston, Ont., 1980), volume 38 of Proc. Sympos. Pure Math., pages 521–628.
Amer. Math. Soc., Providence, R.I., 1982.
[14] A. Connes. Noncommutative Geometry. Academic Press, Inc., 1994.
[15] A. Connes and H. Moscovici. Cyclic cohomology, the Novikov conjecture and hyperbolic
groups. Topology, 29(3):345–388, 1990.
[16] A. Connes and H. Moscovici. The local index formula in noncommutative geometry. Geom.
Funct. Anal., 5(2):174–243, 1995.
[17] H. Cordes. Spectral theory of linear differential operators and comparison algebras, volume 76
of London Mathematical Society Lecture Note Series. Cambridge University Press, Cam-
bridge, 1987.
[18] H. Cordes. On essential selfadjointness of powers and comparison algebras. J. Anal. Math.,
58:61–97, 1992. Festschrift on the occasion of the 70th birthday of Shmuel Agmon.
[19] H. Cordes and S. H. Doong. The Laplace comparison algebra of spaces with conical and
cylindrical ends. In Pseudodifferential operators (Oberwolfach, 1986), volume 1256 of Lecture
Notes in Math., pages 55–90. Springer, Berlin, 1987.
[20] H. Cordes and R. McOwen. The C ∗ -algebra of a singular elliptic problem on a noncompact
Riemannian manifold. Math. Z., 153(2):101–116, 1977.
[21] C. Debord and J.-M. Lescure. K-duality for stratified pseudomanifolds. Geom. Topol.,
13(1):49–86, 2009.
[22] C. Debord, J.-M. Lescure, and V. Nistor. Groupoids and an index theorem for conical pseudo-
manifolds. J. Reine Angew. Math., 628:1–35, 2009.
[23] V. Georgescu. On the spectral analysis of quantum field Hamiltonians. J. Funct. Anal.,
245(1):89–143, 2007.
[24] A. Gorokhovsky and J. Lott. Local index theory over foliation groupoids. Adv. Math.,
204(2):413–447, 2006.
[25] E. Hunsicker, V. Nistor, and J. Sofo. Analysis of periodic Schrödinger operators: regularity
and approximation of eigenfunctions. J. Math. Phys., 49(8):083501, 21, 2008.
[26] J. Kaminker and X. Tang. Hopf algebroids and secondary characteristic classes. J. Noncom-
mut. Geom., 3(1):1–25, 2009.
[27] Yu. Kordyukov. Index theory and noncommutative geometry on manifolds with foliations.
Uspekhi Mat. Nauk, 64(2(386)):73–202, 2009.
[28] N. Landsman. Quantization and the tangent groupoid. In Operator algebras and mathematical
physics (Constanţa, 2001), pages 251–265. Theta, Bucharest, 2003.
[29] R. Lauter, B. Monthubert, and V. Nistor. Pseudodifferential analysis on continuous family
groupoids. Doc. Math., 5:625–655 (electronic), 2000.
[30] R. Lauter, B. Monthubert, and V. Nistor. Spectral invariance for certain algebras of pseudo-
differential operators. J. Inst. Math. Jussieu, 4(3):405–442, 2005.
[31] R. Lauter and V. Nistor. Analysis of geometric operators on open manifolds: a groupoid
approach. In Quantization of singular symplectic quotients, volume 198 of Progr. Math.,
pages 181–229. Birkhäuser, Basel, 2001.
224
[32] H. Li, A. Mazzucato, and V. Nistor. Analysis of the finite element method for transmis-
sion/mixed boundary value problems on general polygonal domains. Electronic Transactions
Numerical Analysis, 37:41–69, 2010.
[33] K. Mackenzie. Lie groupoids and Lie algebroids in differential geometry, volume 124 of Lecture
Notes Series. London Mathematical Society, 1987.
[34] R. Melrose. Geometric scattering theory. Stanford Lectures. Cambridge University Press,
Cambridge, 1995.
[35] R. Melrose and V. Nistor. K-theory of C ∗ -algebras of b-pseudodifferential operators. Geom.
Funct. Anal., 8(1):88–122, 1998.
[36] B. Monthubert. Groupoı̈des et calcul pseudo-différentiel sur les variétés à coins. PhD thesis,
Université Paris 7, 1998.
[37] B. Monthubert. Pseudodifferential calculus on manifolds with corners and groupoids. Proc.
Amer. Math. Soc., 127(10):2871–2881, 1999.
[38] B. Monthubert and V. Nistor. A principal symbol topological index theorem for manifolds
with corners. to appear in Compositio Math.
[39] B. Monthubert and F. Pierrot. Indice analytique et groupoı̈des de Lie. C. R. Acad. Sci. Paris
Sér. I Math., 325(2):193–198, 1997.
[40] H. Moscovici and R. Stanton. R-torsion and zeta functions for locally symmetric manifolds.
Invent. Math., 105(1):185–216, 1991.
[41] H. Moscovici and A. Verona. Harmonically induced representations of nilpotent Lie groups.
Invent. Math., 48(1):61–73, 1978.
[42] P. Muhly, J. Renault, and D. Williams. Equivalence and isomorphism for groupoid C *-
algebras. J. Operator Theory, 17(1):3–22, 1987.
[43] V. Nistor. Groupoids and the integration of Lie algebroids. J. Math. Soc. Japan, 52:847–868,
2000.
[44] V. Nistor, A. Weinstein, and P. Xu. Pseudodifferential operators on groupoids. Pacific J.
Math., 189:117–152, 1999.
[45] M. Pflaum, H. Posthuma, and X. Tang. On the algebraic index for Riemannian étale
groupoids. Lett. Math. Phys., 90(1-3):287–310, 2009.
[46] B. Ramazan. Limite classique de C ∗ -algèbres de groupoı̈des de Lie. C. R. Acad. Sci. Paris
Sér. I Math., 329(7):603–606, 1999.
[47] E. Schrohe and B.-W. Schulze. Boundary value problems in Boutet de Monvel’s algebra
for manifolds with conical singularities. I. In Pseudo-differential calculus and mathematical
physics, volume 5 of Math. Top., pages 97–209. Akademie Verlag, Berlin, 1994.
[48] B. Schulze. Boundary value problems and singular pseudo-differential operators. Wiley-
Interscience Series in Pure and Applied Mathematics. Chichester: John Wiley & Sons., 1998.
[49] R. Strichartz. Harmonic analysis as spectral theory of Laplacians. J. Funct. Anal., 87(1):51–
148, 1989.
[50] M. Taylor. Gelfand theory of pseudodifferential operators and hypoelliptic operators. Trans.
Amer. Math. Soc., 153:495–510, 1971.
[51] E. van Erp. The Atiyah-Singer index formula for subelliptic operators on contact manifolds.
Part I. Ann. of Math. (2), 171(3):1647–1681, 2010.
[52] E. Vasselli. Bundles of C ∗ -categories. II. C ∗ -dynamical systems and Dixmier-Douady invari-
ants. J. Funct. Anal., 257(2):357–387, 2009.
Institut de Mathmatiques de Toulouse, Université Paul Sabatier, Laboratoire Emile

Picard, F-31062 Toulouse CEDEX 4
E-mail address: bertrand.monthubert@math.univ-toulouse.fr
Pennsylvania State University, Math. Dept., University Park, PA 16802

E-mail address: nistor@math.psu.edu
Volume 546, 2011
Relative pairings and the Atiyah-Patodi-Singer index

formula for the Godbillon-Vey cocycle
Hitoshi Moriyoshi and Paolo Piazza
Dedicated to Henri Moscovici on the occasion of his sixty-fifth anniversary
Abstract. We describe a Godbillon-Vey index formula for longitudinal Dirac

operators on a foliated bundle (X, F ) with boundary; in particular, we define
a Godbillon-Vey eta invariant on (∂X, F∂ ), that is, a secondary invariant for
longitudinal Dirac operators on type III foliations. Our theorem generalizes the
classic Atiyah-Patodi-Singer index formula for (X, F ). Moreover, employing
the Godbillon-Vey index as a pivotal example, we explain a new approach to
higher index theory on geometric structures with boundary. This is heavily
based on the interplay between the absolute and relative pairing of K-theory
and cyclic cohomology for an exact sequence of Banach algebras, which in
the present context takes the form 0 → J → A → B → 0 with J dense and
holomorphically closed in C ∗ (X, F ) and B depending only on boundary data.
1. Introduction
Connes’ index theorem for G-proper manifolds [1], with G an étale groupoid,
unifies under a single statement most of the existing (longitudinal) index theorems.
We shall focus on a particular case of such a theorem, that of foliated bundles.
Thus, let N be a closed compact manifold. Let Γ → Ñ → N be a Galois Γ-cover.
Let T be a smooth oriented compact manifold with an action of Γ which is assumed
to be by diffeomorphisms, orientation preserving and locally faithful, as in [14]. Let
Y = Ñ ×Γ T and let (Y, F) be the associated foliation. (This is an example of G-
proper manifold with G equal to the groupoid T Γ.) Let D be a Γ-equivariant
family of Dirac operators on the fibration Ñ × T → T ; such a family induces a
longitudinal Dirac operator on (Y, F).
2010 Mathematics Subject Classification. Primary: 58J20. Secondary: 58J22, 58J42, 19K56.
Key words and phrases. Foliations, foliated bundles, Godbillon-Vey invariant, groupoids,
Godbillon-Vey cyclic cocycle, index classes, relative pairing, excision, Atiyah-Patodi-Singer higher
index theory, Godbillon-Vey eta invariant.
Research partially supported by Japan Society for the Promotion of Science (JSPS),
Grants-in-Aid for Scientific Research; the 21st century COE program at Keio University; Is-
tituto Nazionale di Alta Matematica Francesco Severi, through the GNSAGA; the Ministero
dell’Istruzione, dell’Università e della Ricerca (MIUR) through the project Spazi di moduli e
Teoria di Lie.
2011
c American
c Mathematical Society
1
225
226
2 HITOSHI MORIYOSHI AND PAOLO PIAZZA
If T = point and Γ = {1} we have a compact manifold and Connes’ index

theorem reduces to the Atiyah-Singer index theorem. If Γ = {1} we simply have
a fibration and the theorem reduces to the Atiyah-Singer family index theorem.
If T = point then we have a Galois covering and Connes’ index theorem reduces
to the Connes-Moscovici higher index theorem. If dim T > 0 and Γ = {1}, then
Connes’ index theorem is a higher foliation index theorem on the foliated manifold
(Y, F).
One particularly interesting higher index is the so-called Godbillon-Vey index;
an alternative treatment of Connes’ index formula in this particular case was given
by Moriyoshi-Natsume in [14]. Subsequently, Gorokhovsky and Lott [4] gave a su-
perconnection proof of Connes’ index theorem, including an explicit formula for the
Godbillon-Vey higher index. Leichtnam and Piazza [7] extended Connes’ index the-
orem to foliated bundles with boundary, using an extension of Melrose b-calculus
and the Gorokhovsky-Lott superconnection approach. Unfortunately, a key as-
sumption in [7] is that the group Γ be of polynomial growth. This excludes many
interesting examples and higher indeces; in particular it excludes the possibility of
proving a Atiyah-Patodi-Singer formula for the Godbillon-Vey higher index.
One primary objective of this article is to illustrate such a result. Complete

proofs will appear in [16].
In tackling the problem we develop what we believe is a new approach to index

theory on manifolds with boundary. This can be summarized as follows. We define
a short exact sequence of Banach algebras
0→J→A→B→0
with J dense and holomorphically closed in C ∗ (X, F) and with B depending only
on boundary data. We prove that there are well defined Dirac index classes, de-
noted respectively Ind(D, D∂ ) ∈ K∗ (A, B) and Ind(D) ∈ K∗ (J), and that these
index classes correspond under excision; the relative class Ind(D, D∂ ) ∈ K∗ (A, B)
is obtained by using the graph projection of D and of Dcyl whereas the index class
Ind(D) ∈ K∗ (J) is obtained through the parametrix of D and the associated re-
mainders. Next, for (certain) cyclic k-cocycles defining a higher index in the closed
case, let us name one of such cocycles τk , we define
• a cyclic k-cocycle on J, still denoted τk ;
• a eta cyclic cocycle σk+1 on B; σk+1 (which thus depends solely on bound-
ary data) is obtained by a sort of suspension procedure involving τk and
a specific 1-cocycle σ1 (Roe’s 1-cocycle);
• a relative cyclic k-cocyle (τkr , σk+1 ), with τkr a cyclic cochain defined from
τk through a regularization à la Melrose.
The index formula in this context is obtained by establishing the equality
Ind(D), [τk ] = Ind(D, D∂ ), [τkr , σk+1 ] .
On the left hand side we have the absolute pairing, which is by definition the
higher index. On the right hand side we have the relative pairing; multiplying the
operator by s > 0, using the definition of the relative pairing and taking the limit
as s ↓ 0 we obtain the right hand side of the Atiyah-Patodi-Singer index formula.
The eta-correction term is obtained through the eta cocycle σk+1 .
RELATIVE PAIRINGS AND THE GODBILLON-VEY HIGHER INDEX THEOREM 227
3
We end this brief introduction by pointing out that relative pairings in K-

theory and cyclic cohomology have already been successfully employed in the study
of geometric and topological invariants of elliptic operators. We particularly wish
to mention here the paper by Lesch, Moscovici and Pflaum [9]; in this interest-
ing article the absolute and relative pairings associated to a suitable short exact
sequence of algebras (this is a short exact sequence of parameter dependent pseu-
dodifferential operators) are used in order to define and study a generalization of
the divisor flow of Melrose on a closed compact manifold, see [12] and also [10].
The results of this article first appeared in [15].
Acknowledgements. Most of this work has been done while the first author was
visiting Sapienza Università di Roma and the second author was visiting Keio Uni-
versity and Nagoya University. We thank the Japan Society for the Promotion of
Science and the 21st century COE program at Keio for sponsoring most of these
visits. Further financial support was provided by Istituto Nazionale di Alta Matem-
atica Francesco Severi and Ministero dell’Istruzione, dell’Università e della Ricerca
(MIUR). Part of this research was also carried out while the two authors were vis-
iting jointly the Chern Institute in Tianjin and the Institut de Mathématiques de
Jussieu in Paris (Équipe Algèbres d’Opérateurs). We thank these institutions for
hospitality and financial support. Finally, it is a pleasure to thank Sergio Doplicher,
Sacha Gorokhovsky, Eric Leichtnam, Henri Moscovici, Toshikazu Natsume, John
Roe and Xiang Tang for helpful discussions.
2. Geometry of foliated bundles.

2.1. Manifolds with boundary. Let now (M, g) be a riemannian manifold
with boundary; the metric is assumed to be of product type in a collar neighborhood
U [0, 1] × ∂M of the boundary. Let M̃ be a Galois Γ-cover of M ; we let g̃ be the
lifted metric. We also consider ∂ M̃ , the boundary of M̃ . Let T be a smooth oriented
compact manifold with an action of Γ by orientation-preserving diffeomorphisms.
We assume that this action is locally faithful, as in [14], that is: if γ ∈ Γ acts as
the identity map on an open set in T , then γ is the identity element in Γ.
Let X0 = M̃ ×Γ T ; this is a manifold with boundary and the boundary ∂X0 is
equal to ∂ M̃ ×Γ T . (X0 , F0 ) denotes the associated foliated bundle. The leaves of
(X0 , F0 ) are manifolds with boundary endowed with a product-type metric. The
boundary ∂X0 inherits a foliation F∂ . The cylinder R×∂X0 also inherits a foliation
Fcyl , obtained by crossing the leaves of F∂ with R. Similar considerations apply to
the half cylinders (−∞, 0] × ∂X0 and ∂X0 × [0, +∞) .
2.2. Manifolds with

cylindrical ends. Notation. We consider Ṽ :=
M̃ ∪∂ M̃ (−∞, 0] × ∂ M̃ , endowed with the extended metric and the obviously
extended Γ action along the cylindrical end. We consider X := Ṽ ×Γ T ; this
is a foliated bundle, with leaves manifolds with cylindrical ends. We denote by
(X, F) this foliation. Notice that X = X0 ∪∂X0 ((−∞, 0] × ∂X0 ); moreover the
foliation F is obtained by extending F0 on X0 to X via the product cylindri-
cal foliation Fcyl on (−∞, 0] × ∂X0 . We can write more suggestively: (X, F) =
(X0 , F0 ) ∪(∂X0 ,F∂ ) (((−∞, 0] × ∂X0 , Fcyl )). For λ > 0 we shall also consider the

finite cyclinder Ṽλ = M̃ ∪∂ M̃ [−λ, 0] × ∂ M̃ and the resulting foliated manifold
(Xλ , Fλ ). Finally, with a small abuse, we introduce the notation: cyl(∂X) :=
228
R × ∂X0 , cyl− (∂X) := (−∞, 0] × ∂X0 and cyl+ (∂X) := ∂X0 × [0, +∞) The folia-
tions induced on cyl(∂X), cyl± (∂X) by F∂ will be denoted by Fcyl , Fcyl
±
.
2.3. Holonomy groupoid. We consider the groupoid G := (Ṽ × Ṽ × T )/Γ

with Γ acting diagonally; the source map and the range map are defined by s[y, y , θ] =
[y , θ], r[y, y , θ] = [y, θ]. Since the action on T is assumed to be locally faithful, we
know that (G, r, s) is isomorphic to the holonomy groupoid of the foliation (X, F).
In the sequel, we shall simply call (G, r, s) the holonomy groupoid. If E → X is a
hermitian vector bundle on X, with product structure along the cylindrical end,
then we can consider the bundle over G equal to (s∗ E)∗ ⊗ r ∗ E.
3. Wiener-Hopf extensions
3.1. Foliation C ∗ -algebras. We consider Cc (X, F) := Cc (G). Cc (X, F) can
also be defined as the space of Γ-invariant continuous functions on Ṽ × Ṽ ×T with Γ-
compact support. More generally we consider Cc (X, F; E) := Cc (G, (s∗ E)∗ ⊗ r ∗ E)
with its well known *-algebra structure given by convolution. We shall often omit
the vector bundle E from the notation.
The foliation C ∗ -algebra C ∗ (X, F; E) is defined by completion of Cc (X, F; E).
See for example [14] where it is also proved that C ∗ (X, F; E) is isomorphic to the
C ∗ -algebra of compact operators of the Connes-Skandalis C(T ) Γ-Hilbert module
E (this is also described in [14]). Summarizing: C ∗ (X, F; E) ∼= K(E) ⊂ L(E).
3.2. Foliation von Neumann algebras. Consider the family of Hilbert
spaces H := (Hθ )θ∈T , with Hθ := L2 (Ṽ × {θ}, Eθ ). Then Cc (Ṽ × T ) is a con-
tinuous field inside H, that is, a linear subspace in the space of measurable sec-
tions of H. Let End(H) the space of measurable families of bounded operators
T = (Tθ )θ∈T , where bounded means that each Tθ is bounded on Hθ . Then End(H)
is a C ∗ -algebra, in fact a von Neumann algebra, equipped with the norm
T := ess. sup{Tθ ; θ ∈ T }
with Tθ the operator norm. We also denote by EndΓ (H) the C ∗ -subalgebra of
End(H) consisting of Γ-equivariant measurable families of operators. This is often
denoted W ∗ (X, F) and named the foliation von Neumann algebra associated to
(X, F). We set CΓ∗ (H) the closure of Γ-equivariant families T = (Tθ )θ∈T ∈ EndΓ (H)
preserving the continuous field Cc (Ṽ × T ). In [14], Section 2 it is proved that
the foliation C ∗ -algebra C ∗ (X, F) is isomorphic to a C ∗ -subalgebra of CΓ∗ (H) ⊂
EndΓ (H) 1. Notice, in particular, that an element in C ∗ (X, F) can be considered
as a Γ-equivariant family of operators.
3.3. Translation invariant operators. Recall cyl(∂X) := R × ∂X0 ≡ (R ×
∂ M̃ ) ×Γ T with Γ acting trivially in the R-direction of (R × ∂ M̃ ). We consider the
foliated cylinder (cyl(∂X), Fcyl ) and its holonomy groupoid Gcyl := ((R × ∂ M̃ ) ×
(R × ∂ M̃ ) × T )/Γ (source and range maps are clear). Let R act trivially on T ; then
(R × ∂ M̃ ) × (R × ∂ M̃ ) × T has a R × Γ-action, with R acting by translation on
itself. We consider the *-algebra Bc (cyl(∂X), Fcyl ) ≡ Bc defined as
(3.1)
{k ∈ C((R×∂ M̃ )×(R×∂ M̃ )×T ); k is R×Γ-invariant, k has R×Γ-compact support}
1The C ∗ -algebra C ∗ (H) was denoted B in [14]

Γ
5
The product is by convolution. An element in Bc defines a Γ-equivariant family

((θ))θ∈T of translation-invariant operators. The completion of Bc with respect to
the obvious C ∗ -norm (the sup over θ of the operator-L2 -norm of (θ)) gives us a
C ∗ -algebra that will be denoted B ∗ (cyl(∂X), Fcyl ) or more briefly B ∗ .
3.4. Wiener-Hopf extensions. Recall the Hilbert C(T ) Γ-module E and
the C ∗ -algebras K(E) and L(E). Since the C(T ) Γ-compact operators K(E) are
an ideal in L(E) we have the classical short exact sequence of C ∗ -algebras
π
0 → K(E) → L(E) −
→ Q(E) → 0
with Q(E) = L(E)/K(E) the Calkin algebra. Let χ0R : R → R be the characteristic
function of (−∞, 0]; let χR : R → R be a smooth function with values in [0, 1] such
that χ(t) = 1 for t ≤ −, χ(t) = 0 for t ≥ 0. Let χ0 and χ be the functions induced
by χ0R and χR on X. Similarly, introduce χ0cyl and χcyl .
Lemma 3.2. There exists a bounded linear map
(3.3) s : B ∗ → L(E)
extending sc : Bc → L(E), sc () := χ0 χ0 . Moreover, the composition ρ = πs
induces an injective C ∗ -homomorphism
(3.4) ρ : B ∗ → Q(E).
We consider Im ρ as a C ∗ -subalgebra in Q(E) and identify it with B ∗ (cyl(∂X), Fcyl )
via ρ. Set
A∗ (X; F) := π −1 (Im ρ) with π the projection L(E) → Q(E).
Recalling the identification C ∗ (X, F) = K(E), we thus obtain a short exact sequence
of C ∗ -algebras:
0 → C ∗ (X, F) → A∗ (X; F) −
→ B ∗ (cyl(∂X), Fcyl ) → 0
π
(3.5)
where the quotient map is still denoted by π. Notice that (3.5) splits as a short exact
sequence of Banach spaces, since we can choose s : B ∗ (cyl(∂X), Fcyl ) → A∗ (X; F)
the map in (3.3) as a section. So
A∗ (X; F) ∼= C ∗ (X, F) ⊕ s(B ∗ (cyl(∂X), Fcyl ))
as Banach spaces.
There is also a linear map t : A∗ (X, F) → C ∗ (X, F) which is obtained as
follows: if k ∈ A∗ (X; F), then k is uniquely expressed as k = a + s() with a ∈
C ∗ (X, F) and π(k) = ∈ B ∗ (cyl(∂X), Fcyl ). Thus, π(k) = [χ0 χ0 ] ∈ Q(E) for one
(and only one) ∈ B ∗ (cyl(∂X), Fcyl ) since ρ is injective. We set
(3.6) t(k) := k − sπ(k) = k − χ0 χ0
Then t(k) ∈ C ∗ (X, F).
4. Relative pairings and the eta cocycle: the algebraic theory

4.1. Introductory remarks. On a closed foliated bundle (Y, F), the Godbillon-
Vey cyclic cocycle is initially defined on the ”small” algebra Ac ⊂ C ∗ (Y, F) of Γ-
equivariant smoothing operators of Γ-compact support (viz. Ac := Cc∞ (G, (s∗ E)∗ ⊗
r ∗ E)). Since the index class defined using a pseudodifferential parametrix is already
well defined in K∗ (Ac ), the pairing between the the Godbillon-Vey cyclic cocycle
and the index class is well-defined.
230
In a second stage, the cocycle is continuously extended to a dense holomorphi-

cally closed subalgebra A ⊂ C ∗ (Y, F); there are at least two reasons for doing this.
First, it is only by going to the C ∗ -algebraic index that the well known properties
for the signature and the spin Dirac operator of a metric of positive scalar curvature
hold. The second reason for this extension rests on the structure of the index class
which is employed in the proof of the higher index formula, i.e. either the graph
projection or the Wassermann projection; in both cases Uc is too small to contain
the index class and one is therefore forced to find an intermediate subalgebra A,
Ac ⊂ A ⊂ C ∗ (Y, F); A is big enough for the two particular index classes to belong
to it but small enough for the Godbillon-Vey cyclic cocycle to extend; moreover,
being dense and holomorphically closed it has the same K-theory as C ∗ (Y, F).
Let now (X, F) be a foliated bundle with cylindrical ends; in this section we
shall select ”small” subalgebras
Jc ⊂ C ∗ (X, F) , Ac ⊂ A∗ (X, F) , Bc ⊂ B ∗ (cyl(∂X), Fcyl ) ,
π
with Jc an ideal in Ac , so that there is a short exact sequence 0 → Jc → Ac −→ c
π
Bc → 0 which is a subsequence of 0 → C ∗ (X, F) → A∗ (X; F) − → B ∗ (cyl(∂X), Fcyl )
→ 0. We shall then proceed to define the relevant cyclic cocycles, relative and
absolute, and study, algebraically, their main properties. As in the closed case, we
shall eventually need to find an intermediate short exact sequence, sitting between
the two, 0 → J → A → B → 0, with constituents big enough for the the two
index classes we shall define to belong to them but small enough for the cyclic
cocycles (relative and absolute) to extend; this is quite a delicate point and it will
be explained in Section 5. We anticipate that in contrast with the closed case the
ideal Jc in the small subsequence will be too small even for the index class defined
by a pseudodifferential parametrix. This has to do with the non-locality of the
parametrix on manifolds with boundary; it is a phenomenon that was explained in
detail in [7].
4.2. Small dense subalgebras. Define Jc := Cc∞ (X, F); see subsection 3.1.
Redefine Bc as
{k ∈ C ∞ ((R×∂ M̃ )×(R×∂ M̃ )×T ); k is R×Γ-invariant, k has R×Γ-compact support}
see subsection 3.3. We now define Ac ; consider the functions χλ , χλcyl induced on
X and cyl(∂X) by the real function χR
(−∞,−λ] . We shall say that k is in Ac if it is
a smooth function on Ṽ × Ṽ × T which is Γ-invariant and for which there exists
λ ≡ λ(k) > 0, such that
• k − χλ kχλ is of Γ-compact support
• there exists ∈ Bc such that χλ kχλ = χλcyl χλcyl on ((−∞, −λ] × ∂ M̃ ) ×
((−∞, −λ] × ∂ M̃ ) × T
Lemma 4.1. Ac is a *-subalgebra of A∗ (X, F). Let πc := π|Ac ; there is a short
exact sequence of *-algebras
π
(4.2) 0 → Jc → Ac −→
c
Bc → 0 .
Remark 4.3. Notice that the image of Ac through t|Ac is not contained in Jc
since χ0 is not even continuous. Similarly, the image of Bc through s|Bc is not
contained in Ac .
7
4.3. Relative cyclic cocycles. Let A be a k−algebra over k = C. Recall the

cyclic cohomology groups HC ∗ (A) [1]. Given a second algebra B together with a
surjective homomorphism π : A → B, one can define the relative cyclic complex
Cλn (A, B) := {(τ, σ) : τ ∈ Cλn (A), σ ∈ Cλn+1 (B)} with coboundary map given by
(τ, σ) −→ (π ∗ σ − bτ, bσ) . A relative cochain (τ, σ) is thus a cocycle if bτ = π ∗ σ and
bσ = 0. One obtains in this way the relative cyclic cohomology groups HC ∗ (A, B).
If A and B are Fréchet algebra, then we can also define the topological (relative)
cyclic cohomology groups. More detailed information are given, for example, in [9].
4.4. Roe’s 1-cocycle. In this subsection, and in the next two, we study a
particular but important example. We assume that T is a point and that Γ = {1},
so that we are really considering a compact manifold X0 with boundary ∂X0 and
associated manifold with cylindrical ends X; we keep denoting the cylinder R×∂X0
by cyl(∂X) (thus, as before, we don’t write the subscript 0). The algebras appearing
in the short exact sequence (4.2) are now given by Jc = Cc∞ (X × X),
Bc = {k ∈ C ∞ ((R×∂X0 )×(R×∂X0 )); k is R-invariant, k has compact R-support} .
Finally, a smooth function k on X × X is in Ac if there exists a λ ≡ λ(k) > 0 such
that
(i) k − χλ kχλ is of compact support on X × X;
(ii) ∃ ∈ Bc such that χλ kχλ = χλcyl χλcyl on ((−∞, −λ]×∂X0 )×(−∞, −λ]×∂X0 ) .
For such a k ∈ Ac we define πc (k) = and we have the short exact sequence of
πc
∗-algebras 0 → Jc → Ac −→ Bc → 0 . Incidentally, in the Wiener-Hopf short exact
π
sequence (3.5), which now reads as 0 → C ∗ (X) → A∗ (X) − → B ∗ (cyl(∂X)) → 0,
∗
the left term C (X) is clearly given by the compact operators on L2 (X).
We shall define below a 0-relative cyclic cocycle associated to the homomor-
phism πc : Ac → Bc . To this end we start by defining a cyclic 1-cocycle σ1 for
the algebra Bc ; this is directly inspired from work of John Roe (indeed, a simi-
larly defined 1-cocycle plays a fundamental role in his index theorem on partioned
manifolds [17]).
Consider the characteristic function χλcyl , λ > 0, induced on the cylinder by the
real function χR(−∞,−λ] . For notational convenience, unless absolutely necessary, we
shall not distinguish between χλcyl on the cylinder cyl(∂X) and χλ on the manifold
with cylindrical ends X.
We define σ1λ : BcR × Bc → C as
(4.4) σ1λ (0 , 1 ) := Tr(0 [χλ , 1 ]) .
One can check that the operators [χλ , 0 ] and 0 [χλ , 1 ] are trace class ∀0 , 1 ∈ Bc
(and Tr[χλ , 0 ] = 0). In particular σ1λ (0 , 1 ) is well defined.
Proposition 4.5. The value Tr(0 [χλ , 1 ]) is independent of λ and will simply
be denoted by σ1 (0 , 1 ). The functional σ1 : Bc × Bc → C is a cyclic 1-cocycle.
4.5. Melrose’ regularized integral. Recall that our immediate goal is to
define a relative cyclic 0-cocycle for the homomorphism πc : Ac → Bc appearing
in the short exact sequence of the previous section. Having defined a 1-cocycle σ1
on Bc we now need to define a 0-cochain on Ac . Our definition will be a simple
adaptation of the definition of the b-trace in Melrose’ b-calculus [11] (but since our
algebra Ac is very small, we can give a somewhat simplified treatment). Recall
232
that for λ > 0 we are denoting by Xλ the compact manifold obtained attaching
[−λ, 0] × ∂X0 to our manifold with boundary X0 .
So, let k ∈ Ac with πc (k) = ∈ Bc . Since is R-invariant on the cylinder
cyl(∂X) = R × ∂X0 we can write (y, y , s) with y, y ∈ ∂X0 , s ∈ R. Set

(4.6) r
τ0 (k) := lim k(x, x)dvolg − λ (y, y, 0)dvolg∂
λ→+∞ Xλ ∂X0
where the superscript r stands for regularized. (The b-superscript would be of course
more appropiate; unfortunately it gets confused with the b operator in cyclic coho-
mology.) It is elementary to see that the limit
exists; in fact, because
of the very par-
ticular definition of Ac the function ϕ(λ) := Xλ k(x, x)dvolg −λ ∂X0 (y, y, 0)dvolg∂
becomes constant for large values of λ. The proof is elementary. τ0r defines a 0-
cochain on Ac .
Remark 4.7. Notice that (4.6) is nothing but Melrose’ regularized integral [11],
in the cylindrical language, for the restriction of k to the diagonal of X × X.
We shall also need the following
Lemma 4.8. If k ∈ Ac then t(k), which is a priori a compact operator, is in
fact trace class and τ0r (k) = Tr(t(k)) .
We remark once again that t(k) is not an element in Jc .
4.6. The regularized integral and Roe’s 1-cocycle define a relative 0-
πc
cocycle. We finally consider the relative 0-cochain (τ0r , σ1 ) for the pair Ac −→ Bc .
Proposition 4.9. The relative 0-cochain (τ0r , σ1 ) is a relative 0-cocycle. It
thus defines an element [(τ0r , σ1 )] in the relative group HC 0 (Ac , Bc ).
There are several proofs of this Proposition; we have stated that σ1 is a cocycle
and what needs to be proved now is that bτ0r = (πc )∗ σ1 . One proof of this equality
employs Lemma 4.8; another one use the Hilbert transform and Melrose’ formula
for the b-trace of a commutator [11], see the next Subsection.
4.7. Melrose’ 1-cocycle and the relative cocycle condition via the b-
trace formula. As we have anticipated in the previous subsection, the equation
bτ0r = πc∗ σ1 is nothing but a compact way of rewriting Melrose’ formula for the
b-trace of a commutator. We wish to explain this point here. Following now the
notations of the b-calculus, we consider the sligthy larger algebras
Abc := Ψ−∞
b,c (X, E) , Bcb := Ψ−∞
b,I,c (N+ ∂X, E|∂ ) , Jcb := ρff Ψ−∞
b,c (X, E)
πb
and 0 −→ Jcb −→ Abc −→ c
Bcb −→ 0, with πcb equal to Melrose’ indicial operator
I(·). Let τ0r be equal to the b-Trace: τ0r := b Tr. Observe that σ1 also defines
a 1-cocyle on Bcb . We can thus consider the relative 0-cochain (τ0r , σ1 ) for the
I(·)
homomorphism Abc −−→ Bcb ; in order to prove that this is a relative 0-cocycle it
remains to to show that bτ0r (k, k ) = σ1 (I(k), I(k )), i.e.
(4.10) b
Tr[k, k ] = Tr(I(k)[χ0 , I(k )])
Recall here that Melrose’ formula for the b-trace of a commutator is

i
(4.11) b
Tr[k, k ] = Tr∂X (∂μ I(k, μ) ◦ I(k , μ)) dμ
2π R
9
with C z → I(k, z) denoting the indicial family of the operator k, i.e. the Fourier
transform of the indicial operator I(k).
Inspired by the right hand side of (4.11) we consider an arbitrary compact
manifold Y , the algebra Bcb (cyl(Y )) and the 1-cocycle

i
(4.12) s1 (, ) := ˆ ◦ ˆ (μ) dμ
TrY ∂μ (μ)
2π R
That this is a cyclic 1-cocyle follows by elementary arguments. Formula (4.12)
defines what should be called Melrose’ 1-cocycle
Proposition 4.13. Roe’s 1-cocycle σ1 and Melrose 1-cocycle s1 coincide:

i
(4.14) σ1 (, ) := Tr([χ0 , ]) = ˆ ◦ ˆ (μ) dμ =: s1 (, )
TrY ∂μ (μ)
2π R
Proposition 4.13 and Melrose’ formula imply at once the relative 0-cocyle con-
dition for (τ0r , σ1 ): indeed using first Proposition 4.13 and then Melrose’ formula
we get:

i
σ1 (I(k), I(k )) := Tr(I(k)[χ0 , I(k )]) = Tr∂X (∂μ I(k, μ) ◦ I(k , μ)) dμ
2π R
= b Tr[k, k ] = bτ0r (k, k ) .
Thus I ∗ (σ1 ) = bτ0r as required.
Conclusions. We have seen the following:
• the right hand side of Melrose’ formula defines a 1-cocyle s1 on Bc (cyl(Y )),
Y any closed compact manifold;
• Melrose 1-cocyle s1 equals Roe’s 1-cocyle σ1
• Melrose’ formula itself can be interpreted as a relative 0-cocyle condition
for the 0-cochain (τ0r , s1 ) ≡ (τ0r , σ1 ).
4.8. Philosophical remarks. We wish to recollect the information obtained
in the last three subsections and start to explain our approach to Atiyah-Patodi-
Singer higher index theory.
On a closed compact orientable riemannian smooth manifold Y let us consider
the algebra ofsmoothing operators Jc (Y ) := C ∞ (Y × Y ). Then the functional
Jc (Y ) k → Y k|Δ dvol defines a 0-cocycle τ0 on Jc (Y ); indeed by Lidski’s the-
orem the functional is nothing but the functional analytic trace of the integral
operator corresponding to k and we know that the trace vanishes on commutators;
in formulae, bτ0 = 0. The 0-cocycle τ0 plays a fundamental role in the proof of the
Atiyah-Singer index theorem, but we leave this aside for the time being.
Let now X be a smooth orientable manifold with cyclindrical ends, obtained
from a manifold with boundary X0 ; let cyl(∂X) = R × ∂X0 . We have then defined
algebras Ac (X), Bc (cyl(∂X)) and Jc (X) fitting into a short exact sequence 0 →
πc
Jc (X) → Ac (X) −→ Bc (cyl(∂X)) → 0.
Corresponding to the 0-cocycle τ0 in the closed case we can define two important
0-cocycles on a manifold with cyclindrical ends X:
• We can consider τ0 on Jc (X) = Cc∞ (X × X); this is well defined and does
define a 0-cocycle . We shall refer to τ0 on Jc (X) as an absolute 0-cocycle.
• Starting with the absolute 0-cocycle τ0 on Jc (X) we define a relative
πc
0-cocycle (τ0r , σ1 ) for Ac (X) −→ Bc (cyl(∂X)). The relative 0-cocycle
r
(τ0 , σ1 ) is obtained through the following two fundamental steps.
234
(1) We define a 0-cochain τ0r on Ac (X) by replacing the integral with

Melrose’ regularized integral.
(2) We define a 1-cocycle σ1 on Bc (cyl(∂X)) by taking a suspension of
τ0 through the linear map δ() := [χ0 , ]. In other words, σ1 (0 , 1 )
is obtained from τ0 ≡ Tr by considering (0 , 1 ) → τ0 (0 [χ0 , 1 ]) ≡
τ0 (0 δ(1 )).
Definition 4.1. We shall also call Roe’s 1-cocycle σ1 the eta 1-cocycle corre-
sponding to the absolute 0-cocycle τ0 .
In order to justify the wording of this definition we need to show that all this
has something to do with the eta invariant and its role in the Atiyah-Patodi-Singer
index formula. This will be explained in Section 6 and Section 7.
4.9. The absolute Godbillon-Vey 2-cyclic cocycle τGV . Let (Y, F), Y =
Ñ ×Γ T , be a foliated bundle without boundary. We assume that T = S 1 . Let
E → Y an hermitian complex vector bundle on Y . Let E be the Γ-equivariant
lift of E to Ñ × T . Let (G, s : G → Y, r : G → Y ) be the holonomy groupoid
associated to Y , G = (Ñ × Ñ × T )/Γ. Consider again the convolution algebra
Cc∞ (G, (s∗ E)∗ ⊗ r ∗ E), of equivariant smoothing families with Γ-compact support.
The notation Ψ−∞c (G, E) is also employed. On Ψ−∞ c (G, E) ≡ Cc∞ (G, (s∗ E)∗ ⊗r ∗ E)
we can define a remarkable 2-cocycle, denoted τGV and known as the Godbillon-Vey
cyclic cocycle. First of all, recall that there is a weight ωΓ defined on the algebra
Ψ−∞c (G; E),

(4.15) ωΓ (k) = Tr(ñ,θ) k(ñ, ñ, θ)dñ dθ .
Y (Γ)
In this formula Y (Γ) is the fundamental domain in Ñ × T for the free diagonal
action of Γ on Ñ × T and we have restricted the kernel k to ΔÑ × T ⊂ Ñ × Ñ × T ,
ΔÑ denoting the diagonal set in Ñ × Ñ , ΔÑ × T ≡ Ñ × T ; Tr(ñ,θ) denotes the trace
on End(E (ñ,θ) ). (If the measure on T is Γ-invariant, then this weight is a trace;
however, we don’t want to make this assumption here.) Recall then the bundle E
on Y × T : this is the same vector bundle as E but with a different Γ-action. See
[14] for details. There is a natural identification Ψ−∞ c (G; E) ≡ Ψ−∞ c (G; E ). We
shall consider the linear space Ψ−∞ c (G; E, E
); using the above identification we
can give Ψ−∞
c (G; E, E
) a natural bimodule structure over Ψ−∞
c (G; E). We shall
be interested in the linear functional 2 defined on the bimodule Ψ−∞ c (G; E, E ) by
the analogue of (4.15). To be quite explicit

(4.16) ωΓ (k) = Tr(ñ,θ) k(ñ, ñ, θ)dñ dθ , k ∈ Ψ−∞ c (G; E, E )
Y (Γ)
where we now identify E

(ñ,θ) and E
(ñ,θ) given that thet are identical vector spaces
(it is only the Γ-actions that are different). We call (4.16) the bimodule trace.
(This name come from the following fundamental property: if k ∈ Ψ−∞ c (G; E, E ),
−∞ −∞
k ∈ Ψc (G; E) ≡ Ψc (G; E ) then ωΓ (kk ) = ωΓ (k k).)
Recall now the two derivations δ2 := [φ, ] and δ1 := [φ̇, ] coming from
the modular automorphism group described in [14]. More precisely, we have a
2This will not be a weight, given that on a bimodule there is no notion of positive element
11
derivation δ2 and a bimodule derivation δ1 ,

(4.17) δ2 : Ψ−∞
c (G, E) → Ψ−∞
c (G; E) , δ1 : Ψ−∞
c (G, E) → Ψ−∞
c (G; E, E ) ,
Definition 4.2. With 1 = dim T , the Godbillon-Vey cyclic 2-cocycle on
Cc∞ (G, (s∗ E)∗ ⊗ r ∗ E) is defined to be
1
(4.18) τGV (a0 , a1 , a2 ) = sign(α) ωΓ (a0 δα(1) a1 δα(2) a2 )
2!
α∈S2
with ωΓ the bimodule trace in (4.16).

The fact that this 3-linear functional is indeed a cyclic 2-cocycle is proved
in [14]. We now go back to a foliated bundle (X, F) with cylindrical ends, with
X := M̃ ×Γ T , as in Section 2. We consider the small subalgebras introduced
in Subsection 4.2. The weight ωΓ is still well defined on Jc (X, F); the 2-cocycle
τGV can thus be defined on Jc (X, F), giving us the absolute Godbillon-Vey cyclic
cocycle.
4.10. The eta 3-coycle σGV corresponding to τGV . Now we apply the
general philosophy explained at the end of the previous Section. Let χ0 be the
usual characteristic function of (−∞, 0] × ∂X0 in cyl(∂X) = R × ∂X0 . Write
cyl(∂X) = (R × ∂ M̃ ) ×Γ T with Γ acting trivially on the R factor. Let cyl(Γ) be a
fundamental domain for the action of Γ on (R × ∂ M̃ ) × T ; finally, let ωΓcyl be the
corresponding weight. We keep denoting this weight by ωΓ . Recall the derivation
δ() := [χ0 , ]; recall that we passed from the absolute 0-cocycle τ0 ≡ Tr to the
1-eta cocycle on the cylindrical algebra Bc by considering (0 , 1 ) → τ0 (0 δ(1 )).
We referred to this operation as a suspension.
We are thus led to suspend definition 4.2, thus defining the following 4-linear
functional on the algebra Bc .
Definition 4.3. The eta cochain σGV associated to the absolute Godbillon-Vey
2-cocycle τGV (a0 , a1 , a2 ) is by definition
1
(4.19) σGV (0 , 1 , 2 , 3 ) = sign(α) ωΓ (0 δα(1) 1 δα(2) 2 δα(3) 3 )
3!
α∈S3
with δ3 () := [χ0 , ]. The eta cochain is a 4-linear functional on Bc (cyl(∂X), Fcyl ))
λ
In fact, we can define, as we did for σ1 , the 3-cochain σGV by employing the
λ λ
characteristic function χ . However, one checks easily that the value of σGV does
not depend on λ. One can prove that this definition is well posed, namely that
each term (0 δα(1) (1 ) δα(2) (2 ) δα(3) (3 )) is of finite weight. We then have the
important
Proposition 4.20. The eta functional σGV is cyclic and it is a 3-cocycle:
b σGV = 0.
r
4.11. The relative Godbillon-Vey cyclic cocycle (τGV , σGV ). We now
apply our strategy as in Subsection 4.8. Thus starting with the absolute cyclic
cocycle τGV on Jc (X, F)
we first consider the 3-linear functional on Ac (X, F) given
r 1 r r
by ψGV (k0 , k1 , k2 ) := 2! α∈S2 sign(α) ωΓ (k0 δα(1) k1 δα(2) k2 ) with ωΓ the regular-
ized weight corresponding to ωΓ
236
r
Next we consider the cyclic cochain associated to ψGV :
r 1 r r r
(4.21) τGV (k0 , k1 , k2 ) := (ψGV (k0 , k1 , k2 ) + ψGV (k1 , k2 , k0 ) + ψGV (k2 , k0 , k1 )) .
3
The next Proposition is crucial:
r
Proposition 4.22. The relative cyclic cochain (τGV , σGV ) ∈ Cλ2 (Ac , Bc ) is a
r
relative 2-cocycle: thus bσGV = 0 (which we already know) and bτGV = (πc )∗ σGV .
For later use we also state the analogue of Lemma 4.8:
Proposition 4.23. Let t : A∗ (X, F) → C ∗ (X, F) be the section introduced in
Subsection 3.4. If k ∈ Ac ⊂ A∗ (X, F) then t(k) has finite weight. Moreover, for
the regularized weight ωΓr : Ac → C we have
(4.24) ωΓr = ωΓ ◦ t
5. Smooth subalgebras
In this section we select important subsequences of 0 → C ∗ (X, F) → A∗ (X; F) →
∗
B (cyl(∂X), Fcyl ) → 0.
5.1. Shatten ideals. Let χΓ be a characteristic function for a fundamental
domain of Γ → M̃ → M . Consider Ψ−∞ c (G; E) =: Jc (X F) ≡ Jc . We shall often
omit the bundle E from the notation.
Definition 5.1. Let k ∈ Jc be positive and self-adjoint. The Schatten norm
||k||m of k is defined as
(5.1) (||k||m )m := sup ||χΓ (k(θ))m χΓ ||1
θ∈T
with the || ||1 denoting the usual trace-norm on the Hilbert space Hθ = L2 (Ṽ × {θ}.
Equivalently
(5.2) (||k||m )m = sup ||χΓ (k(θ))m/2 ||2HS .
θ∈T
with || ||HS denoting the usual Hilbert-Schmidt norm. In general, we set ||k||m :=
|| (kk∗ )1/2 ||m . The Schatten norm of k ∈ Jc is easily seen to be finite for any
m ≥ 1; we define Im (X, F) ≡ Im as the completion of Jc with respect to || ||m
One can prove that Im is a Banach algebra and an ideal inside C ∗ (X, F).
Moreover:
Proposition 5.3. The weight ωΓ extends continuously from Jc ≡ Cc∞ (G) to
I1 .
We shall now introduce the subalgebra of C ∗ (X, F) that will be used in the
proof of our index theorem. Consider on the cylinder R × Y (with cylindrical
variable s) the functions

(5.4) fcyl (s, y) := 1 + s2 gcyl (s, y) = 1 + s2 .
We denote by f and g smooth functions on X equal to fcyl and gcyl on the open
subset (−∞, 0) × Y ; f and g are well defined up to a compactly supported function.
We set
(5.5) Jm (X, F) := {k ∈ Im | gk and kg are bounded}
We shall often simply write Jm .
13
Proposition 5.6. Jm is a subalgebra of Im and a Banach algebra with the

norm
(5.7) kJm := km + gkC ∗ + kgC ∗ .
Moreover Jm is holomorphically closed in Im (and, therefore, in C ∗ (X, F)).
5.2. Schatten extensions. Let (Y, F), Y := Ñ ×Γ T , be a foliated T -bundle
without boundary; for example Y = ∂X ≡ ∂X0 . Consider (cyl(Y ), Fcyl ) the asso-
ciated foliated cylinder. Recall the function χ0cyl (often just χ0 ), the function on
the cylinder induced by the characteristic function of (−∞, 0] in R. Notice that
the definition of Schatten norm also apply to (cyl(Y ), Fcyl ), viewed as a foliated
T -bundle with cylindrical ends. Let Ψ−p −p
R,c (Gcyl ) ≡ Ψc (Gcyl /RΔ ) be the space of
R × Γ-equivariant families of pseudodifferential operators of order −p on the fi-
bration (R × Ñ ) × T → T with R × Γ-compact support. Consider an element
∈ Ψ−pc (Gcyl /RΔ ); then we know that defines a bounded operator from the
Sobolev field E k to the Sobolev field E k+p . See [14], Section 3. Let us denote, as in
[14], the operator norm of a bounded operator L from E k to E j as Lj,k ; notice the
reverse order. For a R × Γ-invariant, R × Γ-compactly supported pseudodifferential
operator of order (−p), P , we consider the norm
(5.8) |||P |||p := max(P −n,−n−p , P n+p,n )
with n a fixed integer strictly greater than dim N . We denote the closure of
−p
Ψ−p
c (Gcyl /RΔ ) with respect to the norm ||| · |||p by OP (cyl(Y ), Fcyl ). We shall
−p
often write OP .
Proposition 5.9. OP−p (cyl(Y ), Fcyl ) is a Banach algebra and a subalgebra of
∗
B (cyl(Y ), Fcyl )
Consider now the bounded linear map ∂3max : B ∗ → EndΓ H given by ∂3max :=
[χ , ]. Consider in B ∗ the Banach subalgebra OP−1 (cyl(Y ), Fcyl ) and consider
0
in EndΓ H the subalgebra Jm (cyl(Y ), Fcyl ). Let ∂3 be the restriction of ∂3max to

OP−1 (cyl(Y ), Fcyl ). Since · ≤ ||| · ||| we see that ∂3 is also bounded. Let
D := { ∈ OP−1 (cyl(Y ), Fcyl ) | ∂3 () ∈ Jm (cyl(Y ), Fcyl )}. One can prove
that ∂3 |D induces a closed derivation δ 3 with domain D. This is clearly a closed
extension of the derivation δ3 , δ3 () = [χ0 , ], considered in Subsection 4.10.
Definition 5.2. If m ≥ 1 we define Dm (cyl(Y ), Fcyl ) as Dom δ 3 endowed with
norm
(5.10) Dm := |||||| + [χ0cyl , ]Jm .
We shall often simply write Dm instead of Dm (cyl(Y ), Fcyl ).
Proposition 5.11. Let m ≥ 1, then Dm is a Banach algebra with respect to
(5.10) and a subalgebra of B ∗ ≡ B ∗ (cyl(Y ), Fcyl ). Moreover, Dm is holomorphically
closed in B ∗ .
The Banach algebra we have defined is still too large for the purpose of extend-
ing the eta cocyle. We shall first intersect it with another holomophically closed
Banach subalgebra of B ∗ .
−1
Observe that there exists an action of R on Ψ−1
c (Gcyl /RΔ ) ⊂ OP (cyl(Y ), Fcyl ) ⊂
∗
B defined by
(5.12) αt () := eits e−its ,
238
with t ∈ R, s the variable along the cylinder and ∈ Ψ−1 c (Gcyl /RΔ ). Note that αt ()
is again (R × Γ)-equivariant. It is clear that |||αt ()||| = ||||||; thus, by continuity,
{αt }t∈R yields a well-defined action, still denoted {αt }t∈R , of R on the Banach
algebra OP−1 (cyl(Y ), Fcyl ). Note that this action is only strongly continuous. Let
∂α : OP−1 → OP−1 be the unbounded derivation associated to {αt }t∈R
(αt () − )
(5.13) ∂α () := lim
t→0 t
By definition
Dom(∂α ) = { ∈ OP−1 | ∂α () exists in OP−1 }.
One can prove that the derivation ∂α is in fact a closed derivation.
We endow Dom(∂α ) with the graph norm
(5.14) |||||| + |||∂α ()||| .
It is not difficult to see that Dom(∂α ) is a Banach algebra with respect to (5.14) and,
obviously, a subalgebra of B ∗ ≡ B ∗ (cyl(Y ), Fcyl ); moreover it is holomorphically
closed in B ∗ .
We can now take the intersection of the Banach subalgebras Dm (cyl(Y ), Fcyl )
and Dom(∂α ):
Dm,α (cyl(Y ), Fcyl ) := Dm (cyl(Y ), Fcyl ) ∩ Dom(∂α )
and we endow it with the norm
(5.15) m,α := |||||| + [χ0cyl , ]Jm + |||∂α ||| .
Being the intersection of two holomorphically closed dense subalgebras, also the
Banach algebra Dm,α (cyl(Y ), Fcyl ) enjoys this property.
We are finally ready√to define the subalgebra we are interested in. Recall the
function fcyl (s, y) = 1 + s2 .
Definition 5.3. If m ≥ 1 we define
(5.16)
Bm (cyl(Y ), Fcyl ) := { ∈ Dm,α (cyl(Y ), Fcyl ) | [f, ] and [f, [f, ]] are bounded} .
This will be endowed with norm
Bm :=m,α + 2[f, ]B ∗ + [f, [f, ]]B ∗
=|||||| + [χ0cyl , ]Jm + |||∂α ||| + 2[f, ]B ∗ + [f, [f, ]]B ∗ .
One can prove that Bm (cyl(Y ), Fcyl ) is a holomorphically closed dense subalge-
bra of B ∗ . We shall often simply write Bm instead of Bm (cyl(Y ), Fcyl ).
Let us go back to the foliated bundle with cylindrical end (X, F). We now
define
(5.17) Am (X, F) := {k ∈ A∗ (X, F); π(k) ∈ Bm (cyl(∂X), Fcyl ), t(k) ∈ Jm (X, F)}
Lemma 5.18. Am (X, F) is a subalgebra of A∗ (X, F).
Now we observe that, as vector spaces,
(5.19) Am ∼ = Jm ⊕ s(Bm ) .
Granted this result, we endow Am with the direct-sum norm:
(5.20) ||k||Am := ||t(k)||m + ||π(k)||Bm
15
Obviously s induces a bounded linear map Bm → Am of Banach spaces.

Proposition 5.21. (Am , || ||Am ) is a Banach algebra. Moreover, Jm is an
ideal in Am and there is a short exact sequence of Banach algebras:
π
(5.22) 0 → Jm (X, F) → Am (X; F) −
→ Bm (cyl(∂X), Fcyl ) → 0 .
Finally, t : A∗ (X, F) → C ∗ (X, F) restricts to a bounded section t : Am (X, F) →
Jm (X, F)
5.3. Derivations. In order to extend continuously the cyclic cocycles τGV and
r
(τGV , σGV ) we need to take into account the modular automorphism group, thus
π
decreasing further the size of the short exact sequence 0 → Jm → Am − → Bm → 0.
Consider the two derivations δ1 and δ2 introduced in Subsection 4.9. Let us consider
first δ2 . Recall the C ∗ -algebra CΓ∗ (H) ⊃ C ∗ (X, F); it is obtained, by definition,
by closing up the subalgebra CΓ,c (H) ⊂ EndΓ (H) consisting of those elements
that preserve the continuous field Cc∞ (Ṽ × T, E). We set Dom (δ2max ) = {k ∈
CΓ,c (H) | [φ, k] ∈ CΓ∗ (H)} and
δ2max : Dom (δ2max ) → CΓ∗ (H), δ2max (k) := [φ, k] .
One can prove that δ2max is closable. Similarly, with self-explanatory notation, the
bimodule derivation
δ1max : Dom (δ1max ) → CΓ∗ (H, H ), δ1max (k) := [φ̇, k] ,
max
with Dom (δ1max ) := {k ∈ CΓ,c (H) | [φ̇, k] ∈ CΓ∗ (H, H )} is closable. Let δ j be
their respective closures; thus, for example,
max max
δ2 : Dom δ 2 ⊂ CΓ∗ (H) −→ CΓ∗ (H)
and similarly for δ1max . Define now
max max
D2 := {a ∈ Dom δ 2 ∩ Jm (X, F) | δ 2 a ∈ Jm (X, F)}
max
and δ 2 : D2 → Jm (X, F) as the restriction of to D2 with values in Jm (X, F).
δ2
One can show that δ 2 is a closed derivation. Define similarly D1 and the closed
derivation δ 1 . We set
(5.23) Jm := Jm ∩ Dom(δ 1 ) ∩ Dom(δ 2 ) .
with Dom(δ 1 ) = D1 and Dom(δ 2 ) = D2 .
Consider next Bm ; we consider the derivations δ1 := [φ̇∂ , ], δ2 := [φ∂ , ]
on the cylinder R × ∂X0 ; we have already encountered these derivations in the
definition of the eta cocycle σGV ; see more precisely Definition 4.3. Consider first
δ2 . Define a closed derivation ∂ 2 by taking the closure of the closable derivation
∂2
Ψ−1 ∗ −1
c (Gcyl /RΔ ) −→ B , with ∂2 () := [φ∂ , ] and with Ψc (Gcyl /RΔ ) endowed with
the norm ||| · |||. One can prove that ∂ 2 |D2 , with
D2 = {b ∈ Dom(∂ 2 ) | ∂ 2 (b) ∈ Bm }
is a closed derivation with values in Bm . We set δ 2 := ∂ 2 |D2 ; thus Dom(δ 2 ) = D2
and δ 2 := ∂ 2 |D2 . A similarly definition of δ 1 and Dom(δ 1 ) can be given. We set
(5.24) Bm := Bm ∩ Dom(δ 1 ) ∩ Dom(δ 2 ) ≡ Bm ∩ D1 ∩ D2 .
We endow Bm with the norm
(5.25) Bm := Bm + δ 1 Bm + δ 2 Bm
240
Proposition 5.26. Bm is holomorphically closed in B ∗ (cyl(∂X), Fcyl ).

Finally, we consider the Banach algebra Am (X, F) which is certainly contained
max
in CΓ∗ (H), given that Ac (X, F) is contained in CΓ,c (H). Consider again δ j and
restrict it to a derivation with values in Am (X, F ):
δ 2 : D2 → Am (X, F )
max max
with D2 = {a ∈ Dom δ 2 | δ 2 a ∈ Am (X, F )} and similarly for δ 1 . We obtain in
this way closed derivations δ 1 and δ 2 with domains Domδ 1 = D1 and Domδ 2 = D2 .
We set
(5.27) Am := Am ∩ Dom(δ 1 ) ∩ Dom(δ 2 ) ∩ π −1 (Bm ) .
Lemma 5.28. The map π sends Am into Bm ; Jm is an ideal in Am and we
obtain a short exact sequence of Banach algebras
π
(5.29) 0 → Jm → Am −
→ Bm → 0
The section s and t restricts to bounded sections s : Bm → Am and t : Am → Jm .
Finally, Jm is holomorphically closed in C ∗ (X, F).
π
5.4. Isomorphism of K-groups. Let 0 → J → A − → B → 0 a short exact
sequence of Banach algebras. Recall that K0 (J) := K0 (J + , J) Ker(K0 (J + ) → Z)
and that K(A+ , B + ) = K(A, B). For the definition of relative K-groups we refer,
π
for example, to [5], [9]. Recall that a relative K0 -element for A −
→ B is represented
by a triple (P, Q, pt ) with P and Q idempotents in Mk×k (A) and pt ∈ Mk×k (B) a
path of idempotents connecting π(P ) to π(Q). The excision isomorphism
(5.30) αex : K0 (J) −→ K0 (A, B)
is given by
αex ([(P, Q)]) = [(P, Q, c)]
with c denoting the constant path. Consider also Jm := Jm ∩ Dom(δ 1 ) ∩ Dom(δ 2 )
and recall that this is a smooth subalgebra of C ∗ (X, F): using also the excision
isomorphism, we obtain
(5.31) K0 (A∗ , B ∗ ) K0 (C ∗ (X, F)) K0 (Jm ) K0 (Am , Bm ) .
5.5. Extended cocycles. Recall, from general theory, that [τGV ] ∈ HC 2 (Jc )
r
and [(τGV , σGV )] ∈ HC 2 (Ac , Bc ) can be paired with elements in K0 (Jc ) and K0 (Ac , Bc )
respectively. See the proof of our index formula below for the definition of the rel-
ative pairing. Introduce now the S p−1 operation and
3
S p−1 τGV =: τ2n and (S p−1 τGV r
, S p−1 σGV ) =: (τ2p
r
, σ(2p+1) ).
2p + 1
We obtain in this way cyclic cocycles and thus classes [τ2p ] ∈ HC 2p (Jc ) and
r
[(τ2p , σ(2p+1) )] ∈ HC 2p (Ac , Bc ).
Proposition 5.32. Let 2n equal to the dimension of the leaves in X = M̃ ×Γ S 1 .
Then the absolute cocycle τ2n extends to a bounded cyclic cocycle on J2n+1 and the
eta cocycle σ(2n+1) extends to a bounded cyclic cocycle on B2n+1 .
Proposition 5.33. Let degS p−1 τGVr
= 2p > m(m−1)2 −2 = m3 −2m2 +m−2,
with m = 2n + 1 and 2n equal to the dimension of the leaves in (X, F). Then the
regularized Godbillon-Vey cochain S p−1 τGV
r r
, which is by definition τ2p , extends to
a bounded cyclic cochain on Am .
17
Summarizing: fix m = 2n + 1, with 2n equal to dimension of the leaves and set
J := Jm , A := Am , B := Bm
Using the above two Propositions we see that there are well defined classes
(5.34) [τ2p ] ∈ HC 2p (J) for 2p ≥ 2n
(5.35) r
[(τ2p , σ(2p+1) )] ∈ HC 2p (A, B) for 2p > m(m − 1)2 − 2 .
6. C ∗ -index classes. Excision

6.1. Dirac operators. We begin with a closed foliated bundle (Y, F), with
Y = Ñ ×Γ T . We are also given a Γ-equivariant complex vector bundle E on
Ñ × T , or, equivalently, a complex vector bundle on Y . We assume that E has
a Γ-equivariant vertical Clifford structure. We obtain in this way a Γ-equivariant
family of Dirac operators (Dθ )θ∈T that will be simply denoted by D. If (X0 , F0 ),
X0 = M̃ ×Γ T , is a foliated bundle with boundary, as in the previous sections, then
we shall assume the relevant geometric structures to be of product-type near the
boundary. If (X, F) is the associated foliated bundle with cylindrical ends, then we
shall extend all the structure in a constant way along the cylindrical ends. We shall
eventually assume M̃ to be of even dimension, the bundle E to be Z2 -graded and the
Dirac operator to be odd and formally self-adjoint. We denote by D∂ ≡ (Dθ∂ )θ∈T
the boundary family defined by D+ . This is a Γ-equivariant family of formally
self-adjoint first order elliptic differential operators on a complete manifold. We
denote by Dcyl the operator induced by D∂ ≡ (Dθ∂ )θ∈T on the cylindrical foliated
manifold (cyl(∂X), Fcyl ); Dcyl is R × Γ-equivariant. We refer to [14] [7] for precise
definitions. In all of this section we shall make the following fundamental
Assumption. There exists > 0 such that ∀θ ∈ T
(6.1) L2 − spec(Dθ∂ ) ∩ (−, ) = ∅
For specific examples where this assumption is satisfied, see [7]. We shall concen-
trate on the spin-Dirac case, but it will be clear how to extend the results to general
Dirac-type operators.
6.2. Index class in the closed case. Let (Y, F) be a closed foliated bundle.
First, we need to recall how in the closed case we can define an index class Ind(D) ∈
K∗ (C ∗ (Y, F)). There are in fact three equivalent description of Ind(D), each one
with its own interesting features:
• the Connes-Skandalis index class, defined by the Connes-Skandalis pro-
jector PQ associated to a pseudodifferential parametrix Q for D; Q can
be chosen of Γ-compact support;
• the Wassermann index class, defined by the Wassermann projector WD ;
• the index class of the graph projection, defined by the graph projection
eD .
It is well known that the three classes introduced above are equal in K0 (C ∗ (Y, F)).
242
6.3. The relative index class Ind(D, D∂ ). Let now (X, F) be a foliated
bundle with cylindrical ends. Let (cyl(∂X), Fcyl ) the associated foliated cylinder.
π
Recall 0 → C ∗ (X, F) → A∗ (X; F) − → B ∗ (cyl(∂X), Fcyl ) → 0, the Wiener-Hopf
extension of the C -algebra of translation invariant operators B ∗ (cyl(∂X), Fcyl );
∗
see Subsection 3.4. We shall be concerned with the K-theory group K∗ (C ∗ (X, F))
and with the relative group K∗ (A∗ (X; F), B ∗ (cyl(∂X), Fcyl )). We shall write more
π
briefly 0 → C ∗ → A∗ − → B ∗ → 0, and K∗ (A∗ , B ∗ ). Recall that a relative K0 -cycle
π
for A∗ −→ B ∗ is a triple (P, Q, pt ) with P and Q idempotents in Mk×k (A∗ ) and
pt ∈ Mk×k (B ∗ ) a path of idempotents connecting π(P ) to π(Q).
Proposition 6.2. Let (X, F) be a foliated bundle with cyclindrical ends, as
above. Consider the Dirac operator on X, D = (Dθ )θ∈T . Assume (6.1). Then
the graph projection eD and the Wassermann projection WD define two relative
classes in K0 (A∗ , B ∗ ). These two classes are equal and fix the relative index class
Ind(D, D∂ ).
The relative classs of Proposition 6.2 are more precisely given by the triples
(6.3)
0 0 0 0
(eD , , pt ) with pt := etDcyl and (WD , , qt ) with qt := WtDcyl ,
0 1 0 1
with t ∈ [1, +∞]. The content of the Proposition is that these two triples do define
elements in K0 (A∗ , B ∗ ) and that these two elements are equal.
6.4. The index class Ind(D). Recall the results in [7] where it is proved that
there is a well defined parametrix Q for D+ , QD+ = Id − S+ , D+ Q = Id − S− ,
with remainders S± in K(E) ≡ C ∗ (X, F). Consequently, there is a well defined
Connes-Skandalis projector PQ . The construction explained in [7] is an extension
to the foliated case of the parametrix construction of Melrose, with particular care
devoted to the non-compactness of the leaves.
Definition 6.1. The index class associated to a Dirac operator on (X, F)
satisfying assumption (6.1) is the Connes-Skandalis index class associated to the
Connes-Skandalis projector PQ . It is denoted by Ind(D) ∈ K0 (C ∗ (X, F)).
6.5. Excision for index classes. The following Proposition plays a funda-
mental role in our approach to higher APS index theory:
Proposition 6.4. Let D = (Dθ )θ∈T be a Γ-equivariant family of Dirac opera-
tors on a foliated manifold with cylindrical ends X = M̃ ×Γ T . Assume that M̃ is
even dimensional. Assume (6.1). Then
(6.5) αex ( Ind(D) ) = Ind(D, D∂ )
7. Index theorems
7.1. Notation. From now on we shall fix the dimension of the leaves of (X, F),
equal to 2n, and set
(7.1) J := J2n+1 , A := A2n+1 and B := B2n+1
so that the short exact sequence in (5.29), for m = 2n + 1, is denoted simply as
(7.2) 0→J→A→B→0
19
This is the intermediate subsequence, between 0 → Jc → Ac → Bc → 0 and

0 → C ∗ (X, F) → A∗ (X, F) → B ∗ (cyl(∂X), Fcyl ) → 0, that we have mentioned in
the introductory remarks in Subsection 4.1.
7.2. Smooth index classes. In Sections 6.3 and 6.4 we stated the existence
of two C ∗ -algebraic index classes: the index class and the relative index class. We
have also seen in Subsection 5.5 that the absolute and relative cyclic cyclic cocycles
πc
τGV and (τGV r
, σGV ) extend from Jc and Ac −→ Bc to the smooth subalgebras J
π
and A −→ B. In order to make use of the latter information, we need to smooth-out
our index classes. This is the content of the following
Theorem 7.3.
1)The Connes-Skandalis projector defines a smooth index class Inds (D) ∈ K0 (J);
moreover, if ι∗ : K0 (J) → K0 (C ∗ (X, F)) is the isomorphism induced by the inclu-
sion ι, then ι∗ (Inds (D)) = Ind(D).
2)The graph projections on (X, F) and (cyl(∂X), Fcyl ) define a smooth relative
index class Inds (D, D∂ ) ∈ K0 (A, B); moreover, if ι∗ : K0 (A, B) → K0 (A∗ , B ∗ ) is
the isomorphism induced by the inclusion ι, then ι∗ (Inds (D, D∂ )) = Ind(D, D∂ ).
s
3)Finally, if αex : K0 (J) → K0 (A, B) is the smooth excision isomorphism, then
(7.4) s
αex (Inds (D)) = Inds (D, D∂ ) in K0 (A, B) .
7.3. The higher APS index formula for the Godbillon-Vey cocycle.
We can now state a APS formula for the Godbillon-Vey cocycle. Let us summarize
our geometric data. We have a foliated bundle with boundary (X0 , F0 ), X0 =
M̃ ×Γ T with T = S 1 We assume that the dimension of M̃ is even and that all
our geometric structures (metrics, connections, etc) are of product type near the
boundary. We also consider (X, F), the associated foliation with cylindrical ends.
We are given a Γ-invariant Z2 -graded hermitian bundle E on the trivial fibration
M̃ ×T , endowed with a Γ-equivariant vertical Clifford structure. We have a resulting
Γ-equivariant family of Dirac operators D = (Dθ ).
Fix m = 2n + 1, with 2n equal to dimension of the leaves and set as before
J := Jm , A := Am , B := Bm
We know that there are well defined index classes
Inds (D) ∈ K0 (J) , Inds (D, D∂ ) ∈ K0 (A, B) ,
the first given in terms of a parametrix Q and the second given in term of the
graph projection eD . Proposition 5.32 and Proposition 5.33 imply the existence of
the following additive maps:
(7.5) · , [τ2p ] : K0 (J) → C , 2p ≥ 2n
(7.6) · r
, [(τ2p , σ(2p+1) )] : K0 (A, B) → C , 2p > m(m − 1)2 − 2 .
Definition 7.1. Let (X0 , F0 ), X0 = M̃ ×Γ S 1 , as above and assume (6.1).

The Godbillon-Vey higher index is the number
(7.7) IndGV (D) := Ind(D), [τ2n ].
Notice that, in fact, IndGV (D) := Inds (D), [τ2p ] for each p ≥ n.
The following theorem is the main results of this paper:
244
Theorem 7.8. Let X0 = M̃ ×Γ S 1 be a foliated bundle with boundary and let

D := (Dθ )θ∈S 1 be a Γ-equivariant family of Dirac operators as above. Assume (6.1)
on the boundary family. Fix 2p > m(m − 1)2 − 2 with m = 2n + 1 and 2n equal to
the dimension of the leaves. Then the following two equalities hold

(7.9) IndGV (D) = Inds (D, D∂ ), [(τ2p
r
, σ2p+1 ] = AS ∧ ωGV − ηGV
X0
with
∞
(2p + 1)
(7.10) ηGV := σ(2p+1) ([p˙t , pt ], pt , . . . , pt )dt , pt := etDcyl ,
p! 0
defining the Godbillon-Vey eta invariant of the boundary family and AS denoting
the form induced on X0 by the (Γ-invariant) Atiyah-Singer form for the fibration

M̃ × S 1 → S 1 and the hermitian bundle E.
Notice that using the Fourier transformation the Godbillon-Vey eta invariant
ηGV does depend only on the boundary family D∂ ≡ (Dθ∂ )θ∈S 1 .
Proof. For notational convenience we set τ2p ≡ τGV , τ2p r

≡ τGV
r
and σ(2p+1) ≡
s
σGV . We also write αex instead of αex . The left hand side of formula (7.9) is,
by definition, the [PQ , e1 ], τGV with PQ the Connes-Skandalis projec-
pairing
0 0
tion and e1 := . Here we have also used the remark that IndGV (D) :=
0 1
Inds (D), [τ2p ] for each p ≥ n. Recall that αex ([PQ , e1 ]) is by definition [PQ , e1 , c],
with c the constant path with value e1 . Since the derivative of the constant path
is equal to zero and since τGV r
|J = τGV , using the obvious extension of (4.24), we
obtain at once the crucial relation
(7.11) αex ([PQ , e1 ]), [(τGV
r
, σGV )] = [PQ , e1 ], [τGV ] .
Now we use the excision formula, asserting that αex ([PQ , e1 ]) is equal, as a relative
class, to [eD , e1 , pt ] with pt := etDcyl . Thus
[eD , e1 , pt ], [(τGV
r
, σGV )] = [PQ , e1 ], [τGV ]
which is the first equality in (7.9) (in reverse order). Using also the definition of
the relative pairing we can summarize our results so far as follows:
IndGV (D) := Inds (D), [τGV ]
≡ [PQ , e1 ], [τGV ]
= αex ([PQ , e1 ]), [(τGV
r
, σGV )]
= [eD , e1 , pt ], [(τGV
r
, σGV )]

1 r (2p + 1) +∞
:= τGV (eD − e1 ) + σGV ([p˙t , pt ], pt , . . . , pt )dt
p! p! 1

1 r (2p + 1) +∞
≡ τGV (
eD ) + σGV ([p˙t , pt ], pt , . . . , pt )dt
p! p! 1
with e = (D + s)−1 . Notice that the convergence at infinity of the integral

+∞ D
1
σGV ([p˙t , pt ], pt , . . . , pt )dt follows from the fact that the pairing is well defined.
21
Replace D by uD, u > 0. We obtain, after a simple change of variable in the

integral,

(2p + 1) +∞ 1 r
σGV ([p˙t , pt ], pt , . . . , pt , pt )dt = −Inds (uD), [τGV ] + τGV (
euD )
p! u p!
But the absolute pairing Inds (uD), [τGV ] in independent of u and of course equal
to IndGV (D); thus

(2p + 1) +∞ 1 r
σGV ([p˙t , pt ], pt , . . . , pt , pt )dt = − IndGV (D) + τGV (
euD )
p! u p!
The second summand of the right hand side can be proved to converge as u ↓ 0 to
X0
AS ∧ ωGV (this employs Getzler rescaling exactly as in [14]). Thus the limit
+∞
(2p + 1)
lim σGV ([p˙t , pt ], pt , . . . , pt , pt )dt
p! u↓0 s

exists 3 and is equal to X0 AS ∧ ωGV − IndGV (D). The theorem is proved
Remark. The classic Atiyah-Patodi-Singer index theorem in obtained proceeding

as above, but pairing the index class with the absolute 0-cocycle τ0 and the relative
index class with the relative 0-cocycle (τ0r , σ1 ). Equating the absolute and the
relative pairing, as above, we obtain an index theorem. It can be proved that this
is precisely the APS index theorem on manifolds with cylindrical ends; in other
words, the eta-term we obtain is precisely the APS eta invariant for the boundary
operator. The classic APS index theorem from the point of view of relating pairing
was announced by the first author in [13]. This approach is also a Corollary of
the main result of the recent preprint of Lesch, Moscovici and Pflaum [8], that
is, the computation of the Connes-Chern character of the relative homology cycle
associated to a Dirac operator on a manifold with boundary in terms of local data
and higher eta cochain for the commutative algebra of smooth functions on the
boundary (see also [3] and [18]). Needless to say, the results in [8] go well beyond
the computation of the index; however, they don’t have much in common with the
non-commutative results presented in this paper.
7.4. Eta cocycles. The ideas explained in the previous sections can be ex-
tended to general cocycles τk ∈ HC k (Cc∞ (G, (s∗ E)∗ ⊗ r ∗ E)); we simply need to
require that these cocycles are in the image of a suitable Alexander-Spanier ho-
momorphism since we can then replace integrals with regularized integrals in the
passage from absolute to relative cocycles. This general theory will be treated else-
where. Here we only want to comment on the particular case of Galois coverings,
since this case illustrates very well the general framework. In this important exam-
ple the techniques of this paper can be used in order to give an alternative approach
to the higher index theory developed in [6], much more in line with the original
treatment given by Connes and Moscovici in their fundamental paper [2].
We now give a very short treatment of this important example, assuming a cer-
tain familiarity with the seminal work of Connes and Moscovici. Let Γ → M̃ → M
be a Galois covering with boundary and let Γ → Ṽ → V be the associated covering
with cylindrical ends. In the closed case higher indeces for a Γ-equivariant Dirac
3the situation here is similar to the one for the eta invariant in the seminal paper of Atiyah-
Patodi-Singer; the regularity there is a consequence of their index theorem
246
operator on M̃ are obtained through Alexander-Spanier cocycles, so we concen-

trate directly on these. Let φ be an Alexander-Spanier p-cocycle; for simplicity we
assume that φ is the sum of decomposable elements given by the cup product of

(i) (i) (i) (i)
Alexander-Spanier 1-cochains: φ = i δf1 ∪ δf2 ∪ · · · ∪ δfp where fj : M̃ → C
is continuous. Here we assume that δfj , δfj (m̃, m̃ ) := (fj (m̃ ) − fj (m̃))
(i) (i) (i) (i)
is Γ-invariant with respect to the diagonal action of Γ on M̃ × M̃ . This is a

non-trivial assumption. We shall omit ∪ from the notation. The cochain φ is a
cocycle (where we recall that for an Alexander-Spanier p-cochain given by a con-
tinuos function φ : M̃ p+1 → C invariant under the diagonal Γ-action, one sets

p+1
δφ(x0 , x1 , . . . , xp+1 ) := 0 (−1)i φ(x0 , . . . , x̂i , . . . , xp+1 )). Always in the closed
case we obtain a cyclic p-cocycle for the convolution algebra Cc∞ (M̃ ×Γ M̃ ) by
setting
1 (i) (i)
(7.12) τφ (k0 , . . . , kp ) = sign(α)ωΓ (k0 δα(1) k1 · · · δα(p) kp ) ,
p! i α∈Sp
(i) (i) (i)

with δj k := [k, fj ]. Notice that [k, fj ] is the Γ-invariant kernel whose value at
(m̃, m̃ ) is given by k(m̃, m̃ )δfj (m̃, m̃ ) which is by definition k(m̃, m̃ )(fj (m̃ ) −
(i) (i)
(i)
fj (m̃)); ωΓ is as usual given by ωΓ (k) = F Trm̃ k(m̃, m̃), with F a fundamental
domain for the Γ-action.
Pass now to manifolds with boundary and associated manifolds with cylindrical
ends. Consider the small subalgebras Jc (Ṽ ), Ac (Ṽ ), Bc (∂ Ṽ × R) appearing in the
(small) Wiener-Hopf extension constructed in Subection 4.2 (just take T =point
πc
there). We write briefly Jc , Ac , Bc and 0 → Jc → Ac −→ Bc → 0. We adopt
the notation of the previous sections. Given φ as above, we can clearly define an
absolute cyclic p-cocycle τφ on Jc . Next, define the (p + 1)-linear functional ψφr on
Ac by replacing the integral in ωΓ with Melrose’ regularized integral. Consider next
the cyclic p-cochain on Ac , call it τφr (k0 , . . . , kp ), defined by
1 r
ψφ (k0 , k1 , . . . , kp ) + ψφr (k1 , . . . , kp , k0 ) + · · · + ψφr (kp , k0 , . . . , kp−1 ) .
p+1
(i)
Finally, introduce the new derivation δp+1 () := [χ0 , ] with χ0 the function on
∂ Ṽ × R induced by the characteristic function of (−∞, 0]. Then the eta cocycle
associated to τφ is given by
(7.13)
1 (i) (i)
σφ (0 , . . . , p+1 ) = sign(α)ωΓ (0 δα(1) 1 · · · δα(p+1) p+1 )
(p + 1)! i α∈Sp+1
It should be possible to prove, using the techniques of this paper, that this is a
cyclic (p + 1)-cocycle for Bc and that (τφr , σφ ) is a relative cyclic p-cocycle for the
pair (Ac , Bc ). σφ is, by definition, the eta cocycle corresponding to τφ .
Proceeding exactly as above, thus introducing suitable smooth algebras, ex-
tending the cyclic cocycles, smoothing out the index classes and equating the
absolute pairing Ind(D̃), [τφ ] with the relative pairing Ind(D̃, D̃∂ ), [τφr , σφ ] one
should obtain a higher (Atiyah-Patodi-Singer)-(Connes-Moscovici) index formula,
23
with boundary correction term given in terms of

∞
σφ ([p˙t , pt ], pt , . . . , pt )dt with pt := etD̃cyl
0
A full treatment of the general theory on foliated bundles, together with this im-
portant particular case will be treated elsewhere.
References
[1] Alain Connes. Noncommutative geometry. Academic Press Inc., San Diego, CA, 1994.
[2] Alain Connes and Henri Moscovici. Cyclic cohomology, the Novikov conjecture and hyperbolic
groups. Topology, 29(3):345–388, 1990.
[3] Ezra Getzler. Cyclic homology and the Atiyah-Patodi-Singer index theorem. In Index theory
and operator algebras (Boulder, CO, 1991), volume 148 of Contemp. Math., pages 19–45.
Amer. Math. Soc., Providence, RI, 1993.
[4] A. Gorokhovsky and John Lott. Local index theory over étale groupoids. J. reine. angew.
Math., 560:151–198, 2003.
[5] Nigel Higson and John Roe. Analytic K-homology. Oxford Mathematical Monographs. Oxford
University Press, Oxford, 2000. Oxford Science Publications.
[6] Eric Leichtnam and Paolo Piazza. The b-pseudodifferential calculus on Galois coverings and a
higher Atiyah-Patodi-Singer index theorem. Mém. Soc. Math. Fr. (N.S.), (68):iv+121, 1997.
[7] Eric Leichtnam and Paolo Piazza. Étale groupoids, eta invariants and index theory. J. Reine
Angew. Math., 587:169–233, 2005.
[8] Matthias Lesch, Henri Moscovici, and Markus Pflaum. Connes-chern character for manifolds
with boundary and eta cochains. Preprint December 2009 (arXiv:0912.0194).
[9] Matthias Lesch, Henri Moscovici, and Markus J. Pflaum. Relative pairing in cyclic cohomol-
ogy and divisor flows. J. K-Theory, 3(2):359–407, 2009.
[10] Matthias Lesch and Markus J. Pflaum. Traces on algebras of parameter dependent pseu-
dodifferential operators and the eta-invariant. Trans. Amer. Math. Soc., 352(11):4911–4936,
2000.
[11] Richard B. Melrose. The Atiyah-Patodi-Singer index theorem, volume 4 of Research Notes
in Mathematics. A K Peters Ltd., Wellesley, MA, 1993.
[12] Richard B. Melrose. The eta invariant and families of pseudodifferential operators. Math.
Res. Lett., 2(5):541–561, 1995.
[13] H. Moriyoshi. The eta invariant, relative cyclic cocycles and quasi-action of groups. Talk in
AMS 1988 Fall Eastern Section Meeting, State College, PA, U.S.A.
[14] Hitoshi Moriyoshi and Toshikazu Natsume. The Godbillon-Vey cyclic cocycle and longitudinal
Dirac operators. Pacific J. Math., 172(2):483–539, 1996.
[15] Hitoshi Moriyoshi and Paolo Piazza. Eta cocycles. preprint July 2009 (arXiv:0907.0173).
[16] Hitoshi Moriyoshi and Paolo Piazza. Eta cocycles, relative pairings and the godbillon-vey
index theorem. preprint January 2011.
[17] John Roe. Partitioning noncompact manifolds and the dual Toeplitz problem. In Operator
algebras and applications, Vol. 1, volume 135 of London Math. Soc. Lecture Note Ser., pages
187–228. Cambridge Univ. Press, Cambridge, 1988.
[18] Fangbing Wu. The higher index theorem for manifolds with boundary. J. Funct. Anal.,
103(1):160–189, 1992.
Graduate School of Mathematics, Nagoya University

E-mail address: moriyosi@math.nagoya-u.ac.jp
Dipartimento di Matematica, Sapienza Università di Roma

E-mail address: piazza@mat.uniroma1.it
Volume 546, 2011
Two exact sequences for lattice cohomology
András Némethi
Dedicated to Henri Moscovici on his 65th birthday.
Abstract. This article is a continuation of [N08], where the lattice coho-

mology of connected negative definite plumbing graphs was introduced. Here
we consider the more general situation of non–degenerate plumbing graphs,
and we establish two exact sequences for their lattice cohomology. The first
is the analogue of the surgery exact triangle proved by Ozsváth and Szabó
for the Heegaard–Floer invariant HF + ; for the lattice cohomology over Z2 –
coefficients it was proved by J. Greene in [Gr08]. Here we prove it over Z,
and we supplement it by some additional properties valid for negative definite
graphs. The second exact sequence is an adapted version which does not mix
the classes of the characteristic elements (spinc –structures); it was partially
motivated by the surgery formula for the Seiberg–Witten invariant obtained in
[BN10]. For this, we define the relative lattice cohomology and we determine
its Euler characteristic in terms of Seiberg–Witten invariants.
1. Introduction
The lattice cohomology {H (Γ)}q≥0 was introduced in [N08]. In its original
q
version, it was associated with any connected negative definite plumbing graph
Γ, or, equivalently, with any oriented 3–manifold, which might appear as the
link of a local complex normal surface singularity. Lattice cohomology (together
with the graded roots) plays a crucial role in the comparison of the analytic and
topological invariants of surface singularities, cf. [N05, N07, N08], see also
[BN10, NN02, N10] for relations with the Seiberg–Witten invariants of the link.
Additionally, the lattice cohomology (conjecturally) offers a combinatorial descrip-
tion for the Heegaard–Floer homology of Ozsváth and Szabó (for this theory see
[OSz03, OSz04, OSz04b] and the long list of articles of Ozsváth and Szabó).
Indeed, in [N08] the author conjectured that ⊕q even/odd Hq (Γ) is isomorphic as
2010 Mathematics Subject Classification. Primary. 32S25, 57M27. Secondary. 14B05, 14J17,
32S45, 57R57.
Key words and phrases. plumbed manifolds, plumbing graphs, 3-manifolds, lattice coho-
mology, surface singularities, rational singularities, Seiberg-Witten invariant, Heegaard-Floer
homology.
The author is partially supported by OTKA Grant K67928.
2011
c 0000
c Mathematical
holder)
1
249
250
2 András Némethi
a graded Z[U ]–module with HFeven/odd

+
(−M (Γ)), where M (Γ) is the plumbed 3–
manifold associated with Γ. (Recall that at this moment there is no combinatorial
definition/characterization of HF + .)
For rational and ‘almost rational’ graphs this correspondence was established
in [OSz03, N05] (see also [NR10] for a different situation and [N08] for related
results). A possible machinery which might help to prove the general conjecture is
based on the surgery exact sequences. They are established for the Heegaard–Floer
theory in the work of Ozsváth and Szabó. Our goal is to prove the analogous exact
sequences for the lattice cohomology. In fact, independently of the above conjecture
and correspondence, the proof of such exact sequences is of major importance, and
they are fundamental in the computation and in finding the main properties of the
lattice cohomologies.
The formal, combinatorial definition of the lattice cohomology permits one to
extend its definition to arbitrary graphs (plumbed 3–manifolds); the connectedness
and negative definiteness assumptions can be dropped. Nevertheless, in the proof of
the exact sequences, we will deal only with non–degenerate graphs (they are those
graphs whose associated intersection form has non–zero determinant).
More precisely, for any graph Γ and fixed vertex j0 , we consider the graphs Γ\j0
and Γ+j0 , where the first one is obtained from Γ by deleting the vertex j0 and adjacent
edges, while the second one is obtained from Γ by replacing the decoration ej0 of
the vertex j0 by ej0 + 1. We will assume that all these graphs are non–degenerate.
Then Theorem 5.5.3 establishes the following long exact sequence.
j0 , Γ and Γ \ j0 are non–degenerate. Then

Theorem A. Assume that the graphs Γ+
q
Aq B Cq
· · · −→ Hq (Γ+
j0 ) −→ H (Γ) −→ H (Γ \ j0 ) −→ H
q q q+1
j0 ) −→ · · ·
(Γ+
is an exact sequence of Z[U ]–modules.

The first 3 terms of the exact sequence (i.e. the H0 –part) were already used in
[OSz03] (see also [N05]), and the existence of the long exact sequence was already
proved over Z2 –coefficients in [Gr08]. Here we establish its validity over Z. In the
proof we not only find the correct signmodifications, but we also replace some key
arguments. (Nevertheless, the proof follows the main steps of [Gr08].)
For negative definite graphs (i.e. when Γ+ j0 , hence Γ and Γ \ j0 too are negative
definite), the above exact sequence has some important additional properties. By
general theory (cf. [N08]), if Γ is negative definite then H0 (Γ) contains a canonical
submodule T and one has a direct sum decomposition H0 = T ⊕ H0red . T is the
analogue of the image of HF ∞ in HF + in the Heegaard–Floer theory. On the other
hand, in Heegaard–Floer theory, by a result of Ozsváth and Szabó, the operator
C0 restricted to T(Γ \ j0 ) is zero (this follows from the fact that the corresponding
cobordism connecting Γ \ j0 and Γ+ j0 comes from a non–negative definite surgery).
The analogue of this result is Theorem 6.1.2:
Theorem B. Assume that Γ+
j0 is negative definite. Consider the exact sequence
0
A0 B C0
0 −→ H0 (Γ+
j0 ) −→ H (Γ) −→ H (Γ \ j0 ) −→ H (Γj0 ) −→ · · ·
0 0 1 +
and the canonical submodule T(Γ\j0 ) of H0 (Γ\j0 ). Then the restriction C0 |T(Γ\j0 )
is zero.
Two exact sequences for lattice cohomology 251
3
Using the above results one proves for a graph Γ with at most n bad vertices (for
the definition, see 6.2) the vanishing Hqred (Γ) = 0 for any q ≥ n (where Hqred = Hq
for q > 0).
H∗ (Γ) has a natural direct sum decomposition indexed by the set of the char-
acteristic element classes (in the case of the Heegaard–Floer, or Seiberg–Witten
theory, they correspond to the spinc –structures of M (Γ)). Namely, H∗ (Γ) =
⊕[k] H∗ (Γ, [k]). In the exact sequence of Theorem A the operators mix these classes.
Theorem 7.2.2 provides an exact sequence which connects the lattice cohomologies
of Γ and Γ \ j0 with fixed (un–mixed) characteristic element classes. More precisely,
for any characteristic element k of Γ, we define a Z[U ]–module {Hqrel (k)}q≥0 , the
relative lattice cohomology associated with (Γ, j0 , k). It has finite Z–rank and it fits
in the following exact sequence:
Theorem C. Assume that Γ and Γ \ j0 are non–degenerate. One has a long exact
sequence of Z[U ]–modules:
Aq Bq Cq
· · · −→ Hqrel (k) −→
rel
Hq (Γ, [k]) −→
rel
Hq (Γ \ j0 , [R(k)]) −→
rel
Hq+1
rel (k) −→ · · ·
where R(k) is the restriction of k (and the operators also depend on the choice of
the representative k).
The existence of such a long exact sequence is predicted and motivated by the
surgery formula for the Seiberg–Witten invariant established in [BN10]: the results
of section 7 resonate perfectly with the corresponding statements of Seiberg–Witten
theory. This allows us to compute the Euler characteristic of the relative lattice
cohomology in terms of the Seiberg–Witten invariants associated with M (Γ) and
M (Γ \ j0 ).
2. Notations and preliminaries

2.1. Notations. First we will introduce the needed notations regarding plumb-
ing graphs and we recall the definition of the lattice cohomology from [N08]. Since
in [loc.cit.] the graphs were connected and negative definite (as the ‘normal’ plumb-
ing representations of isolated complex surface singularity links), at the beginning
we will start with these assumptions.
Let Γ be such a plumbing graph with vertices J and edges E; we set |J | =
s and we fix an order on J . Γ can also be codified in the lattice L, the free
Z–module generated by {Ej }j∈J and the ‘intersection form’ {(Ei , Ej )}i,j , where
(Ei , Ej ) for i = j is 1 or 0 according as whether (i, j) is an edge or not; and
(Ei , Ei ) is the decoration of the vertex i, usually denoted by ei . (The graph is
negative definite if this form is so.) The graph Γ may have cycles, but we will
assume that all the genus decorations are zero (i.e. we plumb S 1 –bundles over S 2 ).
The associated plumbed 3–manifold M (Γ) is not necessarily a rational homology
sphere; this happens exactly when the graph is a tree. Let L be the dual lattice
{l ∈ L ⊗ Q : (l , L) ⊂ Z}; it is generated by {Ej∗ }j , where (Ej∗ , Ei ) = −δij (the
negative of the Kronecker delta). Moreover,
1
Char := {k ∈ L : χk (l) := − (k + l, l) ∈ Z for all l ∈ L}
2
denotes the set of characteristic elements of L (or Γ).
Following Ozsváth and Szabó, we set T0+ for the Z[U ]–module Z[U, U −1 ]/U Z[U ]
with grading deg(U −d ) = 2d (d ≥ 0). More generally, for any r ∈ Q one defines
252
Tr+ , the same module as T0+ , but graded (by Q) in such a way that the (d + r)–
homogeneous elements of Tr+ are isomorphic to the d–homogeneous elements of
T0+ . (E.g., for m ∈ Z, T2m
+
= Z[U, U −1 ]/U −m+1 Z[U ].)
2.2. The lattice cohomology associated with k ∈ Char [N08]. L ⊗ R =

Zs ⊗Z R has a natural cellular decomposition into cubes. The set of zero–dimensional
cubes is provided by the lattice points L. Any l ∈ L and subset I ⊂ J of cardinality
qdefines a q–dimensional cube, which has its vertices in the lattice points (l +

j∈I Ej )I , where I runs over all subsets of I. On each such cube we fix an
orientation. This can be determined, e.g., by the order (Ej1 , . . . , Ejq ), where j1 <
· · · < jq , of the involved base elements {Ej }j∈I . The set of oriented q–dimensional
cubes defined in this way is denoted by Qq (0 ≤ q ≤ s).
Let Cq be the free Z–module generated by oriented cubes q ∈ Qq . Clearly,
for each q ∈ Qq , the oriented boundary ∂q has the form k εk kq−1 for some
εk ∈ {−1, +1}. Here, in this sum, we write only those (q − 1)–cubes which appear
with non–zero coefficient. One sees that ∂ ◦ ∂ = 0, but, obviously, the homology of
the chain complex (C∗ , ∂) is trivial: it is just the homology of Rs . In order to get a
more interesting (co)homology, one needs to consider a weight function w : Qq → Z
(0 ≤ q ≤ s). In the present case this will be defined, for each k ∈ Char fixed, by
w(q ) := max{χk (l) : l is a vertex of q }.
Once the weight function is defined, one considers F q , the set of morphisms
HomZ (Cq , T0+ ) with finite support on Qq . Notice that F q is, in fact, a Z[U ]–module
by (p ∗ φ)(q ) := p(φ(q )) (φ ∈ F q , p ∈ Z[U ]). Moreover, F q has a 2Z–grading:
φ ∈ F q is homogeneous of degree 2d ∈ Z if for each q ∈ Qq with φ(q ) = 0, φ(q )
is a homogeneous element of T0+ of degree 2d − 2 · w(q ).
Next, one defines δ : F q → F q+1 . For this, fix φ ∈ F q and we show how δ(φ)
acts on a cube q+1 ∈ Qq+1 . First write ∂q+1 = k εk kq , then set
k
(2.2.1) (δ(φ))(q+1 ) := εk U w(q+1 )−w(q ) φ(kq ).
k
∗
One verifies that δ ◦ δ = 0, i.e. (F , δ) is a cochain complex (with δ homogeneous
of degree zero); its homology is denoted by {Hq (Γ, k)}q≥0 .
(F ∗ , δw ) has a natural augmentation too. Indeed, set mk := minl∈L {χk (l)}.
Then one defines the Z[U ]–linear map : T2m +
k
−→ F 0 such that (U −mk −s )(l) is
−mk +χk (l)−s
the class of U in T0 for any integer s ≥ 0. Then is injective and
+
homogeneous of degree zero, δ ◦ = 0. The homology of the augmented cochain

complex
δ δ
0 −→ T2m
+
k
−→ F 0 −→ F 1 −→ . . .
is called the reduced lattice cohomology H∗red (Γ, k). For any q ≥ 0, both Hq and
Hqred admit an induced graded Z[U ]–module structure and Hq = Hqred for q > 0.
Note that provides a canonical embedding of T2m +
k
into H0 . Moreover, one
has a graded Z[U ]–module isomorphism H0 = T2mk ⊕ H0red , and H∗red has finite
+
Z–rank.
Remark 2.2.2. For the definition of the lattice cohomology for more general
weight functions and graphs with non–zero genera, see [N08].
5
2.3. Reinterpretation of the lattice cohomology. If k = k + 2l for some

l ∈ L then H∗ (Γ, k) and H∗ (Γ, k ) are isomorphic up to a degree shift, cf. [N08,
(3.3)]. In fact, all the modules {H∗ (Γ, k)}k can be packed into only one object,
more in the spirit of [OSz03]. This was used in [Gr08] too.
In this way, for any fixed k ∈ Char, L is identified with the sublattice k + 2L ∈
Char, and with the notation l := k + 2l one has χk (l) = − 18 (l , l ) + 18 (k, k). In
particular, up to a shift in degree, for each fixed k, l → χk (l) and l → − 18 (l , l )
define the same weight function (on the cubes, see below), hence the same coho-
mology. In fact, we will even modify this weight function by s/8 (in this way the
blowing up will induce a degree preserving isomorphism, cf. §3), and set:
k2 + s
(2.3.1) w : Char → Q, w(k) := − (s = |J |).
8
The q–cubes q ∈ Qq are associated with pairs (k, I) ∈ Char × P(J ), |I| = q (here
P(J ) denotes the power set of J ), and have the form {k + 2 j∈I Ej )I , where I
runs over all subsets of I. The weights are defined by

(2.3.2) w(q ) = w((k, I)) = max

w(k + 2 Ej ) .
I ⊂I
j∈I
Moreover, F q are elements of HomZ (Qq , T0+ ) with finite support. Similarly as
above, F q is a Z[U ]–module with a Q–grading: φ ∈ F q is homogeneous of degree
r if for each q ∈ Qq with φ(q ) = 0, φ(q ) is a homogeneous element of T0+ of
degree r − 2 · w(q ).
It is convenient to consider the module of (infinitely supported) homological
cycles too: let Fq be the direct product of Z≥0 ×Qq copies of Z (considered already in
[OSz03] for q = 0). We write the pair (m, ) as U m . Fq becomes a Z[U ]–module
by U (U m ) = U m+1 . Clearly F q = HomZ[U] (Fq , T0+ ), i.e. φ(U ) = U φ() for
any φ.
δ : F q → F q+1 is defined as in (2.2.1) using the new weight function, or by
δ(φ)() = φ(∂()), where for = (k, I) = (k, {j1 , . . . , jq }) one has
(2.3.3)
q

∂(k, I) = (−1)l U w(k,I)−w(k,I\jl ) (k, I\jl )−U w(k,I)−w(k+2Ejl ,I\jl ) (k+2Ejl , I\jl ) .
l=1
The cohomology of (F ∗ , δ) is denoted by H∗ (Γ). Since the vertices of a cube belong

to the same class Char/2L (where a class has the form [k] = {k + 2l}l∈L ⊂ Char),
H∗ (Γ) has a natural direct sum decomposition
H∗ (Γ) = ⊕[k]∈Char/2L H∗ (Γ, [k]).
In fact, if [k1 ] = [k2 ] then w(k1 ) − w(k2 ) ∈ Z.
Since Γ is negative definite, for each class [k] ∈ Char/2L one has a well–defined
rational number
k2 + s
d[k] := − max = 2 · min w(k).
k∈[k] 4 k∈[k]
Then (cf. 2.2) one has a direct sum decomposition:

(2.3.4) H0 (Γ, [k]) = Td[k]
+
⊕ H0red (Γ, [k]).
Sometimes we write T := ⊕[k] Td[k]
+
for the canonical submodule of H0 , which satisfies
H = T ⊕ Hred .
0 0
254
Following [N05, N08, N10] we define the Euler characteristic of H∗ (Γ, [k]) by

(2.3.5) eu(H∗ (Γ, [k])) := −d[k]/2 + (−1)q rankZ Hqred (Γ, [k]).
q≥0
Remark 2.3.6. For−m any φ ∈ F and ≥ 0 set

q
U − ∗ φ ∈ F q defined as follows:
if φ() = m≥0 am, U , then (U ∗ φ)() = m≥0 am, U −m− . Notice that
−
U (U −1 ∗ φ) = φ (but, in general, U −1 ∗ (U φ) = φ).

For any ∈ Qq let ∨ be that element of F q which sends to 1 ∈ T0+ and
any other element to zero. Then any element of F q is a finite Z–linear combination
of elements of type U − ∗ ∨ .
2.4. Generalizations. Graphs which are not negative definite. Since
the definition of the lattice cohomology is purely algebraic/combinatorial, its def-
inition can be considered for any graph, or even for any lattice L with fixed basis
elements {Ej }j . Nevertheless, in this note we assume that the graphs are non–
degenerate. Of course, doing this generalization, some of the properties of the
lattice cohomology associated with connected negative definite graphs will not sur-
vive. Here we wish to point out some of the differences.
First, we notice that dropping the connectedness of the graph basically has no
effect (this fact already was used in the main inductive proofs of [OSz03] and [N05,
(8.3)]). Dropping the negative definiteness is more serious. In order to explain this,
we will introduce the following spaces: for any fixed class [k] and any real number r
we define Sr as the union of all cubes ∈ Qq with all vertices in [k] and w() ≤ r.
If Γ is negative definite then Sr is compact for any r. In particular, the weight
function takes its minimum on it, and for each class [k] = Char/2L a decomposition
H0 (Γ, [k]) = Td([k])
+
⊕ H0red (Γ, [k]) can be defined. This property will not survive
for general graphs. In fact, if the components of Sr are not compact, then we
even might have the vanishing H0 (Γ) = 0 (and this can happen simultaneously
with the non–vanishing of H1 ), see e.g. (2.4.1). Also, in [N08, Theorem 3.1.12]
the lattice cohomology is recovered from the simplicial cohomology of the spaces
Sr . In the general case, the adapted version of that theorem (with similar proof),
valid for any lattice, states that the same formula is valid once if we replace the
cohomology groups H q (Sr , Z) by the cohomology groups with compact support
Hcq (Sr , Z). Namely
H∗ (Γ, [k]) Hc∗ (Sr , Z).
r∈w(k)+Z
We wish to emphasize that replacing the cohomology with cohomology with com-
pact support is a conceptual modification: the two groups have different functorial
properties.
In fact, exactly this second point of view is crucial in the definition of the lattice
cohomology (and in the definition of the maps in the surgery formula treated next):
basically we mimic the infinitely supported homology and the (dual) cohomology
with compact support, and the corresponding functors associated with them.
Since the surgery exact sequence considered in the next section is valid for
any non–degenerate lattice, it is very convenient to extend the theory to arbitrary
graphs (or at least for non–degenerate ones) in order to have a larger flexibility
for computations. Nevertheless, we have to face the following crucial problem. An
oriented plumbed 3–manifold M (Γ) has many different plumbing representations Γ.
They are connected by the moves of the (oriented) plumbing calculus. For negative
7
definite graphs the only moves are the blowups (and their inverses) by (rational)
(−1)-vertices. In [N08, (3.4)] it is proved that the lattice cohomology is stable
with respect to these moves, see also §3 here. On the other hand, if we enlarge our
plumbing graphs, this stability condition will not survive: the same 3–manifold can
be represented by many different lattices with rather different lattice cohomologies.
More precisely, for negative definite graphs one conjectures (cf. [N08]) that
from the lattice cohomology one can recover (in a combinatorial way) the Heegaard–
Floer homology of Ozsváth–Szabó. In particular, the lattice cohomology carries a
geometric meaning depending only on M (Γ). This geometric meaning is lost in the
context of general graphs (or, at least, it is not so direct).
Example 2.4.1. S 3 can be represented by a graph with one vertex, which has
decoration −1. Computing the lattice cohomology of this graph we get Hq = 0 for
q = 0 and H0 = H0red = T0+ . (This is the Heegaard–Floer homology HF + (S 3 ).)
On the other hand, it is instructive to compute the lattice cohomology of another
graph, which has one vertex, but now with decoration +1. Its lattice cohomology
is Hq = 0 for q = 1 and H1 = T−1/2
+
. This graph also represents S 3 , and the two
graphs can be connected by a sequence of non–empty graphs and ±1–blowups and
blowdowns. In particular, we conclude that the lattice cohomology is not stable
with respect to blowing up/down (+1)–vertices.
Remark 2.4.2. Replacing negative definite graphs with arbitrary non–degenerate

ones we can also adopt the definition of the weight function (2.3.1) by modifying
to
(2.4.3)
−k2 + 3σ + 2s
w† : Char → Q, w† (k) := (σ = signature of ( , ), s = |J |),
8
a more common expression associated with (4–manifold intersection) forms which
are not negative definite. In this note we will not do this, but the interested reader
preferring this expression might shift all the weights below by 3(σ + s)/8.
3. Blowing up Γ
3.1. Since in the main construction we will need some of the operators induced
by blowing down, we will make explicit the involved morphisms.
The next discussion provides a new proof of the stability of the lattice coho-
mology in the case when we blow up a vertex (for the old proof see [N08, (3.4)]).
Starting from now, the graph is neither necessarily connected nor necessarily
negative definite. Nevertheless, we will assume that the intersection form is non–
degenerate.
We assume that Γ is obtained from Γ by ‘blowing up the vertex j0 ’. More
precisely, Γ denotes a graph with one more vertex and one more edge: we glue to
the vertex j0 by the new edge the new vertex Enew with decoration −1 (and genus
0), while the decoration of Ej0 is modified from ej0 into ej0 − 1, and we keep all
the other decorations. We will use the notations L(Γ), L(Γ ), L (Γ), L (Γ ). Let
w : Char(Γ ) → Q be the weight function of Γ defined similarly as in (2.3.1). (We
may use the following convention for the ordering of the indices: if j = j0 , then
j < j0 < jnew .) The following facts can be verified:
256
3.1.1. Consider the maps π∗ : L(Γ ) → L(Γ) defined by

π∗ ( xj Ej + xnew Enew ) = xj E j ,
and π ∗ : L(Γ) → L(Γ ) defined by

π∗( xj E j ) = xj Ej + xj0 Enew .
Then (π ∗ x, x )Γ = (x, π∗ x )Γ . This shows that
(π ∗ x, π ∗ y)Γ = (x, y)Γ and (π ∗ x, Enew )Γ = 0
for any x, y ∈ L(Γ). Both π∗ and π ∗ extend over L ⊗ Q and L .
3.1.2. Define the (nonlinear) map: c : L (Γ) → L (Γ ), c(l ) := π ∗ (l ) + Enew .
Then c(Char(Γ)) ⊂ Char(Γ ) and c induces an isomorphism between the orbit
spaces Char(Γ)/2L(Γ) Char(Γ )/2L(Γ ). Moreover, for any k ∈ Char(Γ) and
k := c(k) one has w (k ) = w(k).
3.1.3. π∗ (Char(Γ )) ⊂ Char(Γ). For any k ∈ Char(Γ ) write k := π∗ (k ).
Then k = π ∗ (k) + aEnew for some odd integer a, and w (k ) − w(k) = a 8−1 ∈ Z≥0 .
2
The maps c and π∗ can be extended to the level of cubes and complexes as
follows.
3.1.4. For any q ≥ 0 and = (k , I) ∈ Qq (Γ ) one defines π∗ ((k , I)) :=
(π∗ (k ), I) ∈ Qq (Γ), provided that jnew ∈ I. By (3.1.3) one has wΓ ()−wΓ (π∗ ()) ≥
0. This defines a homological morphism π∗h : Fq (Γ ) → Fq (Γ) by
w ()−w (π ())
U Γ Γ ∗
π∗ () if jnew ∈ I,
(3.1.5) π∗h () =
0 else.
Using (2.3.3) one verifies that π∗h ◦ ∂ = ∂ ◦ π∗h , hence π∗h is morphism of homological
complexes. In particular, its dual π∗c : F q (Γ) → F q (Γ ), defined by π∗c (φ) = φ ◦ π∗h ,
satisfies π∗c ◦ δ = δ ◦ π∗c , hence it is a morphism of complexes as well.
3.1.6. Before we extend c to the level of complexes, notice that for any k ∈
Char(Γ)
c(k) + 2Ej if j = j0 ,
c(k + 2Ej ) =
c(k) + 2Ej + 2Enew if j = j0 .
In particular, the pair c(k) and c(k +2Ej0 ) does not form a 1–cube in Γ . Hence,
the
mapping (k, I) → (c(k), I), sending the vertex k+2 j∈I Ej into c(k)+2 j∈I Ej ,
does not commute with the boundary operator. The ‘right’ operator ch : Fq (Γ) →
Fq (Γ ) associates with ch ((k, I))

(c(k), I) if j0 ∈ I,
(c(k), I) + U w(k,I)−w(k+2Ej0 ,I0 ) (c(k) + 2Ej0 , I0 ∪ jnew ) if I = I0 ∪ j0 , j0 ∈ I0 .
In fact, by (3.1.7) below, w(k + 2Ej0 , I0 ) above can be replaced by
w (c(k) + 2Ej0 , I0 ), or even by w (c(k) + 2Ej0 , I0 ∪ jnew ).
By (3.1.2) and a computation one gets (where in the third line I0 is any subset of
J with j0 ∈ I0 ):
(3.1.7)
w((k, I)) = w ((c(k), I)),
w((k + 2Ej0 , I0 )) = w ((c(k) + 2E
j0 , I0 )),
w (c(k) + 2 j∈I Ej + 2Ej0 ) = w (c(k) + 2 j∈I Ej + 2Ej0 + 2Enew ).

0 0
9
These and a (longer) computation shows that ch commutes with the boundary
operator ∂.
3.1.8. Using π∗ ◦ c, the definitions and the first equation of (3.1.7) one gets
π∗h ◦ ch = idF∗ (Γ) . On the other hand, ch ◦ π∗h is not the identity, but it is homotopic
to idF∗ (Γ ) . Indeed, we define the homotopy operator K : F∗ (Γ ) → F∗+1 (Γ ) as
follows. Write any k as cπ∗ (k) + 2aEnew for some a ∈ Z. Then define K((k, I)) as
0 if either jnew ∈ I or a = 0. Otherwise take for K((k, I))

sign(a) · U w(k,I)−w(cπ∗ (k)+2lEnew ,I∪jnew ) (cπ∗ (k) + 2lEnew , I ∪ jnew ),
where the summation is over l ∈ {0, 1, . . . , a − 1} if a > 0 and l ∈ {a, . . . , −1} if
a < 0. (The exponents are non–negative because of (3.1.3).) Then, again by a
computation, ∂ ◦ K − K ◦ ∂ = id − ch ◦ π∗h .
In particular, π∗h and ch induce (degree preserving) isomorphisms of the corre-
sponding lattice cohomologies. In the sequel the operator π∗h will be crucial.
4. Comparing Γ and Γ \ j0
4.1. Notations, remarks. We consider a non–degenerate graph Γ as in 3.1,
and we fix one of its vertices j0 ∈ J . The new graph Γ \ j0 is obtained from Γ by
deleting the vertex j0 and all its adgacent edges. We will denote by L(Γ), L (Γ), L(Γ\
j0 ), L (Γ \ j0 ) the corresponding lattices.
The operator i : L(Γ \ j0 ) → L(Γ), i(Ej,Γ\j0 ) = Ej,Γ identifies L(Γ \ j0 )
with a sublattice of L(Γ). The dual operator (restriction) is R : L (Γ) → L (Γ \
∗ ∗
j0 ), R(Ej,Γ ) = Ej,Γ\j 0
for j = j0 and R(Ej∗0 ,Γ ) = 0. It satisfies (iQ (x), y)Γ =
(x, R(y))Γ\j0 for any x ∈ L (Γ \ j0 ) and y ∈ L (Γ). (Here, Ej,Γ ∗ ∗
respectively Ej,Γ\j 0

are the usual dual generators
of L considered in Γ, respectively in Γ \ j0 .)
Recall that l = ∗
j aj Ej,Γ is characteristic if and only if aj ≡ ej (mod 2).
In particular, R sends characteristic elements into characteristic elements. On the
other hand, iQ does not necessarily preserve characteristic elements.
Although R(Char(Γ)) ⊂ Char(Γ \ j0 ), this operator cannot be extended to the
level of cubes. Indeed, notice that

∗
(4.1.1) R(Ej0 ) = − Ej,Γ\j 0
,
(j,j0 )∈EΓ
where the sum runs over the adjacent vertices of j0 in Γ. In particular, the endpoints
of the 1–cube (k, j0 ) are sent into R(k) and R(k) + 2R(Ej0 ), which cannot be
expressed as combinations of 1–cubes in Γ \ j0 . In the next subsection we will
consider another operator, which can be extended to the level of cubes (and which,
in fact, operates in a different direction than R, cf. 2.4).
4.2. The B–operator. Consider b : L (Γ \ j0 ) → L (Γ) defined by

∗ ∗
aj Ej,Γ\j 0
→ aj Ej,Γ .
j j
Clearly, if k ∈ Char(Γ \ j0 ) then b(k) + aj0 Ej∗0 ,Γ ∈ Char(Γ) for any aj0 with
aj0 ≡ ej0 (mod 2). In order to see how it operates on cubes, notice that in Γ \ j0
for any j ∈ J (Γ \ j0 ) one has

∗ ∗
Ej,Γ\j0 = −ej Ej,Γ\j 0
− Ei,Γ\j 0
,
(i,j)∈EΓ\j0
258
and a similar relation holds in Γ too. Therefore, one gets that

(4.2.1) b(Ej ) = Ej + (Ej0 , Ej )Γ · Ej∗0 ,Γ .
Therefore,
(4.2.2) b(l ) = iQ (l ) + (Ej0 , iQ (l ))Γ · Ej∗0 ,Γ .
In particular, b is a ‘small’ modification of iQ , but this modification is enough to
extend it to the level of cubes. Although the vertices of (k, I) are not sent to the
vertices of a cube (provided that I contains elements adjacent to j0 in Γ), cf. (4.2.1),
nevertheless if we consider all the shifts in the direction of the ‘error’ of (4.2.1) we
get a well–defined operator
Bq : Fq (Γ \ j0 ) → Fq (Γ)
given by

(4.2.3) Bq ((k, I)) := (b(k) + aj0 Ej∗0 ,Γ , I).
aj0 ≡ej0 (mod 2)
(Here we keep the notation ‘B’, since this notation was used in similar contexts, as
in [OSz03, N05, Gr08].)
Bq induces a morphism B q : F q (Γ) → F q (Γ \ j0 ) via (B q φ)() = φ(Bq ).
A straightforward (slightly long) computation shows that Bq commutes with the
boundary operator ∂, hence B ∗ ◦ δ = δ ◦ B ∗ too. In particular, one gets a well–
defined morphism of Z[U ]–modules B∗ : H∗ (Γ) → H∗ (Γ \ j0 ).
4.3. The sign–modified B–operator. In the exact sequences considered in

the next section we will need to modify the B–operator by a sign (compare with the
end of §2 of [OSz03], where the case q = 0 is discussed). The definition depends
on some choices.
For each L–orbit [k] := k +2L(Γ\j0 ) ⊂ Char(Γ\j0 ) we will fix a representative
r[k] ∈ [k]. Hence, for any characteristic element k, with orbit [k], k−r[k] ∈ 2L(Γ\j0 ),
∗
hence (k − r[k] , Ej,Γ\j 0
) ∈ 2Z for any j = j0 . In particular, (k − r[k] , R(Ej0 )) ∈ 2Z
for R(Ej0 ) defined in (4.1.1). Then the modified operator
B q : Fq (Γ \ j0 ) → Fq (Γ)
is given by

(4.3.1) B q ((k, I)) := (−1)n(k,aj0 )/2 · (b(k) + aj0 Ej∗0 ,Γ , I),
aj0 ≡ej0 (mod 2)
where
n(k, aj0 ) := (k − r[k] , R(Ej0 )) + aj0 + ej0 .
∗ ∗
Then, again, B commutes with the boundary operator, hence B : F q (Γ) →
q
F q (Γ \ j0 ) defined by (B φ)() = φ(B q ) commutes with δ, hence it also defines
∗
a Z[U ]–module morphism B : H∗ (Γ) → H∗ (Γ \ j0 ).
∗
Remark 4.3.2. The morphisms B∗ and B do not preserve neither the gradings
nor the orbits Char/2L (i.e. they do not split into a direct sum with respect to
these orbits).
11
5. The surgery exact sequence

5.1. Notations. Let Γ be a non–degenerate graph as above with s = |J | ≥ 2.
We fix a vertex j0 ∈ J . Associated with j0 we consider two other graphs, namely
Γ \ j0 and Γ+
j0 . The second one is obtained from Γ by modifying the decoration ej0
of the vertex j0 into ej0 + 1. In order to stay in our category of objects, we will
assume that both graphs Γ \ j0 and Γ+ j0 are non–degenerate.
Our goal is to establish a long exact sequence connecting the lattice cohomolo-
gies of these three graphs, similar to one which is valid for the Heegaard–Floer
cohomologies provided by surgeries. As usual, in order to define a long exact se-
quence, we need first to determine a short exact sequence of complexes.
The graphs Γ and Γ\j0 will be connected by the B–operator, cf. 4.3. We define
the ‘A’–operator connecting Γ and Γ+ b
j0 as follows. Let Γ be the graph obtained
from Γ by attaching a new vertex jnew (via a single new edge) to j0 , such that the
decoration of the new vertex is −1. Then, for the pair Γ, Γb (since Γ = Γb \ jnew ),
we can apply the construction and results of §4. In particular, we get morphisms
of complexes: B∗ : F∗ (Γ) → F∗ (Γb ) and B ∗ : F ∗ (Γb ) → F ∗ (Γ). Next, since Γ+
j0 is
obtained from Γb by blowing down the new vertex jnew , by § 3 we get morphisms
∗ ∗
of complexes: π∗h : F∗ (Γb ) → F∗ (Γ+j0 ) and π∗ : F (Γj0 ) → F (Γ ). By composition
c + b
we get the A–operators

∗ ∗ ∗ ∗
A∗ := π∗h ◦ B∗ : F∗ (Γ) → F∗ (Γ+
j0 ), and A := B ◦ π∗ : F (Γj0 ) → F (Γ).
c +
In particular, we can consider the short sequence of complexes

∗
A∗
0 → F ∗ (Γ+ ∗ ∗ B
(5.1.1) j0 ) −→ F (Γ) −→ F (Γ \ j0 ) → 0.
Theorem 5.1.2. The short sequence of complexes (5.1.1) is exact.

The proof is given in several steps.
5.2. The injectivity of Aq . Take an arbitrary non–zero φ ∈ F q (Γ+ j0 ). In
order to prove that Aq φ = 0, we need to find f ∈ Fq (Γ) such that φ(Aq (f )) = 0.
Let N be the smallest non–negative integer such that U N +1 φ = 0. Then replacing
φ by U N φ we may assume that U φ = 0. In particular, in f (or in Aq (f )) any term
whose coefficient has the form U n (n > 0) is irrelevant.
Since φ = 0, there exists (k̄, I) ∈ Qq (Γ+j0 ) with φ((k̄, I)) = 0. Write

∗ ∗
(5.2.1) k̄ = aj Ej,Γ+ + E
j ,Γ+
,
j0 0 j0
j∈J
with aj − ej even for all j ∈ J . Since φ is finitely supported, we may assume that
(5.2.2) aj0 is minimal with the property φ((k̄, I)) = 0.
We wish to construct f ∈ Fq (Γ) such that φ(Aq (f ) − (k̄, I)) = 0. For the weight
functions in Γb and Γ+ b +
j0 we will use the notations w , respectively w .
First, set

∗
(5.2.3) k := aj Ej,Γ ∈ Char(Γ).
j
Then
∗ ∗
B0 (k) = kab , where kab = aj Ej,Γb + aEnew .
a≡1 j
260
Set also
∗ ∗
ka := π∗ (kab ) = aj Ej,Γ+ + aE
j ,Γ+
.
j0 0 j0
j
Notice that k1 = k̄. Since π ∗ ka = kab − aEnew , by (3.1.3)

(5.2.4) wb (kab ) = w+ (ka ) + (a2 − 1)/8.

Moreover, B((k, I)) = a≡1 (kab , I) and π∗ ((kab , I)) = (ka , I).
For any I ⊂ I write EI = j∈I Ej . Let δI be 1 if j0 ∈ I , and zero otherwise.
Since
1 1
(5.2.5) wb (kab + 2EI ) − wb (kab ) = aj − (EI , EI )Γb ,
2
2
j∈I
and
1 1 1
(5.2.6) w+ (ka + 2EI ) − w+ (ka ) = aj + δI a − (EI , EI )Γ+ ,
2
2 2 j0
j∈I
we get
a2 − 1 a−1
(5.2.7) wb (kab + 2EI ) − w+ (ka + 2EI ) = − δI · .
8 2
Assume that I
j0 . Then, by (5.2.7), we get that wb (kab , I)−w+ (ka , I) = (a2 −1)/8,
hence
a2 −1 a2 −1
(5.2.8) Aq ((k, I)) = U 8 (ka , I) + (k̄, I) + U 8 (ka , I).
a≡1, a<0 a≡1, a≥3
Both sums are killed by φ, the first one by (5.2.2), the second one by U φ = 0, hence
φ(Aq ((k, I)) − (k̄, I)) = 0.
Next, assume that I j0 . In that case, again by (5.2.7), we get that wb (kab , I)−
w (ka , I) > 0 whenever a ∈ {−1, 1, 3}. For the other three special values we have
+
the following facts. It is convenient to define for each k (represented as in (5.2.3))

the integer

(5.2.9) M (k) := max

{ aj − (EI , EI )Γb }.
I
j∈I
• For a = 1 one has wb (kab , I) − w+ (ka , I) = 0 always.

• If a = 3, then the right hand side of (5.2.7) is 1 − δI , hence wb (kab , I) −
w (ka , I) = 0 if and only if maxI { wb (kab + 2EI ) } can be realized by some I with
+
I j0 . By (5.2.5) this is equivalent to

(5.2.10) M (k) can be realized by some I with I j0 .
• If a = −1, then the right hand side of (5.2.7) is δI , hence wb (kab , I) −
+
w (ka , I) = 0 if and only if
(5.2.11) M (k) can be realized by some I with I j0 .
Assume that in the case I j0 for a = 3 one has wb (kab , I)−w+ (ka , I) > 0. Then
by an identical argument as in the case I j0 we get that φ(Aq ((k, I))−(k̄, I)) = 0.
Hence, the remaining final case is when there exists at least one I ⊂ I which
contains j0 and satisfies (5.2.10). Then
Aq ((k, I) = (k̄, I) + (k̄ + 2Ej∗ ,Γ+ , I) (mod U and ker(φ)).
0 j0
13
Now, we will consider k + 2Ej∗0 ,Γ instead of k. Via the operator Aq it provides

the same cubes as k (where the index set will have a shift a → a − 2) but with
different U m –coefficients. Notice that by (5.2.10) and (5.2.11), M (k + 2Ej∗0 ,Γ ) can
be realized only by subsets I with I j0 , hence (5.2.11) will fail. Therefore,
Aq ((k + 2Ej∗0 ,Γ , I) = (k̄ + 2Ej∗ ,Γ+ , I) + (k̄ + 4Ej∗ ,Γ+ , I) (mod U).
0 j0 0 j0
More generally, for any positive integer , by the same argument

Aq ((k + 2Ej∗0 ,Γ , I) = (k̄ + 2Ej∗ ,Γ+ , I) + (k̄ + (2 + 2)Ej∗ ,Γ+ , I) (mod U).
0 j0 0 j0
Since φ is finitely supported, φ((2+2)Ej∗ ,Γ+ , I)) = 0 for some ≥ 0, let us consider
0 j0
the minimal such . Then Aq modulo U and ker(φ) restricted to the relevant finite
dimensional spaces looks like an ( + 1) × ( + 1) upper triangular matrix whose
diagonal is the identity matrix. Since this is invertible over Z, the result follows:
some linear combinations of the elements Aq ((k + 2tEj∗0 ,Γ , I), 0 ≤ t ≤ , yield (k̄, I)
modulo U and ker(φ).
q
5.3. The surjectivity of B . We provide the same argument as [Gr08]: For
any fixed a ≡ ej0 (mod 2),
q
B U − ∗ (b(k) + aEj∗0 ,Γ , I)∨ = ±U − ∗ (k, I)∨ .
Since the collection of U − ∗ (k, I)∨ generates F q (Γ \ j0 ) (over Z), the surjectivity
follows.
q
5.4. B ◦ Aq = 0. Take an arbitrary (k, I) ∈ Qq (Γ \ j0 ). Then, by (5.2.7), one
has
c+ej a2 −1
π∗ b(k) + (a + c)Ej∗ ,Γ+ , I .
0
(Aq ◦ B q )((k, I)) = (−1)N (k)+ 2 ·U 8
0 j0
a≡1 c≡ej0
Those pairs (a, c) for which a2 − 1 and a + c are fixed hit the same element of Fq .
Write a = 2i + 1. Then the two solutions of (a2 − 1)/8 = i(i + 1)/2 = t satisfy
i1 + i2 = −1. Since the corresponding two c values satisfy 2i1 + c1 = 2i2 + c2 , one
gets that (c2 − c1 )/2 is odd. Hence the terms cancel each other in pairs.
q
5.5. ker B ⊂ im Aq . Set ker U m := {φ ∈ F q (Γ) : U m φ = 0}. Notice that
the inclusion
q
(5.5.1) ker U m ∩ ker B ⊂ im Aq
for m = 1 (together with (5.4)) implies by induction its validity for any m (cf.
[Gr08]). Indeed, assume that the inclusion is true for m − 1 and set φ ∈ ker U m ∩
q
ker B . Then U φ = Aq (ψ) for some ψ. Moreover, φ̃ := φ−Aq (U −1 ∗ψ) ∈ ker U . On
q
the other hand, by (5.4), φ̃ ∈ ker B too. Therefore, by (5.5.1) applied for m = 1,
we get φ̃ ∈ im Aq , hence φ ∈ im Aq too.
q q q
Notice that ∪m (ker U m ∩ ker B ) = ker B , hence this would prove ker B ⊂
im Aq too.
q
Next, we show (5.5.1) for m = 1. First notice that ker U ∩ ker B is generated
by elements of type (k, I)∨ +(k +2Ej∗0 ,Γ , I)∨ where I j0 , and (k, I)∨ where I j0 .
In the first case (i.e. I j0 ), let us write k as in (5.2.3), and set k̄ as in (5.2.1).
Then by (5.2.7) and (5.2.8) one has
Aq ((k, I)) = (k̄ − 2Ej∗ ,Γ+ , I) + (k̄, I) (mod U),
0 j0
262
Aq ((k + 2Ej∗0 ,Γ , I)) = (k̄, I) + (k̄ + 2Ej∗ ,Γ+ , I) (mod U),

0 j0
and (k̄, I) does not appear in any other term with non–zero coefficient (mod U).
Hence Aq (k̄, I)∨ = (k, I)∨ + (k + 2Ej∗0 ,Γ , I)∨ .
Next, we fix an element of type (k, I)∨ with I j0 . It belongs to the collection
{(k(i), I)}a∈Z , where k(i) = k+2iEj∗0 ,Γ , which will be treated simultaneously via the
discussions of (5.2). Write k as in (5.2.3) and set k̄ via (5.2.1); it is also convenient
to write k̄(i) := k̄ + 2iEj∗ ,Γ+ . Notice that k(i) has coefficients {{aj }j=j0 , aj0 + 2i}.
0 j0
Therefore, for i 0 (5.2.10) will fail and (5.2.11) is satisfied. Let i0 − 1 be maximal
when (5.2.10) fails. Then for i = i0 both conditions are satisfied, and for i > i0
only (5.2.10). Therefore,
(5.5.2)
Aq ((k(i), I)) = (k̄(i − 1), I) + (k̄(i), I) (mod U) if i < i0 ,
Aq ((k(i0 ), I)) = (k̄(i0 − 1), I) + (k̄(i0 ), I) + (k̄(i0 + 1), I) (mod U)
Aq ((k(i), I)) = (k̄(i), I) + (k̄(i + 1), I) (mod U) if i > i0 .
This reads as
⎧
⎨ (k(i), I)∨ + (k(i + 1), I)∨ if i < i0
∨
q
A ((k̄(i), I) ) = (k(i), I)∨ if i = i0
⎩
(k(i), I)∨ + (k(i − 1), I)∨ if i > i0
Taking finite linear combinations we get that any (k(i), I)∨ is in the image of Aq .
This ends the proof of Theorem (5.1.2). As a corollary we get:
j0 , Γ and Γ \ j0 are non–degenerate.

Theorem 5.5.3. Assume that the graphs Γ+
Then
q
Aq B Cq
· · · −→ Hq (Γ+
j0 ) −→ H (Γ) −→ H (Γ \ j0 ) −→ H
q q q+1
j0 ) −→ · · ·
(Γ+
is an exact sequence of Z[U ]–modules.
6. The exact sequence for negative definite graphs

6.1. Preliminaries. For any graph Γ we write det(Γ) for the determinant of
the negative of the intersection form associated with Γ. If Γ is negative definite
then det(Γ) is obviously positive.
If Γ is negative definite then Γ \ j0 is automatically so for any j0 . Nevertheless,
this is not the case for Γ+ +
j0 (although, if Γj0 is negative definite then Γ is so too).
Lemma 6.1.1. Assume that Γ is negative definite. Then Γ+ j0 is negative definite

if and only if det(Γ) > det(Γ\j0 ). If these conditions are satisfied then Ej∗0 ,Γ ∈ L(Γ)
and (Ej∗0 ,Γ )2 ∈ Z.
Proof. Γ+ +
j0 is negative definite if and only if det(Γj0 ) is positive (provided that
Γ \ j0 is negative definite). But det(Γj0 ) = det(Γ) − det(Γ \ j0 ). The last statement
+
follows from −(Ej∗0 ,Γ )2 = det(Γ \ j0 )/ det(Γ), which cannot be an integer.
In the sequel we assume that all the graphs are negative definite, but not
necessarily connected. The next theorem is an addendum to Theorem 5.5.3. Since
its proof is rather long, it will be published elsewhere.
15
Theorem 6.1.2. Assume that Γ+

j0 is negative definite. Consider the exact se-
quence
0
A0 B C0
0 −→ H0 (Γ+
j0 ) −→ H (Γ) −→ H (Γ \ j0 ) −→ H (Γj0 ) −→ · · ·
0 0 1 +
0
and the canonical submodule T(Γ \ j0 ) of H0 (Γ \ j0 ). Then T(Γ \ j0 ) ⊂ im B ; or,
equivalently, the restriction C0 |T(Γ \ j0 ) is zero.
6.2. Graphs with n bad vertices. We say that a negative definite connected
graph is ‘rational’ if it is the plumbing representation of a the link of a rational sin-
gularity (or, the resolution graph of a rational singularity). They were characterized
combinatorially by Artin; for more details see [N99, N05].
We fix an integer n ≥ 0. We say that a negative definite graph has at most
n ‘bad’ vertices if we can find n vertices {jk }1≤k≤n , such that replacing their dec-
orations ejk by some more negative integers ejk ≤ ej0 we get a graph whose all
connected components are rational. (Notice that this is a generalization of the no-
tion of ‘bad’ vertices of [OSz03]. A graph with at most one bad vertex is called
‘almost rational’ in [N05, N08]. Any ‘star–shaped’ graph, i.e. normal form of a
Seifert manifold, has at most one bed vertex, namely the ‘central’ vertex.)
Theorem 6.2.1. If Γ has at most n bad vertices then Hqred (Γ) = 0 for q ≥ n.
This is a generalization of [N08, (4.3.3)], where it is proved for n = 1 (compare
also with the vanishing theorems of [OSz03, N05]).
Proof. We run induction over n. If n = 0, then all the components of Γ are
rational. By [N05], their reduced lattice cohomology vanishes. This fact remains
true for more components too, since the cohomology of a tensor product of two
acyclic complexes is acyclic.
Assume now that the statement is true for n−1 and take Γ with n bad vertices.
Let j be one of them. Let Γj (−) be the graph obtained from Γ by replacing the
decoration ej by ej − ( ≥ 0). Then consider the long exact sequence (5.5.3)
associated with Γj (−), Γj (− − 1) and Γ \ j0 , for all ≥ 0. Then, by the inductive
step, we get that Hq (Γ) = Hq (Γj (−)) for all and q ≥ n. (Here, in the case n = 1,
Theorem 6.1.2 is also used.) Since for 0 the graph Γj (−) has only n − 1 bad
vertices, all these modules vanish.
In fact, the above statement can be improved as follows.
Theorem 6.2.2. Assume that Γ has at most n ≥ 2 bad vertices {jk }1≤k≤n such
that Γ \ j1 has at most (n − 2) bad vertices. Then Hqred (Γ) = 0 for q ≥ n − 1.
Proof. The proof is same as above, if one eliminates first the vertex j1 .
See [N05, (8.2)(5.b)] for a graph Γ with 2 bad vertices {j1 , j2 } such that Γ \ j1
has only rational components.
7. The ‘relative’ surgery exact sequence

7.1. Preliminaries. The motivation for the next exact sequence is two–fold.
First, the exact sequence (5.5.3) mixes the classes of the characteristic elements.
(Note that the surgery exact sequence valid for the Heegaard–Floer theory — which
is one of out models for the theory — does the same.) These classes, in topolog-
ical language, correspond to the spinc –structures of the corresponding plumbed
264
3–manifolds. It would be desirable to have a surgery exact sequence which does

not mix them, and allows inductively the computation of each H∗ (Γ, [k]) for each
[k] independently.
The second motivation is the main result of [BN10]. This is a surgery for-
mula for the Seiberg–Witten invariant of negative definite plumbed 3–manifolds; it
compares these invariants for Γ and Γ \ j0 for fixed (non–mixed) spinc –structures.
The third term in the main formula of [BN10] comes from a ‘topological’ Poincaré
series associated with the plumbing graph, and its nature is rather different than
the other two terms.
Here our goal is to determine an exact sequence connecting H∗ (Γ, [k̄]) and
∗
H (Γ \ j0 , [R(k̄)]) (where R(k̄) is the restriction of k̄, see 4.1) with the newly defined
third term, playing the role of the relative cohomology. Its relationship with the
Poincaré series used in [BN10] will also be treated.
7.2. The ‘relative’ complex and cohomology. We consider a non–degenerate
graph Γ and j0 one of its vertices.
We fix [k] ∈ Char(Γ \ j0 )/2L(Γ \ j0 ) and a characteristic element km ∈ [k] with
w(km ) = mink∈[k] w(k). Furthermore, we fix a0 satisfying a0 ≡ ej0 (mod 2). Then

ka0 := i(km ) + (i(km ), Ej0 ) + a0 Ej∗0 ∈ Char(Γ). In fact, for any k ∈ [k] one gets

that i(k ) + (i(km ), Ej0 ) + a0 Ej∗0 ∈ Char(Γ) and it is an element of [ka0 ]. For
simplicity we write r0 for (i(km ), Ej0 ) + a0 .
We define
B0,rel : F0 (Γ \ j0 , [k]) → F0 (Γ, [ka0 ])
by

(7.2.1) B0,rel (k ) = i(k ) + (i(km ), Ej0 ) + a0 Ej∗0 = i(k ) + r0 Ej∗0 .
This extends to the level of complexes
B∗,rel : (F∗ (Γ \ j0 , [k]), ∂) → (F∗ (Γ, [ka0 ]), ∂)
by B∗,rel ((k, I)) = (B0,rel (k), I). Its dual
∗
Brel : (F ∗ (Γ, [ka0 ]), δ) → (F ∗ (Γ \ j0 , [k]), δ)
∗
is defined by Brel (φ) = φ ◦ B∗,rel .
By a similar argument as in (5.3) we get that
∗
Brel : F ∗ (Γ, [ka0 ]) → F ∗ (Γ \ j0 , [k]) is surjective.
∗ ∗ ∗
We define the ‘relative’ complex Frel
= Frel (Γ, j0 , [k], a0 ]) via ker(Brel ). Let the
∗ ∗ ∗
cohomology of the complex (Frel , δ) be Hrel = Hrel (Γ, j0 , [k], a0 ). It is a graded
Z[U ]–module. We refer to it as the relative lattice cohomology.
∗ ∗
Note that both Frel and Brel depend on the choice of the representative ka0 of
[ka0 ] and are not invariants merely of the classes [ka0 ] and [k].
Theorem 7.2.2. One has the short exact sequence of complexes
∗ A∗ B∗
0 −→ Frel −→
rel
F ∗ (Γ, [ka0 ]) −→
rel
F ∗ (Γ \ j0 , [k]) −→ 0,
which provides a long exact sequence of Z[U ]–modules:
Aq Bq Cq
· · · −→ Hqrel −→
rel
Hq (Γ, [ka0 ]) −→
rel
Hq (Γ \ j0 , [k]) −→
rel
Hq+1
rel −→ · · ·
(Again, the relative cohomology modules and the operators in the above exact se-
quence depend on the choice of ka0 , and not only on its class [ka0 ].)
17
In the case when Γ is negative definite, the restriction of B0rel to Td[k

+
a ] has its
0
image in Td[k]
+
.
Proposition 7.2.3. Assume that Γ (but not necessarily Γ+

j0 ) is negative defi-
nite. Then
B0rel : Td[k
+
a ]
→ Td[k]
+
is onto.
0
In particular, H∗rel has finite rank over Z. Moreover, one has an exact sequence of
finite Z–modules:
0 −→ H0rel −→ H0red (Γ, [ka0 ]) ⊕ Zn −→ H0red (Γ \ j0 , [k]) −→
H1rel −→ H1red (Γ, [ka0 ]) −→ H1red (Γ \ j0 , [k]) · · ·

where n := wΓ (i(km ) + r0 Ej∗0 ) − min{ wΓ | [ka0 ]} ∈ Z≥0 .
Proof. First note that
1 + r02 (Ej∗0 )2
wΓ (i(k ) + r0 Ej∗0 ) = wΓ\j0 (k ) − .
8
This applied for k = km provides
d[ka0 ] d[k] 1 + r02 (Ej∗0 )2
(7.2.4) n+ = − .
2 2 8
For any l ≥ 0 and k̄ ∈ [ka0 ] set φ̄l ∈ F 0 (Γ, [ka0 ]) defined by
φ̄l (k̄) = U −l−d[ka0 ]/2+wL (k̄) (k̄ ∈ [ka0 ]).
For different l ≥ 0 they generate Td[k
+
a0 ]
⊂ H0 (Γ, [ka0 ]). Similarly, set {φl }l≥0 in
F 0 (Γ \ j0 , [k]), where

φl (k ) = U −l−d[k]/2+w(k ) (k ∈ [k]).
They generate Td[k]
+ 0
. ¿From the above identities and from the definition of Brel one
gets for any l ≥ 0:
B0rel (φ̄n+l ) = φl .
Hence the restriction B0rel : Td[k
+
a0 ]
→ Td[k]
+
is onto and the Z–rank of its kernel is
n.
The reader is invited to recall the definition of the Euler characteristic of the
lattice cohomology from (2.3.5). We define the Euler characteristic of the relative
lattice cohomology by

eu(H∗rel ) := (−1)q rankZ Hqrel .
q≥0
Then the exact sequence of Proposition 7.2.3 and equation (7.2.4) provide
Corollary 7.2.5. With the notation r0 := (i(km ), Ej0 ) + a0 one has
1 + r02 (Ej∗0 )2
eu(H∗rel (Γ, j0 , [k], a0 )) = eu(H∗ (Γ, [ka0 ])) − eu(H∗ (Γ \ j0 , [k])) − .
8
Fix any l ∈ L(Γ \ j0 ), then ka0 + 2l = i(km + 2l) + r0 Ej∗0 , hence we also get
266
Corollary 7.2.6. For any l ∈ L(Γ \ j0 ) one has

(ka0 + 2l)2Γ + |J |
eu(H∗rel (Γ, j0 , [km ], a0 )) = eu(H∗ (Γ, [ka0 ])) −
8
(7.2.7)
(km + 2l) 2
+ |J \ j0 |
−eu(H∗ (Γ \ j0 , [km ])) +
Γ\j0
.
8
7.3. Reinterpretation. Above, we started with an element km of the class
[k], and we constructed one of its extensions ka0 . Since ka0 + i(2l) = i(km + 2l) +
r0 Ej∗0 , we have km + 2l = R(ka0 + 2l) for any l ∈ L(Γ \ j0 ).
This procedure can be inverted. Indeed, let us fix any k̄ ∈ Char(Γ) (which plays
the role of ka0 +i(2l)). Then set R(k̄) and define r0 by the identity r0 Ej∗0 = k̄−iR(k̄).
Finally, define
B0,rel : F0 (Γ \ j0 , [R(k̄)]) → F0 (Γ, [k̄]) by B0,rel (k ) = i(k ) + r0 Ej∗0 ,
∗
whose kernel is the relative complex Frel with cohomology H∗rel (Γ, j0 , k̄). Then
(7.2.7) reads as
(k̄)2Γ + |J |
eu(H∗rel (Γ, j0 , k̄)) = eu(H∗ (Γ, [k̄])) −
8
(7.3.1)
(R(k̄))2Γ\j0 + |J \ j0 |
− eu(H∗ (Γ \ j0 , [R(k̄)])) + .
8
The identity (7.3.1) depends essentially on the choice of the choice of k̄ ∈ Char(Γ).
In fact, even if we fix the class [k̄], the choice of the representative k̄ from the
class [k̄] provides essentially different identities of type (7.3.1): not only the terms
eu(H∗rel (Γ, j0 , k̄)), (k̄)2Γ and (R(k̄))2Γ\j0 depend on the choice of k̄, but even the class
[R(k̄)].
7.4. The connection with the topological Poincaré series. Let KΓ ∈ L

denote the canonical characteristic element of Γ defined by the adjunction formulae
(KΓ + Ej , Ej ) + 2 = 0 for all j ∈ J . Similarly, one defines KΓ\j0 ∈ Char(Γ \ j0 ).
Note that Char(Γ) = K + 2L (Γ) and KΓ\j0 = R(KΓ ). The next result computes
eu(H∗rel (Γ, j0 , K + 2l )) in terms of l via the coefficients of a series associated with
Γ.
Consider the multi-variable Taylor expansion Z(t) = pl tl at the origin of
∗
(7.4.1) (1 − tEj )δj −2 ,
j∈J
l
where for any l = j lj Ej ∈ L we write tl = j tjj , and δj is the valency of j.
This lives in Z[[L ]], the submodule of formal power series Z[[t±1/d ]] in variables
±1/d
{tj }j , where d = det(Γ). The series Z(t) was used in several articles studying
invariants of surface singularities, see [CDG04, CDG08, CHR04, N08b, N08c,
N10] for different aspects.

For any series S(t) ∈ Z[[L ]], S(t) = l cl tl , we have the natural decomposi-
tion

S= Sh , where Sh := cl tl .
h∈L /L l : [l ]=h
19
In particular, for any fixed class [l ] ∈ L /L, one can consider the component Z[l ] (t)
of Z(t). In fact, see e.g. [N08b, (3.1.20)],
1 ∗
(7.4.2) Z[l ] (t) = ρ([l ])−1 · (1 − ρ([Ej∗ ])tEj )δj −2 ,
d ρ∈(L /L) j∈J

where (L /L) is the Pontrjagin dual of L /L.
Furthermore, once the vertex j0 of Γ is fixed, for any class [l ] ∈ L /L we set

H[l ],j0 (t) := Z[l ] (t) tj0 =td ∈ Z[[t]].
tj =1 for j=j0
i
Let S(t) = i≥0 ci t be a formal power series. Suppose that for some positive
pn−1
integer p, the expression i=0 ci is a polynomial Pp (n) in the variable n. Then
the constant term of Pp (n) is independent of p. We call this constant term the
periodic constant of S and denote it by pc(S) (cf. [NO09]).

Proposition 7.4.3. Fix the vertex j0 of Γ and write H[l ],j0 (t) as i≥0 ci ti .

(a) If l = j aj Ej∗ = j lj Ej ∈ L (Γ) with all aj sufficiently large then

ci = eu(H∗rel (Γ, j0 , K + 2l )).
i<dlj
0

(b) Take ¯l = j ¯lj Ej ∈ L (Γ) with ¯lj 0 ∈ [0, 1). Then
pc(H[l̄ ],j0 ) = eu(H∗rel (Γ, j0 , K + 2¯l )).
Proof. Use (7.3.1) from above and the identities (3.2.7) and (3.2.13) from
[N10].
7.5. The connection with the Seiberg–Witten invariants. Let M (Γ)
be the oriented plumbed 3–manifold associated with Γ, and −M (Γ) the same 3–
manifold with opposite orientation. In is known that the spinc –structures of M (Γ)
(and of −M (Γ) too) can be identified with Char/2L, see e.g. [N05, N08]. Let
sw(M (Γ), [k̄]) be the Seiberg–Witten invariant of M (Γ) associated with the spinc –
structure [k̄]. Then, by Theorem B of the Introduction of [N10] for a negative
definite graph Γ one has
(7.5.1) eu(H∗ (Γ, [k̄])) = sw(−M (Γ), [k̄]).
Hence the above statements can be reinterpreted in terms of Seiberg–Witten in-
variants as well.
Example 7.5.2. Let Γ be the next graph, where all the vertices have decoration
−2 except the j0 vertex which has −3.
j0
s s s s s s s s
s
det(Γ) = det(Γ \ j0 ) = 1, hence for both graphs we have only one class.
Moreover, Γ\j0 is rational (an E8 –graph), hence min wΓ\j0 = 0, H∗red (Γ\j0 ) = 0
and eu(H∗ (Γ \ j0 )) = 0.
On the other hand, Γ is minimally elliptic, min wΓ = 0, H∗red (Γ) = H0red (Γ) =
Z(0) , the rank one Z–module concentrated at degree zero. Hence eu(H∗ (Γ)) = 1.
268
By the long exact sequence, we get Hqrel (Γ, j0 , k̄) = 0 for any k̄ and q > 0.
It is easy to see that KΓ = −Ej∗0 , and (Ej∗0 )2 = −1. Therefore, if k̄ = K + 2l ,
and r0 := −(k̄, Ej∗0 ) and lj 0 := −(l , Ej∗0 ), then r0 = 2lj 0 − 1. By (7.2.5) or (7.3.1)
r02 + 7 lj (lj − 1)
rank H0rel (Γ, j0 , k̄) = eu(H∗ (Γ, j0 , k̄)) = =1+ 0 0 .
8 2
By a computation one obtains
1 − t6 1 − t + t2
H[0],j0 (t) = = = 1 + t + 2t2 + 3t3 + 4t4 + · · · .
(1 − t3 )(1 − t2 )(1 − t) (1 − t)2
Then
lj 0 (lj 0 − 1)
ci = 1 + 1 + 2 + 3 + · · · + (lj 0 − 1) = 1 + ,
2
i<lj
0
hence (7.4.3)(a) follows. In order to realize part (b), we have to take l̄ with l̄j 0 = 0,
hence in this case eu(H∗ (Γ, j0 , k̄)) = 1. But one also has
1 − t + t2
pc = 1.
(1 − t)2
Example 7.5.3. Assume that Γ is a star–shaped graph with central vertex j0
and we fix k̄ = KΓ .
Since all the connected components of Γ \ j0 are strings (i.e. rational graphs),
for them (cf. [N08])
(KΓ\j0 )2 + |J \ j0 |
eu(H∗ (Γ \ j0 , [KΓ\j0 ])) − = 0.
8
Moreover, Hq (Γ, [K]) = Hqrel (Γ, j0 , [K]) = 0 for q > 0, and
K 2 + |J |
rank H0rel (Γ, j0 , [K]) = eu(H∗ (Γ, [K]) −
8
(7.5.4) K 2 + |J |
= rank H0 (Γ, [K]) − min wΓ −
8
= rank H0 (Γ, [K]) − min χK .
This equals the periodic constant of H[0],j0 (t) by [N08, BN10]. Moreover, if
(X, o) is a weighted homogeneous normal surface singularity with minimal good
resolution graph Γ, then its geometric genus pg equals the last term of (7.5.4), cf.
e.g. [NN04, N05]. Hence pg (X, o) = rank H0rel (Γ, j0 , [K]).
References
[BN10] Braun, G. and Némethi, A., Surgery formula for Seiberg–Witten invariants of negative
definite plumbed 3-manifolds, Journal für die Reine und Angewandte Mathematik 638 (2010),
189–208.
[CDG04] Campillo, A., Delgado, F. and Gusein-Zade, S. M., Poincaré series of a rational surface
singularity, Invent. Math. 155(1) (2004), 41–53.
[CDG08] Guseı̆n-Zade, S. M., Delgado, F. and Campillo, A., Universal abelian covers of rational
surface singularities and multi-index filtrations, Funk. Anal. i Prilozhen. 42(2) (2008), 3–10.
[CHR04] Cutkosky, S. D., Herzog, J. and Reguera, A., Poincaré series of resolutions of surface
singularities, Trans. Amer. Math. Soc. 356(5) (2004), 1833–1874.
[Gr08] Greene, J., A surgery triangle for lattice cohomology, arXiv:0810.0862.
21
[N99] Némethi, A., Five lectures on normal surface singularities, lectures delivered at the Summer
School in Low dimensional topology Budapest, Hungary, 1998; Bolyai Society Math. Studies
8 (1999), 269-351.
[N05] Némethi, A., On the Ozsváth-Szabó invariant of negative definite plumbed 3-manifolds,
Geometry and Topology 9 (2005), 991-1042.
[N07] Némethi, A., Graded roots and singularities, in Singularities in geometry and topology,
World Sci. Publ., Hackensack, NJ, 2007, 394–463.
[N08] Némethi, A., Lattice Cohomology of Normal Surface Singulariites, Publ. of RIMS, Kyoto
University, 44 (2) (2008), 507-543.
[N08b] Némethi, A., Poincaré series associated with surface singularities, in Singularities I: Al-
gebraic and Analytic Aspects, Contemporary Math. 474 (2008), 271–299.
[N08c] Némethi, A., The cohomology of line bundles of splice-quotient singularities,
arXiv:0810.4129.
[N10] Némethi, A., The Seiberg–Witten invariants of negative definite plumbed 3–manifolds,
Alg.Geo.arXiv:1003.1254, to appear in the Journal of EMS.
[NN02] Némethi, A. and Nicolaescu, L.I., Seiberg-Witten invariants and surface singularities,
Geometry and Topology, 6 (2002), 269–328.
[NN04] Némethi, A. and Nicolaescu, L.I., Seiberg-Witten invariants and surface singularities II
(singularities with good C∗ -action), Journal of London Math. Soc. 69(2) (2004), 593–607.
[NO09] Némethi, A. and Okuma, T., On the Casson Invariant Conjecture of Neumann-Wahl,
Journal of Algebraic Geometry 18 (2009), 135–149.
[NR10] Némethi, A. and Román, F., The lattice cohomology of S−d 3 (K), submitted.
[OSz03] Ozsváth, P.S. and Szabó, Z., On the Floer homology of plumbed three-manifolds, Geom-
etry and Topology 7 (2003), 185–224.
[OSz04] Ozsváth, P.S. and Szabó, Z., Holomorphic disks and topological invariants for closed
three-manifolds, Ann. of Math. (2) 159(3) (2004), 1027–1158.
[OSz04b] Ozsváth, P.S. and Szabó, Z., Holomorphic triangle invariants and the topology of sym-
plectic four-manifolds, Duke Math. J. 121(1) (2004), 1–34.
[R04] Rustamov, R., A surgery formula for renormalized Euler characteristic of Heegaard Floer
homology, math.GT/0409294.
A. Rényi Institute of Mathematics, 1053 Budapest, Reáltanoda u. 13-15, Hungary.

E-mail address: nemethi@renyi.hu
Volume 546, 2011
Cup products in Hopf cyclic cohomology with coefficients in

contramodules
Bahram Rangipour
Dedicated to Henri Moscovici, with highest admiration and appreciation.
Abstract. We use stable anti Yetter-Drinfeld contramodules to improve the

cup products in Hopf cyclic cohomology. The improvement fixes the lack of
functoriality of the cup products previously defined and show that the cup
products are sensitive to the coefficients.
1. Introduction
Hopf cyclic cohomology was invented by Alain Connes and Henri Moscovici as a
computational tool for computing the index cocycle of hypoelliptic operators on
manifolds [6]. One of the objectives of the theory was to study the cyclic cocycles
generated by a symmetric system, in the sense of noncommutative geometry, which
is usually given by an action or a coaction of a Hopf algebra on an algebra or a
coalgebra. The main tool for transferring such cocycles to the cyclic complex of
algebras is a characteristic map defined in [6]. The characteristic map is based on
an invariant trace on the algebra of functions on the manifold in question. However
in many situations the invariant trace does not exist, for example see [4]. For such
cases the invariant cyclic cocycles play the role of the invariant trace and one defines
a higher version of the characteristic map [7, 10]. By the generalization of Hopf
cyclic cohomology [12, 11] that allows one to take advantage of coefficients for
Hopf cyclic cohomology, the invariant cyclic cocycles are understood as examples
of Hopf cyclic cocycles. As a result, one generalizes the characteristic map to a cup
product [16]. Similarly, the ordinary cup product in algebras was also generalized to
another type of cup product in Hopf cyclic cohomology by replacing cycles and their
characters with twisted cycles and their twisted characters. In [19, 14], by a direct
application of the cyclic Eilenberg-Zilber theorem ( c.f. [17, 9]), the cup products
were reconstructed and simplified. Finally, it is shown that all cup products defined
in [16, 19, 14, 10] are the same at the level of cohomology [15].
The suitable coefficients for Hopf cyclic cohomology mentioned above is called the
stable anti Yetter-Drinfeld (SAYD) modules [11, 13]. It has both module and
2010 Mathematics Subject Classification. Primary 19D55, 16E40; Secondary 16T05.

Key words and phrases. Hopf cyclic cohomology, cup products, contramodules.
2011
c
1
271
272
2 BAHRAM RANGIPOUR
comodule structure over the Hopf algebra in question, with two compatibilities
made up of composition of action and coaction. However, it is proved that Hopf
cyclic cohomology works with a generalization of SAYD modules called SAYD con-
tramodules [1]. Contramodules for coalgebras was introduced in [8]. A right con-
tramodule of a coalgebra C is a vector space M together with a C-linear map
α : hom(C, M) → M making the diagrams (2.18) commutative.
A SAYD contramodule M is a module and contramodule together with two com-
patibilities made up of α and the action of H on M. As an example, if M is a
SAYD module over H then homk (M, C) is a SAYD contramodule over H.
In this paper, building on the methods we developed in [19], we generalize the
cup products defined in the same paper by using SAYD contramodules coefficients.
By Theorem 4.2 and Theorem 4.3 we show that the cup products are sensitive to
the coefficients. In Section 2 we recall Hopf cyclic cohomology with coefficients in
SAYD modules and contramodules. In Section 3 we define the cup products for a
compatible pair of SAYD modules and contramodules. Here a compatible pair is
a pair of SAYD module and contramodule endowed with a pairing with values in
the ground field and compatible with respect to actions and coactions. Finally, in
Section 4 we generalize the results of Section 3 for arbitrary coefficients without
any compatibility between them. The range of the new cup products is ordinary
cyclic cohomology of algebras with coefficients in vector spaces.
In this note a Hopf algebra is denoted by a sextuple (H, μ, η, Δ, ε, S), where μ, η,
Δ, ε, and S are multiplication, unit, comultiplication, counit, and antipode respec-
tively. We use the Sweedler notation for comultiplications and coactions i.e., for
coalgebras we use Δ(c) = c(1) ⊗ c(2) , for comodules we use (a) = a<0> ⊗ a<1> and
for coefficients we use (m) = m<−1> ⊗ m<0> . All algebras, coalgebras and Hopf
algebras are over the field of complex numbers C. The unadorned tensor product
⊗ denotes ⊗C .
We would like to thank Tomasz Brzeziński for Remark 4.1. We are also grateful to
the referee for his careful reading of the manuscript and his valuable comments.
2. Hopf cyclic cohomology with coefficients

2.1. Stable anti Yetter-Drinfeld-module. For the reader’s convenience,
we briefly recall the definition of Hopf cyclic cohomology of coalgebras and algebras
under the symmetry of Hopf algebras with coefficients in SAYD modules [12], and
with coefficients in SAYD contramodules [1].
Let us recall the definition of SAYD modules over a Hopf algebra from [11]. Given
a Hopf algebra H, we say that M is a right-left SAYD module over H if M is a
right module and left module over H with the following compatibilities.
(2.1) M (m · h) = S(h(3) )m<−1> h(1) ⊗ m<0> · h(2)
(2.2) m<0> · m<−1> = m.
The other three flavors, i.e., left-left, left-right, and right-right are defined similarly
[11].
Let C be a H-module coalgebra, that is, a coalgebra endowed with an action, say
from the left, of H such that its comultiplication and counit are H-linear, i.e.,
(2.3) Δ(h · c) = h(1) · c(1) ⊗ h(2) · c(2) , ε(h · c) = ε(h)ε(c).
CUP PRODUCTS IN CUP
HOPFPRODUCTS IN HOPF CYCLIC
CYCLIC COHOMOLOGY COHOMOLOGY
WITH 273
COEFFICIENTS IN CONTRAMODULES
3
Having the datum (H, C, M ), where C is an H-module coalgebra and M an right-

left SAYD over H, one defines in [12] a cocyclic module {CH
n
(C, M ), ∂i , σj , τ }n≥0
as follows.
(2.4) n
CH (C, M ) := M ⊗H C ⊗n+1 , n ≥ 0,
with the following cocyclic structure,
(2.5) n
∂i : CH (C, M ) → CH
n+1
(C, M ), 0≤i≤n+1
(2.6) n
σj : CH (C, M ) → CH
n−1
(C, M ), 0 ≤ j ≤ n − 1,
(2.7) τ : CH (C, M ) → CH (C, M ),
n n
defined explicitly as follows, where we abbreviate c̃ := c0 ⊗ · · · ⊗ cn ,

(2.8) ∂i (m ⊗H c̃) = m ⊗H c0 ⊗ · · · ⊗ Δ(ci ) ⊗ · · · ⊗ cn ,
(2.9) ∂n+1 (m ⊗H c̃) = m<0> ⊗H c0(2) ⊗ c1 ⊗ · · · ⊗ cn ⊗ m<−1> · c0(1) ,
(2.10) σi (m ⊗H c̃) = m ⊗H c0 ⊗ · · · ⊗ (ci+1 ) ⊗ · · · ⊗ cn ,
(2.11) τ (m ⊗H c̃) = m<0> ⊗H c1 ⊗ · · · ⊗ cn ⊗ m<−1> · c0 .
It is checked in [12] that the above graded module defines a cocyclic module.
Similarly, an algebra is called H-module algebra if it is an H-module and its algebra
structure is H-linear. In other words, for any a, b ∈ A and any h ∈ H we have
(2.12) h · (ab) = (h(1) · a)(h(2) · b), h · 1A = ε(h)1A .
Let A be a H-module algebra. One endows M ⊗ A⊗n+1 with the diagonal action
of H and forms CH n
(A, M ) := HomH (M ⊗ A⊗n+1 , C) as the space of H-linear
maps. It is checked in [12] that the following defines a cocyclic module structure
on C n (A, M ).
(∂i ϕ)(m ⊗ ã) = ϕ(m ⊗ a0 ⊗ · · · ⊗ ai ai+1 ⊗ · · · ⊗ an+1 ),
(∂n+1 ϕ)(m ⊗ ã) = ϕ(m<0> ⊗ (S −1 (m<−1> ) · an+1 )a0 ⊗ a1 ⊗ · · · ⊗ an ),
(σi ϕ)(m ⊗ ã) = ϕ(m ⊗ a0 ⊗ · · · ⊗ ai ⊗ 1 ⊗ · · · ⊗ an−1 ),
(τ ϕ)(m ⊗ ã) = ϕ(m<0> ⊗ S −1 (m<−1> ) · an ⊗ a0 ⊗ · · · ⊗ an−1 ).
∗
The cyclic cohomology of this cocyclic module is denoted by HCH (A, M ).
An algebra is called an H-comodule algebra if it is an H comodule and its algebra
structure is H-colinear, which means that
(2.13) (h · a)<0> ⊗ (h · a)<1> = h(1) · a<0> ⊗ h(2) · a<1> .
Similar to the other case, one defines H C n (A, M ) to be the space of all colinear
maps from A⊗n+1 to M . One checks that the following defines a cocyclic module
structure on H C n (A, M ).
(2.14) (∂i ϕ)(ã) = ϕ(a0 ⊗ · · · ⊗ ai ai+1 ⊗ · · · ⊗ an+1 ),

(2.15) (∂n+1 ϕ)(ã) = ϕ(an+1<0> a0 ⊗ a1 · · · ⊗ an−1 ⊗ an ) · an+1<−1> ,
(2.16) (σi ϕ)(ã) = ϕ(a0 ⊗ · · · ⊗ ai ⊗ 1 ⊗ · · · ⊗ an−1 ),
(2.17) (τ ϕ)(a0 ⊗ · · · ⊗ an ) = ϕ(an<0> ⊗ a0 ⊗ · · · ⊗ an−1 ⊗ an−1 ) · an<−1> .
274
4 BAHRAM RANGIPOUR
The cyclic cohomology of this cocyclic module is denoted by H

HC ∗ (A, M ).
2.2. SAYD contramodule. Let us recall SAYD contramodules from [1]. A

right contramodule of a coalgebra H is a vector space M together with a C-linear
map α : Hom(H, M) → M making the following diagrams commutative
Hom(H,α)
Hom(H, Hom(H, M) / Hom(H, M)
Θ α
Hom(Δ,M)
Hom(H ⊗ H, M) / Hom(H, M) α / M,
Hom(ε,M)
(2.18) Hom(C, M) / Hom(H, M)
KKK
KKK ss
sss
KKK ss
y ss
α
K% s
M,
where Θ is the standard isomorphism given by Θ(f )(h ⊗ h ) = f (h)(h ).

Definition 2.1 ([1]). A left-right anti-Yetter-Drinfeld (AYD) contramodule M is
a left H-module (with the action denoted by a dot) and a right H-contramodule
with the structure map α, such that, for all h ∈ H and f ∈ Hom(H, M),
h·α(f ) = α (h(2) ·f (S(h(3) )(−)h(1) )) .
M is said to be stable provided that, for all m ∈ M, α(rm ) = m, where rm : H →
M, h → h·m.
We refer the reader to [1] for more details on SAYD contramodules. If M is an
AYD module, then its dual M = M ∗ is an AYD contramodule (with the sides in-
terchanged) and SAYD modules correspond to SAYD contramodules. For example,
let M be a right-left AYD module (2.1); the dual vector space M = M ∗ is a right
H-module by m ⊗ h → m · h,
(h·f )(m) = f (m·h),
for all h ∈ H, f ∈ M = Hom(M, C) and m ∈ M , and a right H-contramodule with
the structure map α(f )(m) = f (m<−1> )(m<0> ), f ∈ Hom(H, M), and m ∈ M [1].
Let A be a left H-module algebra and M be a left-right SAYD contramodule over
H. We let CH n
(A, M) be the space of left H-linear maps
(2.19) HomH (A⊗n+1 , M),

and, for all 0 ≤ i, j ≤ n, define ∂i : CH
n−1
(A, M ) → CH
n
(A, M), σj : CH
n+1
(A, M) →
CH (A, M), τ : CH (A, M) → CH (A, M), by
n n n
(2.20) ∂i (ϕ)(a0 ⊗ · · · ⊗ an ) = ϕ(a0 ⊗ · · · ⊗ ai ai+1 ⊗ · · · ⊗ an ), 0 ≤ i < n,

(2.21) ∂n (ϕ)(a0 ⊗ · · · ⊗ an ) = α ϕ S −1 (−)·an a0 ⊗ a1 ⊗ · · · ⊗ an−1 ,
(2.22) σj (ϕ)(a0 ⊗ · · · ⊗ an ) = ϕ(a0 ⊗ · · · ⊗ aj ⊗ 1A ⊗ aj+1 ⊗ · · · ⊗ an ),

(2.23) τ (ϕ)(a0 ⊗ · · · ⊗ an ) = α ϕ S −1 (−)·an ⊗ a0 ⊗ · · · ⊗ an−1 .
CUP PRODUCTS IN CUP
WITH 275
5
∗
It is shown in [1] that the above operators define a cocyclic module on CH (A, M).
We denote the cyclic cohomology of CH (A, M) by HCH (A, M). For M = M ∗ ,
∗ ∗
∗ ∗
where M is a SAYD module over H, it is easy to see that CH (A, M) CH (A, M ).
Indeed, let M be a right-left SAYD module and M := Hom(M, C) be the corre-
sponding right-left SAYD contramodule. We define the following maps;
I : CH
n
(A, M ) → CH
n
(A, M), J : CH
n
(A, M) → CH
n
(A, M ),
I(φ)(a0 ⊗ · · · ⊗ an )(m) = φ(m ⊗ a0 ⊗ · · · ⊗ an ),
J (φ)(m ⊗ a0 ⊗ · · · ⊗ an ) = φ(a0 ⊗ · · · ⊗ an )(m).
Proposition 2.2. The above map I is an isomorphism of cocyclic modules.
Proof. It is obvious that I and J are inverse to one another. We shall check
that I commutes with the cyclic structures. It is easy to see that faces, except
possibly the very last one, and degeneracies commute with I. So it suffices to
check that I commutes with the cyclic operators. Indeed,
I ◦ τ (φ)(a0 ⊗ · · · ⊗ an )(m) = τ (φ)(m ⊗ a0 ⊗ · · · ⊗ an )
= φ(m<0> ⊗ S −1 (m<−1> ) · an ⊗ a0 ⊗ · · · ⊗ an−1 )
= I(φ)(S −1 (m<−1> ) · an ⊗ a0 ⊗ · · · ⊗ an−1 )(m<0> )
= τ ◦ I(φ)(a0 ⊗ · · · ⊗ an )(m).

3. Cup products in Hopf cyclic cohomology

In this section we use the same strategy as in [19, 14] to generalize the cup prod-
ucts constructed in the same references. Via these new cup products one has the
luxury to construct cyclic cocycles by using a compatible pair of SAYD modules
and contramodules rather than only a SAYD module.
3.1. Module algebras paired with module coalgebras. Let A be an H-

module algebra and C be a H-module coalgebra acting on A in the sense that there
is a map
(3.1) C ⊗ A → A,
such that for any h ∈ H, any c ∈ C and any a, b ∈ A one has
(3.2) (h · c) · a = h · (c · a)
(3.3) c · (ab) = (c(1) · a)(c(2) · b)
(3.4) c(1) = (c)1
One constructs a convolution algebra B = HomH (C, A), which is the algebra of
all H-linear maps from A to C. The unit of this algebra is given by η ◦ , where
η : C → A is the unit of A. The multiplication of f, g ∈ B is given by
(3.5) (f ∗ g)(c) = f (c(1) )g(c(2) )
Definition 3.1. Let (M, α) be a left-right SAYD contramodule and N be a right-
left SAYD module over H. We call (N, M) compatible if there is a pairing between
276
6 BAHRAM RANGIPOUR
M and N such that
(3.6) n · h | m = n | h · m,
(3.7) n | α(f ) = n<0> | f (n<−1> ),
for all m ∈ M, n ∈ N , f ∈ Hom(H, M), and h ∈ H.
Let (N, M) be compatible as above. We have the following cocyclic modules defined
in (2.19) . . . (2.23), and (2.4) . . . (2.8) respectively:
∗ ∗
(3.8) (CH (A, M), ∂i , σj , τ ), and (CH (C, N ), ∂i , σj , τ ).
We define a new bicocyclic module by tensoring these cocycle modules over C. The
new bigraded module has in its bidegree (p, q)
(3.9) p,q
Ca−c := HomH (A⊗p+1 , M) ⊗ (N ⊗H C ⊗q+1 ),
→ → →
with horizontal structure ∂ i = Id ⊗∂i , σ j = Id ⊗σj , and τ = Id ⊗τ and vertical
→ → →
p,q
structure ↑∂i = ∂i ⊗ Id, ↑σj = σj ⊗ Id, and ↑τ = τ ⊗ Id. Obviously (Ca−c , ∂, σ, τ ,↑
∂, ↑σ, ↑τ ) defines a bicocyclic module.
Now let us define the map
(3.10)
∗,∗
Ψ : Dq (Ca−c ) → Hom(B ⊗q+1 , C),
Ψ(φ ⊗ (n ⊗ c0 ⊗ · · · ⊗ cq ))(f 0 ⊗ · · · ⊗ f q ) = n | φ(f 0 (c0 ) ⊗ · · · ⊗ f q (cq )).
∗,∗ ∗,∗
Here D(Ca−c ) denotes the diagonal of the bicocyclic module Ca−c . It is a cocyclic
q,q
module whose qth component is C and its cocyclic structure morphisms are
→ → →
∂i := ∂ i ◦ ↑∂i , σj := σ j ◦ ↑σj , and τ := τ ◦ ↑τ .
Proposition 3.2. The map Ψ is a well-defined map of cyclic modules.
Proof. First let us show that Ψ is well-defined. Indeed, by using the facts
that M and N are compatible, f i are H-linear, φ is equivariant and (3.2) holds,
we see that,
Ψ(φ ⊗ (n ⊗ h(1) c0 ⊗ · · · ⊗ h(n+1) cn ))(f 0 ⊗ · · · ⊗ f n )

= n | φ(f 0 (h(1) · c0 ) ⊗ · · · ⊗ f n (h(n+1) · cn )
= n | h · φ(f 0 (c0 ) ⊗ · · · ⊗ f n (cn ))
= nh | φ(f 0 (c0 ) ⊗ · · · ⊗ f n (cn ))
= Ψ(φ ⊗ (n · h ⊗ c0 ⊗ · · · ⊗ cn ))(f 0 ⊗ · · · ⊗ f n ).
Next, we show that Ψ commutes with cocyclic structure morphisms. To this end,
we need only to show the commutativity of Ψ with zeroth cofaces, the last code-
generacies and the cyclic operators because these operators generate all cocyclic
structure morphisms. We check it only for the cyclic operators and leave the rest
to the reader. Let τB denote the cyclic operator of the ordinary cocyclic module of
CUP PRODUCTS IN CUP
WITH 277
7
the algebra B.
Ψ(τ (ϕ) ⊗ τ (n ⊗ c0 ⊗ · · · ⊗ cq ))(f 0 ⊗ · · · ⊗ f q )

= Ψ(τ ϕ ⊗ (n<0> ⊗ c1 ⊗ · · · ⊗ cq ⊗ n<−1> c0 ))(f 0 ⊗ · · · ⊗ f q )
= n<0> | τ (ϕ)(f 0 (c1 ) ⊗ · · · ⊗ f q−1 (cn ) ⊗ f q (n<−1> c0 ))

= n<0> | α ϕ S −1 (−)·f q (n<−1> c0 ) ⊗ f 0 (c1 ) ⊗ · · · ⊗ f q−1 (cn )

= n<0> | ϕ S −1 (n<−1> )·f q (n<−2> c0 ) ⊗ f 0 (c1 ) ⊗ · · · ⊗ f q−1 (cn )

= n | ϕ f q (c0 ) ⊗ f 0 (c1 ) ⊗ · · · ⊗ f q−1 (cq )
= τB Ψ(ϕ ⊗ (n ⊗ c0 ⊗ · · · ⊗ cq ))(f 0 ⊗ · · · ⊗ f q ).
Here in the passage from fourth line to the fifth one we use (3.7).

Let C := p,q≥0 C p,q be a bicocyclic module. With T ot(C) designating the total

mixed complex T ot(C)n = p+q=n C
p,q
, we denote by Tot(C) the associated
normalized subcomplex, obtained by retaining only the elements annihilated by
all degeneracy operators. Its total boundary is bT + BT , with bT and BT defined
as follows:
→
p+1
→
q+1
bp = (−1) i
∂ i, ↑bq = (−1)i ↑∂i ,
i=0 i=0
→
(3.11) bT = b p+ ↑bq ,
p+q=n
→
p−1
→i → →
q−1
→
Bp = ( (−1)(p−1)i τ ) σ p−1 τ , ↑Bq = ( (−1)(q−1)i ↑τ i ) σ q−1 ↑τ,
i=0 i=0
→
(3.12) BT = B p + ↑Bq .
p+q=n
The total complex of a bicocyclic module C is a mixed complex, i.e, b2T = BT2 =
bT BT +BT bT = 0. As a result its cyclic cohomology is well-defined. By means of the
analogue of the Eilenberg-Zilber theorem for bi-paracyclic modules [9, 17], the diag-
onal mixed complex (D(C), bD , BD ) and the total mixed complex (Tot C, bT , BT )
can be seen tobe quasi-isomorphic in both Hochschild and cyclic cohomology.
Here D(C) := q≥0 C q,q is a cocyclic module and therefore a mixed complex with
(co)boundaries,

q+1
→
bD := (−1)q ↑∂i ◦ ∂ i ,
i=0
(3.13) q−1

(q−1)i →i → →
BD := (−1) τ ↑τ i
σ q−1 ↑σq−1 τ ↑τ.
i=0
278
8 BAHRAM RANGIPOUR
At the level of Hochschild cohomology

the quasi-isomorphism is implemented by
the Alexander-Whitney map AW := p+q=n AWp,q : Tot(C)n → D(C)n ,
AWp,q : C p,q −→ C p+q,p+q
→ → →
(3.14) AWp,q = (−1)p+q ↑∂0 ↑∂0 . . . ↑∂0 ∂ n ∂ n−1 . . . ∂ q+1 .

q times
Using a standard homotopy operator H, this can be supplemented by a cyclic

Alexander-Whitney map AW := AW ◦ B ◦ H : Dn → Tot(C)n+2 , and thus up-
graded to an S-map AW = (AW, AW ), of mixed complexes. The inverse quasi-
isomorphisms are provided by the shuffle maps Sh := D(C)n → Tot(C)n , resp.
Sh = (Sh, Sh ), which are discussed in detail in [9, 17].
Let c be a (b, B) cocycle in Tot(C)n , n = p + q. Hence the class of AW (c) in
HC n (D(C)) is well defined.
Now we consider the inclusion ι : A → B = HomH (C, A), defined by ι(a)(c) = c · a.
We see that ι(ab)(c) = c · (ab) = (c(1) · a)(c(2) · b) = (ι(a) ∗ ι(b))(c), and ι(1A )(c) =
c · 1A = ε(c)1A = 1B (c). Hence ι is an algebra map and in turn induces a map at
the level of cyclic cohomology groups:
ι : HC ∗ (B) → HC ∗ (A).
Theorem 3.3. Let H be a Hopf algebra, A be an H-module algebra, C be an H-
module coalgebra acting on A, and (N, M) be a compatible pair of an SAYD module
and contramodule over H. Then Ψ̂ := ι ◦ Ψ ◦ AW defines a cup product at the level
of cyclic cohomology groups:
p q
(3.15) Ψ̂ := ι ◦ Ψ ◦ AW : HCH (A, M) ⊗ HCH (C, N ) → HC p+q (A).
p q
Proof. Let [φ] ∈ HCH (A, M) and [ω] ∈ HCH (C, N ). Without loss of gener-
ality one assumes that φ and ω are both cyclic cocycles, i.e.
→ →
b (φ) =↑b(ω) = 0, τ (φ) = (−1)p φ, ↑τ (ω) = (−1)q ω.
∗,∗ p+q
This implies that φ ⊗ ω is a (b, B) cocycle in Tot(Ca−c ) . Hence AW (φ ⊗ ω)
∗,∗
defines a class in HC p+q (D(Ca−c )). Finally, since ι and Ψ are both cyclic maps,
the transferred cochain ι ◦ Ψ(AW (φ ⊗ ω)) defines a class in HC p+q (A).
3.2. Module algebras paired with comodule algebras. Let H be a Hopf
algebra, A a left H-module algebra, B a left H-comodule algebra, and (N, M) be
a compatible pair of SAYD module and contramodule over H. One constructs a
crossed product algebra whose underlying vector space is A ⊗ B with 1 > 1 as its
unit and the following multiplication:
(3.16) (a > b)(a > b ) = a (b<−1> · a ) > b<0> b
Now consider the two cocyclic modules
∗
(3.17) (CH (A, M), ∂i , σj , τ ), and (H C ∗ (B, N ), ∂i , σj , τ )
introduced in [1] and [12] respectively while are recalled in (2.14) . . . (2.17) and
(2.19). . . (2.23). We define a bicocyclic module by tensoring these cocyclic modules
p,q
over C. The (p, q)-bidegree component Ca−a of this new bicocyclic module is given
by
(3.18) H
Hom(B ⊗q+1 , N ) ⊗ HomH (A⊗p+1 , M),
CUP PRODUCTS IN CUP
WITH 279
9
→ → →
with horizontal structure morphisms ∂ i = Id ⊗∂i , σ j = Id ⊗σj , and τ = Id ⊗τ
and vertical structure morphisms ↑∂i = ∂i ⊗ Id, ↑σj = σj ⊗ Id, and ↑τ = τ ⊗ Id.
Now we define a new morphism
∗,∗ n
(3.19) Φ : D(Ca−a ) → C n (A > B),
defined by
(3.20) Φ(ψ ⊗ φ)(a0 > b0 ⊗ · · · ⊗ an > bn ) =
= ψ(b0<0> ⊗ · · · ⊗ bn<0> ) | φ(S −1 (b0<−1> · · · bn<−1> ) · a0 ⊗ · · ·
(3.21) . . . ⊗ S −1 (bn<−n−1> ) · an ).
∗,∗
Proposition 3.4. The map Φ defines a cyclic map between the diagonal of Ca−a
∗
and the cocyclic module C (A > B).
Proof. We show that Φ commutes with the cyclic structure morphisms. We
shall check it for the first face operator and the cyclic operator and leave the rest to
the reader. Let us denote the cyclic structure morphisms of the algebra A > B by
∂iA>B , σjA>B and τ A>B . First we show that Φ commutes with the zeroth cofaces.
→
Φ( ∂ 0 ↑∂0 (ψ ⊗ φ))(a0 > b0 ⊗ · · · ⊗ an+1 > bn+1 )
= Φ(∂0 φ ⊗ ∂0 ψ))(a0 > b0 ⊗ · · · ⊗ an+1 > bn+1 )
= ∂0 ψ(b0<0> ⊗ · · · ⊗ bn+1<0> ) | ∂0 φ(S −1 (b0<−1> · · · bn+1<−1> ) · a0 ⊗ · · ·
· · · ⊗ S −1 (bn+1<−n−2> ) · an+1 )
= ψ(b0<0> b1<0> ⊗ · · · ⊗ bn+1<0> )
| φ(S −1 (b0<−1> · · · bn+1<−1> ) · a0 S −1 (b1<−2> · · · bn+1<−2> ) · a1 ⊗ · · ·
· · · ⊗ S −1 (bn+1<−n−2> ) · an+1 )
= ψ(b0<0> b1<0> ⊗ · · · ⊗ bn+1<0> )
| φ(S −1 (b0<−1> b1<−1> · · · bn+1<−1> ) · (a0 b0<−2> · a1 ) ⊗ · · ·
· · · ⊗ S −1 (bn+1<−n−1> ) · an+1 )
= Φ(ψ ⊗ φ)(a0 b0<−1> a1 > b0<0> b1 ⊗ a2 > b2 ⊗ · · · ⊗ an+1 > bn+1 )
= ∂0A>B Φ(ψ ⊗ φ)(a0 > b0 ⊗ · · · ⊗ an+1 > bn+1 ).
Now we show that Φ commutes with cyclic operators.
(3.22)
→
Φ( τ ↑τ (ψ ⊗ φ))(a0 > b0 ⊗ · · · ⊗ an > bn )
= Φ(τ ψ ⊗ τ φ)(a0 > b0 ⊗ · · · ⊗ an > bn )
= τ ψ(b0<0> ⊗ · · · ⊗ bn<0> ) | τ φ(S −1 (b0<−1> · · · bn<−1> )a0 ⊗ · · ·
· · · ⊗ S −1 (bn<−n−1> )an )
= ψ(bn<0> ⊗ b0<0> ⊗ · · ·
· · · ⊗ bn−1<0> ) · bn<−1> | α(φ(S −1 (−)S −1 (bn<−n−2> ) · an ⊗
S −1 (b0<−1> · · · bn−1<−1> bn<−2> ) · a0 ⊗ · · · ⊗ S −1 (bn−1<−n> bn<−n−1> ) · an−1 )).
280
10 BAHRAM RANGIPOUR
Using (3.7), and the fact that ψ is H-colinear, one has:

(3.23)
(3.22) = [ψ(bn<0> ⊗ b0<0> ⊗ · · · ⊗ bn−1<0> ) · bn<−1> ]<0> |
φ(S −1 ([ψ(bn<0> ⊗ b0<0> ⊗ · · · ⊗ bn−1<0> ) · bn<−1> ]<−1> )(S −1 (bn<−n−2> ) · an )⊗
⊗ S −1 (b0<1> · · · bn−1<−1> bn<−2> ) · a0 ⊗ · · · ⊗ S −1 (bn−1<−n+1> bn<−n−1> ) · an−1 ).
Using the fact that N is an AYD module we have,
(3.24)
(3.23) = ψ(bn<0> ⊗ b0<0> ⊗ · · · ⊗ bn−1<0> ) · bn<−3> |
φ((S −1 (S(bn<−2> )bn<−1> b0<−1> · · · bn−1<−1> )bn<−4> )(S −1 (bn<−n−2> ) · an )⊗
⊗ S −1 (b0<1> · · · bn−1<−1> bn<−5> ) · a0 ⊗ · · · ⊗ S −1 (bn−1<−n+1> bn<−n−4> ) · an−1 ).
Using (3.6) and the facts that φ is H-linear and M is an AYD contramodule we
see
(3.25)
(3.24) = ψ(bn<0> ⊗ b0<0> ⊗ · · · ⊗ bn−1<0> )bn<−1> |
φ(S −1 (b0<−1> · · · bn−1<−1> )bn<−2> S −1 (bn<−n−2> )an ⊗
⊗ S −1 (b0<1> · · · bn−1<−1> bn<−3> )a0 ⊗ · · · ⊗ S −1 (bn−1<−n+1> bn<−n−2> )an−1 )
= ψ(bn<0> ⊗ b0<0> ⊗ · · · ⊗ bn−1<0> ) | φ(S −1 (bn<−1> b0<−1> · · · bn−1<−1> )an ⊗
⊗ S −1 (b0<1> · · · bn−1<−1> )a0 ⊗ · · · ⊗ S −1 (bn−1<−n+1> )an−1 )
= Φ(φ ⊗ ψ)(an > bn ⊗ a0 > b0 ⊗ · · · ⊗ an−1 > bn−1 )
= τ A>B Φ(φ ⊗ ψ)(a0 > b0 ⊗ · · · ⊗ an > bn ).

Theorem 3.5. Let H be a Hopf algebra, A be an H-module algebra, B be an H-

comodule algebra, (N, M) be a compatible pair of SAYD module and contramodule.
Then the map Φ̂ := Φ ◦ AW defines a cup product:
p
(3.26) Φ̂ := Φ ◦ AW : H
HC q (B, N ) ⊗ HCH (A, M) → HC p+q (A > B).
Proof. The proof is similar to the proof of Theorem 3.3. Let [φ] ∈ H HC p (A, M)
q
and [ψ] ∈ HCH (B, N ). Without loss of generality one assumes that φ and ψ are
→ →
both cyclic cocycle, i.e, b (φ) =↑b(ψ) = 0, τ (φ) = (−1)p φ and ↑τ (ψ) = (−1)q ψ.
∗,∗ p+q
This implies that ψ ⊗ φ is a (b, B) cocycle in Tot(Ca−a ) . Hence AW (ψ ⊗ φ)
p+q ∗,∗
defines a class in HC (D(Ca−a )). Finally, since Φ is a cyclic map, the transferred
cochain Φ(AW (ψ ⊗ φ)) defines a class in HC p+q (A > B).
4. Cup products for incompatible pairs

In this section we generalize the cup products defined in (3.15) and (3.26) to the
case of incompatible coefficients. The target of the cup product in the new case is
the ordinary cyclic cohomology of algebras with coefficients in a module produced
out of the two incompatible coefficients.
CUP PRODUCTS IN CUP
WITH 281
11
Let M be a SAYD contramodule and let N be a SAYD module over a Hopf algebra
H. We define L(N, M) to be the coequalizer
/ / L(N, M),
(4.1) N ⊗H Hom(H, M) / N ⊗H M
where the equalized maps are n ⊗ m → n ⊗ α(f ), and n ⊗ m → n<0> ⊗ f (n<−1> ).
Remark 4.1. In fact L(N, M) is the usual (contra)tensor product defined by Posit-
selski [18, page 96]. One considers the H-coring C := H ⊗ H with the usual coring
structure and identifies N with a left H ⊗ H-comodule [2]. In the same fashion
one identifies M with a right C-contramodule. Then L(N, M) is identified with the
(contra)tensor N ⊗C M.
Now let A and C satisfy (3.1)...(3.4). We recall that the algebra B is HomH (C, A)
∗,∗
with the convolution multiplication and that Ca−c is the bicocyclic module defined
in (3.9). We define
∗,∗ q
(4.2) Ψ̃ : D(Ca−c ) → Hom(B ⊗q+1 , L(N, M)),
Ψ̃(φ ⊗ (n ⊗ c0 ⊗ · · · ⊗ cq ))(f 0 ⊗ · · · ⊗ f q ) = n ⊗H φ(f 0 (c0 ) ⊗ · · · ⊗ f q (cq )).
By a similar argument as in the proof of Proposition 3.2 one shows that Ψ̃ is a

cyclic map. One proves the following theorem with a similar proof as of Theorem
3.3
Theorem 4.2. Let H be a Hopf algebra, A be an H-module algebra, C be an H-
module coalgebra acting on A, and (N, M) be a not necessarily compatible pair of
SAYD module and contramodule over H. Then ι ◦ Ψ̃ ◦ AW defines a cup product
in the level of cyclic cohomology:
p q
ι ◦ Ψ̃ ◦ AW : HCH (A, M) ⊗ HCH (C, N ) → HC p+q (A, L(N, M)).
One notes that the range of this cup product is the ordinary cyclic cohomology of
the algebra A with coefficients in the vector space L(N, M). One also notes that
if (N, M) is compatible then E : L(N, M) → C defined by E(n, m) = n, m is a
map of vector spaces. As a result we get a cyclic map
Ẽ : Hom(A⊗∗ , L(N, M)) → Hom(A⊗∗ , C), Ẽ(ϕ) = E ◦ ϕ.
So we cover the old cup product as Ψ = Ψ̃ ◦ Ẽ, where Ψ is defined in (3.10).

Now let us generalize the other cup product for algebra-algebra in a similar fashion
as the case of algebra-coalgebra. Let A be a left H-module algebra, B be a left
H-comodule algebra, (N, M) be a pair of SAYD module and contramodule over
∗,∗
H, and A > B be the crossed product algebra defined in 3.16. Let also Ca−a be
the bicocyclic module defined in (3.18) . We define
(4.3)
∗,∗ q
Φ̃ : D(Ca−a ) → Hom((A > B)⊗q+1 , L(N, M)),
Φ̃(ψ ⊗ φ)(a0 > b0 ⊗ · · · ⊗ aq > bq )
= ψ(b0<0> ⊗ · · · ⊗ bq<0> ) ⊗H φ(S −1 (b0<−1> · · · bq<−1> )a0 ⊗ · · · ⊗ S −1 (bq<−q−1> )aq ).
Similarly we prove that Φ̃ is cyclic and induces a map on the level of cyclic coho-
mologies:
282
12 BAHRAM RANGIPOUR
Theorem 4.3. Let H be a Hopf algebra, A be an H-module algebra, B be an H-

comodule algebra, (N, M) be a compatible pair of SAYD module and contramodule.
Then the map Φ̃ ◦ AW defines a cup product in the level of cyclic cohomology:
q
Φ̃ ◦ AW : H
HC p (B, N ) ⊗ HCH (A, M) → HC p+q (A > B, L(N, M)).
References
[1] Brzeziński, T., Hopf-cyclic homology with contramodule coefficients. To appear in Quantum
Groups and Noncommutative Spaces, M Marcolli and D Parashar (eds) Vieweg Verlag (Max-
Planck Series), Preprint 2008, arxiv:0806.0389.
[2] Brzeziński, T., Flat connections and (co)modules, New Techniques in Hopf Algebras and
Graded Ring Theory, Universa Press, Wetteren, 2007 pp. 35-52. arxiv:math.QA/0608170.
[3] Connes, A., Noncommutative differential geometry. Inst. Hautes Etudes Sci. Publ. Math. No.
62 (1985), 257–360.
[4] Connes, A. and Moscovici, H., Background independent geometry and Hopf cyclic cohomology,
arXiv:math.QA/0505475.
[5] Connes, A. and Moscovici, H., Transgressions of the Godbillon-Vey class and Rademacher
functions,79–107, Aspects Math., E37, Vieweg, Wiesbaden, 2006.
[6] Connes, A. and Moscovici, H., Hopf algebras, cyclic cohomology and the transverse index
theorem, Commun. Math. Phys. 198 (1998), 199-246.
[7] Crainic, M., Cyclic cohomology of Hopf algebras. J. Pure Appl. Algebra 166 (2002), no. 1-2,
29–66.
[8] Eilenberg, S. and Moore, J.C., Foundations of relative homological algebra, Mem. Amer. Math.
Soc. 55 (1965).
[9] Getzler, E. and Jones, J. D. S., The cyclic homology of crossed product algebras, J. reine
angew. Math. 445 (1993), 163–174.
[10] Gorokhovsky, A., : Secondary characteristic classes and cyclic cohomology of Hopf algebras,
Topology 41(5), 993-1016 (2002)
[11] Hajac, P. M. , Khalkhali, M. , Rangipour, B. and Sommerhäuser Y., Stable anti-Yetter-
Drinfeld modules. C. R. Math. Acad. Sci. Paris 338 (2004), no. 8, 587–590.
[12] Hajac, P. M. , Khalkhali, M. , Rangipour, B. and Sommerhäuser Y., Hopf-cyclic homology
and cohomology with coefficients. C. R. Math. Acad. Sci. Paris 338 (2004), no. 9, 667–672.
[13] Jara, P. and Stefan, D., Hopf-cyclic homology and relative cyclic homology of Hopf-Galois
extensions. Proc. London Math. Soc. ( 3)93, 138–174 (2006).
[14] Kaygun, A., Products in Hopf-cyclic cohomology. Homology, Homotopy Appl. 10 (2008), no.
2, 115–133.
[15] Kaygun, A., Uniqueness of pairings in Hopf-cyclic cohomology, Journal of K-Theory, Vol. 6
(2010), No. 1, pp.1-21.
[16] Khalkhali, M. and Rangipour, B., Cup products in Hopf cyclic cohomology. C. R. Math.
Acad. Sci. Paris, 340(1):9-14, 2005.
[17] Khalkhali M. and Rangipour, B., On the generalized cyclic Eilenberg-Zilber theorem, Canad.
Math. Bull. 47 (2004), no. 1, 38–48.
[18] Positselski, L., Homological algebra of semimodules and semicontramodules. Semi-infinite
homological algebra of associative algebraic structures. arXiv:0708.3398.
[19] Rangipour, B., Cup products in Hopf cyclic cohomology via cyclic modules. Homology, Ho-
motopy Appl. 10 (2008), no. 2, 273–286.
Department of Mathematics and Statistics,University of New Brunswick, Fredericton,

NB, CANADA, E3B 5A3
E-mail address: bahram@unb.ca
Contemporary
Volume 546,Mathematics
2011
Algebras of p-symbols,
noncommutative p-residue,
and the Brauer group
Mariusz Wodzicki
Dedicated to Henri Moscovici on the occasion of his sixty-fifth birthday.
A BSTRACT. Importance of the pseudodifferential symbol calculus extends far be-

yond the fundamental role it is known to play in Global and Microlocal Analysis.
In this article, we demonstrate that algebras of symbols contribute to subtle phe-
nomena in characteristic p > 0.
A perfect fit between Smooth Geometry and de Rham Theory in characteristic

zero leads many to interpret the situation in characteristic p > 0 as an apparent
failure of de Rham Theory in positive characteristic. Smoothness, equated with
the existence of local coordinates, i.e., of an étale map from a neighborhood of an
arbitrary point to the affine space A n , is a concept independent of the ground ring.
What however is very much dependent on the ground ring k and its characteristic
is the geometry of the affine space itself which provides a local model for Smooth
Geometry after all.
Local calculations in Smooth Geometry rely on the fact that the affine spaces
are objects of the category of commutative unipotent algebraic groups. When the
ground ring is a field of characteristic zero, this category is equivalent to the cate-
gory of finite-dimensional vector spaces, all objects are semisimple, and the addi-
tive group G a , which corresponds to the one-dimensional affine space A1 , is the
sole simple object.
In contrast, over a ring of characteristic p > 0, the line is not even semisimple:
G a fits for example into the nontrivial extension of algebraic group schemes
F
(0.1) Ga u u Ga u x G a,1
where F : G a → G a , the Frobenius morphism, corresponds to the k-algebra endo-
morphism of O(G a ) = k[z] which sends z to z p .
1991 Mathematics Subject Classification. Primary 58J42, 16H05, 16K50; Secondary 14G17, 47L80,
58B34.
Key words and phrases. Pseudifferential symbols, Azumaya algebras, Brauer group, noncommuta-
tive residue.
The author was supported in part by NSF Grants DMS-0503401 and DMS-1001846.
1
283
2284 MARIUSZ WODZICKI
The kernel of the Frobenius morphism G a,1 , on the other hand, is a simple
object. It resembles, however, a circle rather than a line even though, formally
speaking, is neither smooth nor of dimension one.
Note the similarities: its module of Kähler differentials Ω1G /k is a free module
a,1
of rank 1 over OG a,1 , its zeroth and first de Rham cohomology groups are free
modules of rank 1 over k, and the Lie algebra of vector fields on G a,1 is the Witt
algebra

Wp ( k ) = k ei , [ ei , e j ] = ( j − i ) ei + j (i, j ∈ Z/pZ ),
i ∈Z/pZ
considered to be a counterpart in positive characteristic to the Lie algebra of vector

fields on a circle. In particular, Wp (k) possesses a universal central extension with
one-dimensional kernel for p > 3.
Note that the Witt algebra is a simple Lie algebra of rank 1 if p > 2, the only
other simple Lie algebra of rank 1 over an algebraically closed field in positive
characteristic being sl2 (k).
An important structure associated with circle is its algebra of pseudodiffer-

ential symbols CS(S1 ) together with the noncommutative residue trace. In present
article we will demonstrate that its counterpart in positive characteristic is equally
deserving of attention.
In Chapter 1 we introduce algebras of p-symbols, Sab (k), parametrized by pairs
of elements a and b of a commutative F p -algebra k. If k = k p , then all Sab (k ) are
isomorphic to the algebra of differential operators on the kernel of Frobenius G a,1 .
We show that each Sab (k) is equipped with a unique trace, a close relative of
the noncommutative residue in characteristic zero. Appropriately, we call it the
noncommutative p-residue.
In Chapter 2 we establish a number of useful tensor identities involving the
algebras of p-symbols and we use these identities in Chapter 3 to prove that each
Sab (k) is an Azumaya algebra. We achieve this by providing an explicit isomor-
phism between Sab (k)⊗ p and a certain algebra of differential operators, cf. (2.13)
below, which shows that the similarity class of Sab (k ) in the Brauer group Br(k ),
which is duly introduced in Chapter 3, is of order p when nontrivial.
In Chapter 3 we also characterize the algebras of p-symbols as being precisely
the central quotients of the Weyl algebra1
kz, ζ
(0.2) A1 ( k ) : = .
([ζ, z] − 1)
In the next chapter we show that the Weyl algebra itself is a nontrivial Azu-
maya algebra over its center,
Z ( A1 (k)) = k [z p , ζ p ] ,
by identifying A1 (k) with the algebra of p-symbols
Sz p ζ p (k [z p , ζ p ]) .
1The term Weyl algebra, first introduced by Dixmier [4] in 1968 (cf. [5], p. 46), may be yet another
example of wrong apellation: algebra of the Canonical Commutation Relations (CCR) was studied by
Dirac in 1926 [3], i.e., two years before CCR appear in Weyl’s book [14]. Apparently the first thorough
investigation of A1 (C ) was carried by Littlewood in 1930–1931 [8], but see also [6].
ALGEBRAS OF p-SYMBOLS, NONCOMMUTATIVE p-RESIDUE 285 3
As a corollary, we obtain a recent theorem of Bezrukavnikov, Mirković and

Rumynin [2] stating that the algebra of PD-differential operators on a smooth
scheme X/S, introduced by Berthelot [1] and studied by himself, Ogus, Vologod-
sky, among others, is an Azumaya algebra over the algebra of functions on the
cotangent space of the Frobenius twist X ( p) /S. Our proof is totally explicit and
elementary.2
An adelic noncommutative residue trace on the Weyl algebra over an arbitrary
ring of coefficients k is presented in Chapter 5.
In Chapter 6 we collect a number of identities for powers of certain elements
in general associative rings and in F p -algebras. With help of these identities we
establish a sufficient condition for triviality of the class of Sab in Br(k). These iden-
tities are also used in an essential way in the final two chapters.
Tensor identities of Chapter 2 are special cases of general identities associated
with certain actions of the symplectic groups Spn (k), n ≥ 2: these are the subject
of Chapter 7. The orbits of the aforementioned actions correspond to elements in
(0.3) p Br( k ) := { β ∈ Br(k) | pβ = 0}
whereas the elements of Spn (k ) provide nontrivial relations in p Br(k).
In the final chapter we represent cyclic p-algebras, which are defined as crossed
products k b Z/pZ of Artin-Schreier extensions k /k and Gal(k /k ) = Z/pZ, as
algebras of p-symbols. By combining this with a classical result of Teichmüller we
deduce that any element of order p in the Brauer group of a field of characteristic
p is represented by a suitable algebra of p-symbols.
Originally we encountered the noncommutative p-residue and the algebras of

pseudodifferential p-symbols in our study of the structure of differential operators
on the algebra of divided-power polynomials, Γk [ x ], as documented in [15].
We would like to conclude this introduction by saying that the noncommuta-
tive residue, nontrivial extension (0.1), and the Cartier operations, are so intimately
connected—they can be thought of as being manifestations of a single phenome-
non.
1. The algebras of p-symbols Sab (k )

Let a and b be a pair of elements of a unital commutative ring k of prime
characteristic p > 0. The latter means that pk = 0 or, equivalently, that k is an
F p -algebra.
We shall denote by Sab (k ) the quotient of the free k-algebra

kz, ζ = Tk Wzζ ,
generated by the free k-module of rank 2 with basis {z, ζ },
Wzζ = kz ⊕ kζ,
by the ideal Iab = Iab (k ) generated by the following three relations
(1.1) [ζ, z] = 1, z p = a, and ζ p = b,
2In [10], Théorème 2, Philippe Revoy proves that A (k ) is a central and separable algebra over
1
k [z p , ζ p ]. This is equivalent to A1 (k) being an Azumaya k [z p , ζ p ]-algebra (cf. [7], Théorème 5.1).
Revoy’s article escaped our notice until the present work has been completed. There is no reference to
Revoy in [2].
and call it the algebra of p-symbols defined by the pair of elements a and b of ground
ring k. We shall omit k from notation when the ground ring is clear from the
context.
The composition law. As a k-module, Sab (k) is free of rank p2 with the mono-
mial basis {zl ζ m }0≤l,m< p where we identify zl ζ m , for 0 ≤ l, m < p, with their
images in Sab (k).
If we identify Sab (k) as a k-module with the commutative k-algebra
Oab := Oa ⊗ Ob ,
where
(1.2) Oc : = k [ t ] / ( t p − c ) ( c ∈ k ),
by sending z to t ⊗ 1 and ζ to 1 ⊗ t, then multiplication in Sab is given by the
familiar law for composition of pseudofifferential symbols.
More precisely, for polynomial symbols α, β ∈ k[z, ζ ], where k denotes an arbi-
trary commutative ring of coefficients, their composition is given by the formula
∞
∑ ∂ζ α ∂z
j [ j]
(1.3) α◦β = β.
j =0
[ j]
Here ∂z denotes the j-th divided-power of ∂:
⎧
⎨ l zl − j if l ≥ j
∂[ j] ( zl ) = j
⎩
0 otherwise
which is a differential operator of order j on k [z]. If j! is invertible in k, then
1 j
∂[ j] = ∂.
j!
j
Since we are assuming pk = 0 the operator ∂ζ is identically zero for j ≥ p.
Thus, the composition law for polynomial symbols in characteristic p is in fact
given by the finite expression
p −1
1 j j
(1.4) α◦β = ∑ ∂ α ∂z β .
j! ζ
j =0
Note that the ideal − c) ⊂ k[t] defining Oc is ∂-invariant, hence the right-
(t p
hand side of (1.4) is well defined for α, β ∈ Oab , and (1.4) is precisely the formula
for multiplication in Sab .
We shall henceforth refer to elements of Sab , represented as elements of Oab
but multiplied according to (1.4), as p-symbols.
Noncommutative p-residue. In view of the remark made in the previous para-
graph, the standard Poisson bracket on the algebra of polynomials k [z, ζ ],
{ f , g} = ∂ζ f ∂z g − ∂z f ∂ζ g,
passes to the quotient algebra Oab thus making it a Poisson algebra. Similarly, the
associated symplectic form on A2 ,
ω = dζ ∧ dz,
ALGEBRAS OF p-SYMBOLS, NONCOMMUTATIVE p-RESIDUE 2875
passes to a differential 2-form on the quotient algebra.

The algebra of differential forms on Oab is a free graded-commutative algebra
over Oab generated by the free Oab -module with basis {dζ, dz} of degree 1,
∗ ∗ ∗
(1.5) ΩO ab /k
ΩO a /k
⊗ k ΩO b /k
Oab ⊗k Λ∗k (dζ, dz).
P ROPOSITION 1.1. The correspondence

2
(1.6) τ : α → [αω ] ∈ HdR (Oab )
which sends a symbol α to the cohomology class of the 2-form αω, is a trace on the algebra
of symbols Sab .
This trace is unique, in the sense that (1.6) induces an isomorphism
Sab 2
(1.7) HdR (Oab )
[Sab , Sab ]
2 (O ) is a free k-module of rank 1 generated by the class of the 2-form
and HdR ab
(1.8) z p−1 ζ p−1 ω.

P ROOF. The commutator formula

(1.9) [α, β] ω = d ρ(α, β)dz + σ(α, β)dζ ,
where
p −1 j −1
1 j − i −1
∑ ∑ ∂ζ
j
(1.10) ρ(α, β) := α (−∂ζ )i ∂z β
j =1
j! i =0
and
p −1 j −1
1 j − i −1
∑ ∑ ∂z
j
(1.11) σ(α, β) := α ∂ζ (−∂z )i β,
j =1
j! i =0
2 (O ).
shows that (1.6) is a k-linear trace on Sab with values in HdR ab
∗
The de Rham cohomology algebra HdR (Oab ) is free graded-commutative, and
generated by the classes of the differential 1-forms
z p−1 dz and ζ p−1 dζ.

2 (O ) is a free k-module generated by the class of (1.8), and thus
In particular, HdR ab
map (1.6) is surjective.
The kernel of (1.6) is a free k-module of rank p2 − 1 with the monomial basis
(1.12) {zl ζ m }0≤l,m≤ p−1; l +m≤2p−3 .

Each basic monomial in (1.12) is a single commutator:

1
[ζ, zl +1 ζ m ] if l p − 1
(1.13) z ζ = l +1
l m
1
− m+1 [z, zl ζ m+1 ] if m p − 1
which demonstrates that the kernel of (1.6) coincides with [Sab , Sab ] and thus cor-
respondence (1.6) induces a k-module isomorphism (1.7).
2 (O ) with k, we can also describe τ as the k-linear func-

By identifying HdR ab
tional τ : Oab → k which sends
α= ∑ clm zl ζ m ∈ Oab , (clm ∈ k),
0≤l,m≤ p−1
to
(1.14) τ (α) := c p−1,p−1 .
When both a and b are invertible in k, then τ (α) is the classical double Cauchy
Residue in z and ζ variables:
1
× the coefficient of α at z−1 ζ −1 .
ab
We shall be refering to τ as the noncommuattive p-residue.
2. Symplectic isomorphisms
Below we establish a number of special k-algebra isomorphisms
(2.1) Sa1 b1 ⊗ · · · ⊗ San bn Sa b ⊗ · · · ⊗ San bn ( n ≥ 1).
1 1
The left-hand-side of (2.1) is a quotient of the tensor algebra

(2.2) Tk (Wz1 ζ 1 ⊕ · · · ⊕ Wzn ζ n )
and, similarly, the right-hand-side is a quotient of the tensor algebra
(2.3) Tk (Wz ζ ⊕ · · · ⊕ Wzn ζ n ).
1 1
If the k-algebra isomorphism, (2.1), is induced by an isomorphism of k-modules,

(2.4) Wz1 ζ 1 ⊕ · · · ⊕ Wzn ζ n Wz ζ ⊕ · · · ⊕ Wzn ζ n ,
1 1
the latter preserves the symplectic form

⊕n
0 1
−1 0
that both sides of (2.4) are equipped with, and we propose to call (2.1) a symplectic
isomorphism.
We shall signal symplectic isomorphisms by employing notation
instead of usual .
Elementary observations. In the following proposition we collect prelimi-
nary observations about algebras Sab .
P ROPOSITION 2.1. (a) For any a, b ∈ k, one has
(2.5) Sab S− a,−b Sb,− a .
(b) The opposite algebra,(Sab )op , is canonically isomorphic to Sba ,
(2.6) (Sab )op Sba .
(c) Algebra Sa0 is canonically isomorphic to the algebra of differential operators, Dk (Oa ).
The latter coincides with the algebra of all k-module endomorphisms,
(2.7) Dk (Oa ) = Endk-mod (Oa ),
and thus is isomorphic to the algebra of p × p matrices M p (k) with coefficients in k.

Under this isomorphism, the noncommutative p-residue trace corresponds to the matrix
trace with the reverse sign:
τ = − Tr .
(d) If b ∈ k ⊂ k, then
p
(2.8) Sab Sa0 .

If b = c p , then the isomorphism in (2.8),
kz, ζ k z, ζ
,
(z p = a, ζ p = b, [ζ, z] = 1) (z p = a, ζ p = 0, [ζ , z] = 1)
is induced by the substitution
ζ → ζ + c.
Tensor identities.
P ROPOSITION 2.2. One has the following canonical symplectic isomorphisms:
(2.9) Sab ⊗ Sa b Sa,b− a ⊗ Sa ,b − a
(2.10) Sab ⊗ Sbc Sa0 ⊗ Sb,c− a
Sab ⊗ Sab Sab ⊗ Sba Sa0 ⊗ Sb0

op
(2.11)
Dk (O a ) ⊗ Dk (Ob ) M p ( k ) ⊗2
⊗(l −1)
S⊗ l
ab Sa0 ⊗ Sb,−la Dk (Oa )⊗(l −1) ⊗ Sb,−la
(2.12)
M p (k)⊗(l −1) ⊗ Sb,−la
⊗p ⊗( p−1)
(2.13) Sab Sa0 ⊗ Sb0 Dk (Oa )⊗( p−1) ⊗ Dk (Ob ) M p (k)⊗ p .
P ROOF. The k-module map
(2.14) ϕ : Wzζ ⊕ Wz ζ → Wzθ ⊕ Wz θ
which sends z and z to themselves, and
(2.15) ζ → θ + z , ζ → θ + z,
induces k-algebra homomorphisms
(2.16) ϕ aba b : Sab ⊗ Sa b → Sa,b− a ⊗ Sa ,b − a ,
while the map inverse to (2.14),
ψ : Wzθ ⊕ Wz θ → Wzζ ⊕ Wz ζ ,
which sends
(2.17) θ → ζ − z , θ → ζ − z,
induces the inverse k-algebra homomorphisms
(2.18) ψa,b− a ,a ,b − a : Sa,b− a ⊗ Sa ,b − a → Sab ⊗ Sa b .
Indeed, if
z p = a, ζ p = b, z p = a , ζ p = b
and
[ζ, z] = 1 = [ζ , z ],
then
(2.19) (ζ − z ) p = b − a , (ζ − z) p = b − a,
(2.20) [ζ − z , z] = 1 = [ζ − z, z ]
and
(2.21) 0 = [ζ − z , ζ − z] = [ζ − z , z ] = [ζ − z, z].
This establishes the existence of a canonical symplectic isomorphism in (2.9).
Isomorphism (2.10) is a special case of (2.9), and (2.11) is a special case of (2.10)
if one takes into account parts (b) and (c) of Proposition 2.1.
Isomorphism (2.12) is proven by induction on l by using (2.9) again:
⊗(l +1) ⊗(l −1)
Sab Sab ⊗ S⊗
ab Sab ⊗ Sa0
l
⊗ Sb,−la
⊗(l −1)
Sa0 ⊗ Sab ⊗ Sb,−la S⊗
a0 ⊗ Sb,−(l +1) a .
l
Finally, isomorphism (2.13) is a special case of (2.12) combined with part (c) of
Proposition 2.1.
R EMARK 2.3. If algebras Sab are thought of as “1-dimensional,” then the tensor
products
(2.22) Sa1 ,...,an ; b1 ,...,bn = Sa1 b1 ⊗ · · · ⊗ San bn
should be considered “n-dimensional” algebras of p-symbols.
R EMARK 2.4. Tensor identities (2.9)–(2.13) are special cases of a general iden-
tity established in Section 6, cf. Theorem 7.1.
3. The Brauer group Br(k ).

Azumaya algebras. Let us recall that a unital algebra A is said to be an Azu-
maya algebra over k if there exist: a unital k-algebra B and a faithfully projective
k-module P such that
A ⊗ Bop Endk-mod ( P).
Since A ⊗ Bop possesses an identity, k-module P must be finitely generated.
In this case we say that algebras A and B are similar, and denote this fact by
A ∼ B.
Similarity is an equivalence relation on the class of Azumaya algebras over a
given ground ring k, and the set of similarity classes of such algebras, equipped
with the multiplication induced by tensor product, forms a group, denoted Br(k ),
which is called the Brauer group of ring k. The inverse of [ A] in Br(k ) is the similar-
ity class of the opposite algebra, [ Aop ].
Several characterisations of Azumaya algebras are provided in Chapter III,
Section 5, of [7] (cf. Théorème 5.1 ibid.)
The following is an immediate corollary of the existence of symplectic isomor-
phisms (2.11) and (2.13) in Proposition 2.2.
C OROLLARY 3.1. For any a and b in k, the algebra of symbols, Sab (k), is an Azumaya
k-algebra and defines an element in
p Br( k ) = { β ∈ Br(k) | pβ = 0}.
R EMARK 3.2. For any n ≥ 1, the correspondence

(3.1) ( a1 , . . . , an ; b1 , . . . , bn ) → Sa1 ,...,an ; b1 ,...,bn
produces a map
k2n → p Br(k),
whose fibers are invariant under the action of the group Sp2n (k) of symplectic
matrices with coefficients in k. We discuss this in detail in Section 7.
In the final Section we prove that the cumulative map

k2n → p Br(k)
n ≥1
is surjective when k is a field, cf. Theorem 8.3.

A characterisation of the algebras of p-symbols. We shall say that a k-algebra
is a central quotient of a k-algebra B, if it is of the form A = B/J for a certain
twosided ideal J ⊆ B and the structural homomorphism k → A identifies k with
the center of A.
P ROPOSITION 3.3. Every k-algebra Sab is a central quotient of the Weyl k-algebra
A1 (k ), cf. (0.2), and vice-versa: every central quotient of A1 (k) is of the form Sab for a
suitable pair of a, b ∈ k.
P ROOF. If we consider Sab as a free k[z]-module of rank p,

Sab = k[z]ζ m ,
0≤ m ≤ p −1
then the inner derivation adz = [z, ] identifies k [z]ζ m with k[z]ζ m−1 , for m > 0,
and annihilates k[z]ζ 0 . In particular, ker adz = k[z]. Similarly, ker adζ = k[ζ ]. It
follows that
k ⊆ Z (Sab ) ⊆ ker adz ∩ ker adζ = k
where Z (Sab ) denotes the center of Sab .
Assume now that a k-algebra A is a central quotient of A1 (k ). We shall identify
z and ζ with their images in A. The commutator identity
(adz ) p = adz p
combined with
[z, [z, ζ ]] = 0
shows that ∈ Z ( A) = k. Similarly for ζ p . Thus, A is a quotient of Sab for a = z p
zp
and b = ζ .
p
Above we demonstrated that Sab is an Azumaya k-algebra, cf. Corollary 3.1. It

remains to apply the following lemma.
L EMMA 3.4. If A = B/J is a central quotient of an Azumaya algebra, then J = 0
and A = B.
10
Indeed, any twosided ideal in an Azumaya k-algebra is of the form J = IB

for some ideal I ⊆ k (cf. [7], Chapter III, Cor. 5.2). The structural homomorphism
k → A is a monorphism in view of the hypothesis that A is a central quotient of B.
Since it factors through the quotient map k → k/I, the latter is injective and thus
I = 0.
4. The Weyl algebra in positive characteristic

The center of the (1-dimensional) Weyl algebra A1 (k ) = k z, ζ /([ζ, z] − 1)
with coefficients in an F p -algebra k contains
(4.1) K = k [ z p , ζ p ].
Viewed as an algebra over K, the Weyl algebra is nothing but the following K-
algebra of p-symbols
A 1 ( k ) = Sz p ζ p ( K ) .
In particular, Z ( A1 (k)) = K, and the Weyl algebra is an Azumaya over its center.
Aided by tensor identity (2.13) we obtain a very precise form of that last state-
ment.
P ROPOSITION 4.1. There exists a canonical isomorphism of K-algebras
A 1 ( k ) ⊗ K p DK ( O a ) ⊗ K ( p −1) ⊗ K DK ( Ob ) M p ( K ) ⊗ K p .
where a = z p , b = ζ p , and K is given by (4.1).

P ROPOSITION 4.2. The n-dimensional Weyl algebra, An (k ) A1 (k)⊗k n , is an
Azumaya algebra over its center

p p p p
K n = k z1 , . . . , z n ; ζ 1 , . . . , ζ n ,
and its similarity class in the Brauer group Br(Kn ) has exactly order p.
P ROOF. In view of Proposition 4.1, it remains only to prove that [ An (Kn )] 0
in Br(Kn ).
Let us consider the homomorphism

p p p p p p
K n = k z 1 , . . . , z n ; ζ 1 , . . . , ζ n → k z 1 , ζ 1 = K1
which sends z j and ζ j to zero for j > 1. The associated base-change functor sends
Kn -algebra An (k) to the K1 -algebra
K1 ⊗Kn An (k) A1 (k) ⊗k M p (k )⊗( p−1) ,
and [ An (k )] ∈ Br(Kn ) is sent to [ A1 (k)] ∈ Br(K1 ).
Let k̄ be the residue field of k at any maximal ideal. The base change functor
associated with the quotient homomorphism k → k̄ sends K1 -algebra A1 (k) to the
K̄1 -algebra A1 (k̄), where
p p
K̄1 = k̄ z1 , ζ 1 .
The latter is a domain. Let F be the field of fractions of K̄1 . The base change functor
associated with the inclusion K̄1 → F sends K̄1 -algebra A1 (k̄ ) to
(4.2) F ⊗K̄1 A1 (k̄) Sz p ζ p ( F ).
1 1
293
The right hand side of (4.2) is an Azumaya F-algebra of dimension p2 over F.

Thus, it is either a central division F-algebra or is isomorphic to M p ( F ). It also
contains A1 (k̄ ), and the latter satisfies the left and the right Ore conditions. This
was first noted in print perhaps by Dudley Ernest Littlewood3 ([8], Thm. XIX,
pp. 219-220; Littlewood considers there only the case of real or complex numbers
4
but his proof of Thm.
XIX applies to any field of coefficients).
Algebra A1 k̄ is a domain. Thus, the ring of fractions

Frac A1 k̄ = { DE−1 | D, E ∈ A1 k̄ , E 0}
is a division ring. Since it contains Sz p ζ p ( F ), the latter cannot be isomorphic to a
1 1
matrix algebra.
This proves that the class of Sz p ζ p ( F ) in Br( F ) is not zero and as a consequence
1 1
also the class of An (Kn ) in Br(Kn ). It also demonstrates that Sz p ζ p ( F ), being a
1 1
division algebra itself, must coincide with the total algebra of fractions of A1 k̄ ,

(4.3) Sz p ζ p ( F ) = Frac A1 k̄ .
1 1
Equality in (4.3) is equivalent to the following property of Weyl algebra A = A1 (k ):

if k is a field of positive characteristic, then for any α ∈ A, there exists
α ∈ A such that αα is a nonzero element of the center of A.
Wedderburn in [13] calls these algebras Hamiltonian since the algebra of quater-
nions at that time was the best known example of such algebras.
R EMARK 4.3. Proposition 4.2 implies that the algebra of the so called PD-
differential operators, introduced by Berthelot [1], is an Azumaya algebra over its
center. This fact seems to have been first noted in print in [2] where it was also
proved (Theorem 2.2.3 ibidem).
R EMARK 4.4. In Section 6 of the
present article we establish a sufficient condi-
tion for the triviality of class Sab in Br(k ) and, when k is a field, we prove it to be
also necessary, cf. Corollary 6.10 and Proposition 6.11 below.
5. A trace on the Weyl algebra

Let k be an arbitrary comutative ring with identity. For any prime p, the com-
position of the reduction modulo p map
A1 ( k ) F p ⊗Z A1 ( k )
with the trace map introduced in Section 1,

F p ⊗Z A1 ( k ) Sz p ζ p F p ⊗Z k [ z p , ζ p ] F p ⊗Z k [ z p , ζ p ] = : F p ⊗ Z k z p , ζ p ,
3Dudley Ernest Littlewood (1903–1979), not to be confused with Hardy’s friend and collaborator,
John Edensor Littlewood (1885–1977).
4In the same year 1933 appeared article [9] in which Öystein Ore introduced and thoroughly
investigated a very general type rings of polynomials of one variable with multiplication twisted by a
certain endomorphism α and a derivation δ acting on the “coefficients”; in particular, Ore proved for
such rings the noncommutative versions of the Euclid Division Algorithm, from which he derived that
such rings of twisted polynomials satisfy the conditions that today bear his name—if and only if α is
an automorphism. Ore’s article was submitted in December 1932, Littlewood’s—in June 1931.
12
cf. (1.14), defines a trace on A1 (k):

(5.1) res p : A1 (k) F p ⊗Z k z p , ζ p .
One has
⎧
⎨ l +1 − 1 m +1 −1
if l = m = −1
p p
l m zp ζp mod p
(5.2) res p z ζ = .
⎩0 otherwise
Note that
res p zl ζ m 0
only for primes dividing the greatest common divisor of l + 1 and m + 1. It follows
that the k-linear map

(5.3) res : A1 (k) → F p ⊗Z k z p , ζ p , res(α) := ∑ res p (α),
p p
where summation extends over all primes, is well defined and annihilates the com-
mutator k-module [ A1 (k), A1 (k)].
Map (5.3) is surjective. Indeed, for i, j ∈ N and a prime p, let π be the product
of all primes different from p which divide the greatest common divisor of i + 1
and j + 1,
π := ∏ q.
q|gcd(i +1,j+1)
qp
If π ∈ Z satisfies
ππ = 1 mod p,
then, for any prime q,

j
zip ζ p if q = p
( i +1) p −1 ( j +1) p −1
resq ππ z ζ = .
0 otherwise
By taking the n-th tensor power of (5.3) we obtain the corresponding trace on
the n-dimensional Weyl algebra
⊗ n
res⊗k n : An (k ) = A1 (k)⊗k n F p ⊗Z k z p , ζ p k .
p
6. Power identities
Two power-of-the-product identities. Let R be a unital ring. Below we adopt
the convention that x0 = 1 for any x ∈ R.
P ROPOSITION 6.1. Let r and s be a pair of elements of R satisfying
(6.1) [[r, s], r ] = 0 = [[r, s]s].
Then
n
(6.2) (rs)n = ∑ anl [r, s]n−l sl rl ( n ≥ 0)
i =0
and
n
(6.3) (rs)n = ∑ bnl [s, r]n−l rl sl ( n ≥ 1)
i =1
295
where
(6.4) anl = φ(nl− l
+1)
(1, 2, . . . , l + 1) ( n ≥ 0)
and
(n−l )
(6.5) bnl = φl (1, 2, . . . , l ) ( n ≥ 1)
(d)
Here φm ∈ Z [ X1 , . . . , Xm ] is the symmetric form of degree d in m variables
∑
(d) d
(6.6) φm ( X1 , . . . , Xm ) = X1 1 · · · X m
dm
.
d,...,dm ≥0
d1 +···+dm =d
P ROOF. Formulae (6.2) and (6.4) are obviously valid for n = 0. By multiplying
both sides of (6.2) on the left by rs, we obtain the following expression for (rs)n+1 ,
n
(rs)n+1 = ∑ anl [r, s]n−l (rs)sl rl
l =0
n n l +1
(6.7) = ∑ anl [r, s]n−l sl+1 rl+1 + ∑ anl [r, s]n−l ∑ sl [r, s]sl −m r l
l =0 l =0 m =0
n +1 n
= ∑ an,l−1 [r, s]n+1−l sl rl + ∑ anl [r, s]n+1−l sl rl
l =1 l =0
which can be re-written as

n +1
(6.8) (rs)n+1 = ∑ (an,l−1 + (l + 1)anl )[r, s]n+1−l sl rl ,
l =0
if we adopt the convention

(6.9) anl = 0 for either l < 0 or l > n.
The latter is compatible with the fact that

1 if l = 0
(6.10) a0l = .
0 otherwise
Induction on n with help of (6.8) demonstrates that formula (6.2) holds for
certain integral coefficients anl satisfying the “boundary” conditions
(6.11) an0 = ann = 1
and the recurrence formula
(6.12) an+1,l = an,l −1 + (l + 1) anl (0 < l < n ).
Note that the coefficients
(n−l )
anl := φl +1 (1, . . . , l + 1)
obviously satisfy boundary conditions (6.11),
(n) (0)
φ1 (1) = 1n = 1, φn+1 (1, . . . , l + 1) = 10 · · · (l + 1)0 = 1,
14
while formula (6.12), for anl , is a consequence of the identity
((n+1)−l ))
(6.13) φl +1 ( X1 , . . . , X l + 1 ) =
(n−(l −1)) (n−l )
φl ( X1 , . . . , Xl ) + Xl +1 φl +1 ( X1 , . . . , Xl +1 )
which holds in Z [ X1 , . . . , Xl +1 ].
Induction on n shows that
anl = anl (0 ≤ l ≤ n ).
This yields equality (6.4).
Multiplication of both sides of (6.2) on the left by s and, on the right, by r,
produces equalities (6.3) and (6.5), respectively.
( p−l )
Arithmetic of the form φl . For a given prime p, let us consider the func-
( p−l )
tions F lp → F p associated with forms φl for 0 ≤ l ≤ p,
( p−l )
(6.14) ( x1 , . . . , xl ) → φl ( x1 , . . . , x l ).
P ROPOSITION 6.2. One has
( p−l )
(6.15) φl (ν1 , . . . , νl ) = 0,
for any 1 < l < p and any l-tuple (ν1 , . . . , νl ) ∈ (F ∗p )l such that
(6.16) νi νj (1 ≤ i j ≤ l ).
P ROOF. Note the identity
( n +1− l ) ( n +1− l )
(6.17) φl −1 ( X1 , . . . , X̂i , . . . , Xl ) − φl −1 ( X1 , . . . , X̂ j , . . . , Xl )
(n−l )
= ( Xi − X j )φl ( X1 , . . . , X l )
in Z [ X1 , . . . , Xl ].
( p−l )
It follows that when νi − νj ∈ F ∗p , then φl (ν1 , . . . , νl ) vanishes if and only if
( p +1− l ) ( p +1− l )
(6.18) φl −1 (ν1 , . . . , ν̂i , . . . , νl ) = φl −1 (ν1 , . . . , ν̂j , . . . , νl ).
Since
( p −1)
φ1 ( ν ) = ν p −1 = 1 (ν ∈ F ∗p ),
we observe that both sides of (6.18) are equal to 1 for l = 2 and any νi , νj ∈ F ∗p .
Induction on l in the range 2 ≤ l ≤ p − 1 proves that both sides of (6.18) are equal
and, indeed, for 3 ≤ l ≤ p − 1, both vanish, provided condition (6.16) is satisfied.

In the rest of this Section we assume that pR = 0, i.e., that R is an F p -algebra.

Under suitable hypotheses relevant to the study of algebras of p-symbols, we
present two formulae for the p-th power of the product and of the sum of two
elements in R.
297
A pth -power-of-the-product identity for an F p -algebra. We begin from the

formula for the p-th power of the product.
P ROPOSITION 6.3. Let p be a prime. For any pair r and s of elements of an F p -
algebra R, satisfying condition (6.1), one has
(6.19) (rs) p = [s, r ] p−1 rs + r p s p .
This is a corollary of formulae (6.3) and (6.5) combined with the congruences
( p−l )
(6.20) φl (1, . . . , l ) = 0 mod p (2 ≤ l ≤ p − 1)
which form a special case of Proposition 6.2.
A pth -power-of-the-sum identity. The following formula is well known even

though probably not in the form presented below.
L EMMA 6.4. Let r0 and r1 be a pair of elements in an F p -algebra R. One has the
following formula
(6.21)
1
∑
p p
(r0 + r1 ) p − (r0 + r1 ) = [rι , . . . [rι p−2 , [r0 , r1 ]] . . . ]
ι : {1,...,p−2}→{0,1}
1 + | supp ι| 1
where | supp ι | is the cardinality of the support of ι,

supp ι = {1 ≤ j ≤ p − 2 | ι j = 1}.

Summation in (6.21) extends over all functions from {1, . . . , p − 2} to {0, 1},
including the case p = 2 when the domain is the empty set.
C OROLLARY 6.5. Let r and s be a pair of elements in an F p -algebra R, satisfying the
commutation relations
l
adr (s), s = [[r, . . . [r, [r, s]] . . . ], s] = 0 (0 < l < p ).

l times
Then
(r + s) p − (r p + s p ) = [r, . . . [r, [r, s]] . . . ].

p − 1 times

D EFINITION 6.6. We shall call a 2n-tuple of elements (z1 , . . . , zn ; ζ 1 , . . . , ζ n ) in
an arbitrary unital ring R, a CCR-system (of length n), if it satisfies the Canonical
Commutation Relations

1 if i = j
(6.22) [ zi , z j ] = [ ζ i , ζ j ] = 0 and [ζ i , z j ] = .
0 otherwise
If n = 1 we shall call it a CCR-pair.
C OROLLARY 6.7. Suppose that (z1 , . . . , zn ; ζ 1 , . . . , ζ n ) is a CCR-system in an alge-
bra A over an F p -algebra k. Then the p-th power of any linear combination of elements of
the system
− −
(6.23) (c1+ ζ 1 + · · · + c+
n ζ n + c1 z1 + · · · c b z n )
p
(ci± ∈ k; 1 ≤ i ≤ n)
16
equals
p p p −p p p p
ζ n + c1 z1 + · · · + c −
p
(6.24) c1+ ζ 1 + · · · + c+
n n zn
if p > 2, and
2 2 2 − 2 2 2 2
(6.25) c1+ ζ 12 + · · · + c+
n ζ n + c1 z1 + · · · + c −
n zn + (c1+ c1− + · · · + c+ −
n cn )
if p = 2. In particular, the p-th power of any linear combination with coefficients in k
p p
belongs to k if all zi and ζ j belong to k.
The following proposition follows easily from Corollary 6.5 and the congru-
ence
( p − 1)! = −1 mod p (Wilson’s Theorem).
P ROPOSITION 6.8. Let f ∈ k[ X ] be a polynomial,
(6.26) f ( X ) = c0 + c1 X + · · · + c p −1 X p −1 ,
over a commutative F p -algebra k. For any CCR-pair z and ζ in an arbitrary k-algebra A,
one has
(6.27) (ζ + f (z)) p = ζ p + f ( p) (z p ) − c p−1
where f ( p) denotes the Frobenius-twist of f :
f ( p ) ( X ) = c 0 + c 1 X + · · · + c p −1 X p −1 .
p p p
(6.28)
As a corollary we obtain some important non-symplectic isomorphisms be-

tween the algebras of p-symbols.
C OROLLARY 6.9. Given a polynomial (6.26), the correspondence
(6.29) z → z , ζ → ζ + f ( z ),
induces an isomorphism of k-algebras
(6.30) Sab (k) Sa,b− f ( p) (a)+c ( k ).
p −1
P ROOF. The homomorphism of free k-algebras kz, ζ → kz , ζ induced by

correspondence (6.29) sends the three generators of ideal Iab , cf. (1.1), to elements
of ideal Iab where b = b − f ( p) ( a) + c p−1 . Thus, it induces a homomorphism from
Sab to Sab . The inverse is induced by the correspondence
z → z, ζ → ζ − f ( z ).

Consider the following Dependency Condition connecting exponents a and b:
there exists a polynomial (6.26) such that

(D)
b = f ( p ) ( a ) − c p −1 .
C OROLLARY 6.10. If the pair of exponents a, b ∈ k satisfies Dependency Condition

(D), then
Sab Dk (Oa ) M p (k ).

299
When the ground ring is a field, this condition is not only sufficient but also
necessary.
P ROPOSITION 6.11. If k is a field, then [Sab ] = 0 in Br(k) if and only if the exponents
satisfy Dependency Condition (D) introduced above.
The necessity of Condition (D) is an immediate consequence of the following
fact.
P ROPOSITION 6.12. Let z and ζ be any CCR-pair in the matrix algebra M p (k) over
an arbitrary field of characteristic p. Then
(6.31) zp ∈ k and ζ p ∈ k,
the set {z, ζ } generates M p (k) as a k-algebra, and the exponents a = z p and b = ζ p
satisfy Condition (D).
Vice-versa, for any pair a, b ∈ k which satisfies Condition (D), there exists a CCR-pair
z, ζ ∈ M p (k ) with z p = a and ζ p = b.
P ROOF. Suppose z and ζ form a CCR-pair in M p (k), and let φ and ψ be their
respective minimal polynomials. Since φ(z) = ψ(ζ ) = 0, we have
0 = [ζ, φ(z)] = φ (z) and 0 = [ ψ ( ζ ), z ] = ψ ( ζ ),
which implies that φ = ψ = 0. Since the degrees of φ and ψ do not exceed p, we
infer that
φ( X ) = X p − a and ψ( X ) = X p − b
for certain a, b ∈ k. This proves (6.31).
Thus the subalgebra A ⊆ M p (k) generated by z and ζ is isomorphic to a quo-
tient of the algebra of p-symbols, Sab . In view of simplicity of the latter, A is iso-
morphic to Sab . Both Sab and M p (k) have the same dimension as vector spaces
over k, hence A = M p (k).
If a ∈ k p , then exponents a and b satisfy Condition (D) with the polynomial
√ p −1
f ( X ) = −bX p−1 + b p a .
If a ∈ k \ k p , then the polynomial X p − a is irreducible and thus coincides with
the minimal polynomial of z. In particular, z is similar to the companion matrix of
Xp − a
⎛ ⎞
0 a
⎜ ⎟
⎜1 . . . ⎟
(6.32) Z=⎜ ⎜ ⎟
. ⎟
⎝ .. 0 ⎠
1 0
and the latter forms with the matrix
⎛ ⎞
0 1
⎜ 0 2 ⎟
⎜ ⎟
(6.33) Ξ=⎜ .. .. ⎟
⎝ . . ⎠
0 p−1
a CCR-pair. This shows that there exists ξ ∈ M p (k) such that [ξ, z] = 1 and ξ p = 0.
In particular, ζ − ξ belongs to the centralizer of z in M p (k ) which coincides with
k [z] ⊂ M p (k) since the centralizer of matrix Z coincides with k[ Z ] ⊂ M p (k). It
18
follows that ξ = ζ − f (z) for some polynomial f ∈ k[ X ] of degree less than p,

whence
0 = ξ p = b − f ( p ) ( a ) + c p −1 ,
in view of identity (6.27). This completes the proof of the first part of Proposition
6.12
In order to show that any pair of exponents satisfying Condition (D) is real-
ized by some CCR-pair in M p (k), note that for a given a ∈ k, matrices Z and Ξ,
cf. (6.32)–(6.33) above, form a CCR-pair with exponents a and 0, respectively.
If b = f ( p) ( a) − c p−1 , then by replacing Ξ with Ξ + f ( Z ), we obtain a CCR-pair
satisfying z p = a and ζ p = b.
7. The general form of a symplectic isomorphism between n-dimensional

algebras of p-symbols
A 2n × 2n-matrix with coefficients in k
+
D E+
C= ,
E− D −
where D ± , E± ∈ Mn (k), induces a homomorphism of free algebras
(7.1) k z1 , . . . , zn ; ζ 1 , . . . , ζ n → kz1 , . . . , zn ; ζ 1 , . . . , ζ n
by sending zj to
− −
+
d1j z1 + · · · + d +
nj zn + e1j ζ 1 + · · · + enj ζ n
and ζ j to
−
+
e1j z1 + · · · + enj
+
zn + d1j ζ 1 + · · · + d−
nj ζ n .
If C ∈ Sp2n (k), then homomorphism (7.1) induces an isomorphism of the corre-

sponding n-dimensional Weyl algebras
(7.2)
∼
An = kz1 , . . . , zn ; ζ 1 , . . . , ζ n /CCRn −→ An = kz1 , . . . , zn ; ζ 1 , . . . , ζ n /CCRn
where CCRn and CCRn denote the ideals generated by the Canonical Commuta-
tion Relations, cf. (6.22).
For a given 2n-tuple
π = ( a1 , . . . , an ; b1 , . . . , bn ) ∈ k2n
define
(7.3) ρC (π ) = ( a1 , . . . , an ; b1 , . . . , bn ) ∈ k2n
where
(7.4) ⎧ p p p
p
⎪ +
⎨ d1j a1 + · · · + d + −
an + e1j −
b1 + · · · + enj bn p>2
nj
a j : = 2 2 2 2
⎪
⎩ d+ a1 + · · · + d+ an + e− b1 + · · · + e− bn + d+ e− + · · · + d+ e−
1j nj 1j nj 1j 1j nj nj p=2
and
301
(7.5) ⎧ p p p
p
⎪ +
⎨ e1j a1 + · · · + enj
+ −
an + d1j b1 + · · · + d− bn p>2
nj
bj := 2 2 2 2 .
⎪
⎩ e+ a1 + · · · + e+ an + d− b1 + · · · + d− bn + d− e+ + · · · + d− e+
1j nj 1j nj 1j 1j nj nj p=2
Equalities (7.4)–(7.5) can be expressed in a more compact form as follows

πC ( p) if p > 2
(7.6) ρC ( π ) : = −
πC + C · C
( 2 ) + if p = 2

where C ( p) = cij denotes the Frobenius twist of C, and C + , C − ∈ Mn,2n (k) are
p
n × 2n matrices representing the top n, and the bottom n rows of C, respectively.

Finally, C + · C − is the row 2n-vector formed by the dot products of columns of C +
with the corresponding columns of C − ,

C + · C − = C1+ · C1− , . . . , C2n
+ −
· C2n .
It follows from Corollary 6.7 that the ideal in An , generated by the elements
p p p p
z1 − a1 , . . . , zn − an , ζ 1 − b1 , . . . , ζ n − bn ,
is being sent by isomorphism (7.2) to the ideal in An generated by the elements
p p p p
z1 − a1 , . . . , zn − an , ζ 1 − b1 , . . . , ζ n − bn .
Thus (7.2) descends to a homomorphism of the corresponding n-dimensional k-
algebras of p-symbols
(7.7) Sa ,...,an ; b ,...,bn −→ Sa1 ,...,an ; b1 ,...,bn .
1 1
The inverse matrix C −1

induces a homomorphism that is inverse to (7.7).
We arrive at the following theorem that describes a general symplectic isomor-
phism between algebras of p-symbols.
T HEOREM 7.1. For any
π = ( a1 , . . . , an ; b1 , . . . , bn ) ∈ k2n ,
and any symplectic matrix C ∈ Sp2n (k), one has the following canonical isomorphisms
(7.8) Sa b ⊗ · · · ⊗ San bn Sa ,...,an ; b ,...,bn Sa1 ,...,an ; b1 ,...,bn Sa1 b1 ⊗ · · · ⊗ San bn
1 1 1 1
where
( a1 , . . . , an ; b1 , . . . , bn ) = ρC (π )
is given by equalities (7.3)–(7.5).
In a compact form, (7.8) can be expressed as

( p > 2) SπC( p)
(7.9) = Sρ C ( π ) Sπ .
( p = 2) SπC(2) +C+ ·C−

As mentioned in Remark 2.4, tensor identities (2.9)–(2.13) of Section 2 are noth-
ing but special cases of identity (7.8) for suitably chosen symplectic matrices C.
20
8. Cyclic p-algebras as algebras of p-symbols

For b ∈ k∗ and c ∈ k, let us denote by (b, c]k the quotient of the free k-algebra
k ζ, η by the ideal Jbc = Jbc (k) generated by the following three relations
(8.1) ζ p = b, ηp = η + c and ζη = (η + 1)ζ.
We shall omit subscript k if the ground ring is clear from the context.
Algebra (b, c] is the crossed product
k[η ]
(8.2) (b, c] = (Z/pZ )
(η p − η − c) b
with the twisting cocyle

∗ 1 if i + j < p
Z/pZ × Z/pZ → k , (i, j) → ,
b if i + j ≥ p
where 0 ≤ i, j < p.
These algebras were perhaps first introduced by Hermann Ludwig Schmid
in his 1934 Ph. D Thesis at Mahrburg ([11], §2) with Helmut Hasse acting as his
advisor. They were studied also by Teichmüller and Witt. Notation adopted here
is the one used by Teichmüller in [12], and is a slight modification of the notation
employed for symbol pairings in Algebraic Number Theory. Algebras (b, c] form
a special class of the so called cyclic p-algebras.
P ROPOSITION 8.1. The correspondence
(8.3) η → zζ, ζ → ζ,
induces an isomorphism of algebras
(8.4) (b, c] Scb−1 ,b .
P ROOF. Proposition 6.3 guarantees that the homomorphism k ζ, η → Scb−1 ,b
induced by correspondence (8.3) descends to a homomorphism of k-algebras
(8.5) (b, c] → Scb−1 ,b .
In order to show that (8.5) is an isomorphism, let us consider the homomorphism
kz, ζ → (b, c] induced by the correspondence
z → z := b−1 ηζ p−1 , ζ → ζ.
Note, that (z , ζ ) is a CCR-pair in (b, c],
[ζ, z ] = b−1 ((η + 1)ζ p − ηζ p ) = 1.
Identity (6.19) is thus applicable and yields
η p = η + (z ) p ζ p
which combined with (8.1) implies that (z ) p b = c. It follows that the homomor-
phism k z, ζ → (b, c] descends to a homomorphism of k-algebras Scb−1 ,b → (b, c]
which supplies the inverse to homomorphism (8.5).
Tensor identities of Proposition 2.2 yield several canonical isomorphisms in-

volving cyclic algebras. We would like to record just one.
303
C OROLLARY 8.2. For any pair b ∈ k∗ , and c ∈ k, there exists a canonical isomor-
phism
(b, c]⊗ p Dk (Ocb−1 )⊗( p−1) ⊗ Dk (Ob ) M p (k)⊗ p .
In the case when the ground ring, k, is a field, we can say more.
T HEOREM 8.3. Any element of order p in the Brauer group, Br(k ), is represented by
an algebra of p-symbols
Sπ = Sa1 ,...,an ; b1 ,...,bn ,
cf. (2.22), for some π = ( a1 , . . . , an ; b1 , . . . , bn ) ∈ k2n and n ≥ 1.
P ROOF. Let us invoke a few known facts about central simple algebras over a
field k of characteristic p > 0. The absolute Frobenius map
(8.6) F : k → k, c → c p ,
induces on Br(k) the endomorphism of multiplication by p. Map (8.6) can be repre-
sented as the canonical inclusion k → k1/p followed by the isomorphism k1/p k.
This means that if the similarity class [ A] ∈ Br(k) of an algebra A has order p in
Br(k ), then A is split by a finitely generated subfield k(u1 , . . . , un ) ⊂ k1/p .
A theorem of Teichmüller ([T], Satz 29) implies existence of b, c ∈ k∗ such
that A ⊗ (b, c] splits over k(u1 , . . . , un−1 ). By applying this factorization argument
repetitively we find that there exists
(b1 , . . . , bn , c1 , . . . , cn ) ∈ (k∗ )2n
such that
A ⊗ (b1 , c1 ] ⊗ · · · ⊗ (bn , cn ]
splits over k, and this means that the opposite algebra, Aop , is similar to
(b1 , c1 ] ⊗ · · · ⊗ (bn , cn ].
It remains now to invoke Proposition 8.1.
References
[1] Pierre Berthelot, Cohomologie cristalline des schémas de caractéristique p > 0, Lecture Notes in Mathe-
matics 451, Springer-Verlag, 1974.
[2] Roman Bezrukavnikov, Ivan Mirković, Dmitriy Rumynin, Localization of modules for a semisimple Lie
algebra in prime characteristic, Annals of Mathematics 167 (2008), 945–991.
[3] Paul Adrien Dirac, On Quantum Algebra, Proc. Cambridge Philos. Soc. 23 (1926), 412–418.
[4] Jacques Dixmier, Sur les algèbres de Weyl, Bull. Soc. Math. France 96 (1968), 209–242.
[5] Kenneth G. Goodearl, Robert B. Warfield, Jr., An Introduction to Noncommutative Noetherian Rings,
Second Edition, London Mathematical Society Student Texts 61, Cambridge University Press, 2004.
[6] William Ogilvy Kermack, William H McCrea, On Professor Whittaker’s solution of differential
equations by definite integrals. Part II. Applications of the methods of non-commutative algebra,
Proc. Edin. Math. Soc. (2), 2 (1931), 220–239.
[7] Max-Albert Knus, Manuel Ojanguren, Théorie de la Descente at Algèbres d’Azumaya, Lecture Notes
in Mathematics 389, Springer-Verlag, 1974.
[8] Dudley Ernest Littlewood, On the Classification of Algebras, Proc. London Math. Soc. (2) 35 (1933),
200–240.
[9] Öystein Ore, Theory of Noncommutative Polynomials, Annals of Mathematics 34 (1933), 480–508.
[10] Philippe Revoy, Algèbres de Weyl en caracteristique p, C. R. Acad. Sci. Paris Sér. A 276 (1973), 225–228.
[11] Hermann Ludwig Schmid, Über das Reziprozitätsgesetz in relativ-zyklischen algebraischen Funktio-
nenkörpern mit endlichem Konstantenkörper, Math. Z. 40 (1935/6), 94–119.
22
[12] Oswald Teichmüller, p-Algebren, Deutsche Mathematik 1 (1936), 362–388 (reprinted in: Oswald Te-
ichmüller, Gesammelte Anhandlungen, herausgegeben von L. V. Ahlfors and F. W. Gehring, Springer-
Verlag, 1982, pp. 120–146).
[13] Joseph Henry Maclagan Wedderburn, Non-commutative domains of integrity, J. f. d. reine u. ange-
wandte Mathematik 167 (1932), 129–141.
[14] Hermann Weyl, Gruppentheorie und Quantenmechanik, S. Hirzel, Leipzig 1928.
[15] Mariusz Wodzicki, Notes on Differential Operators, Fall 2007, 66p.
D EPARTMENT OF M ATHEMATICS , U NIVERSITY OF C ALIFORNIA , B ERKELEY, C ALIFORNIA 94720-

3840
E-mail address: wodzicki@math.berkeley.edu
Volume 546, 2011
Large scale geometry and its applications
Guoliang Yu
Dedicated to Henri Moscovici on the occasion of his sixty fifth birthday
1. Introduction
Large scale geometry is the study of geometric objects observed from a great
distance. The idea of large scale geometry played a major role in the work of Mostow
and Margulis on rigidity of lattices [Mo] [Ma] and the work of Gromov and others
in geometric group theory [G1]. Inspired by the work of Connes-Moscovici [CM],
Roe introduced a large scale geometric method in studying index theory of elliptic
operators on noncompact manifolds [R1]. Roe’s large scale index theory has led
to several important development in K-theory and index theory with applications
to problems in differential geometry and topology such as the Novikov conjecture.
More recently, Guentner, Tessera and myself introduced a large scale geometric
method in studying rigidity of manifolds [GTY]. The purpose of this article is to
give a survey on the large scale geometric aspect of these work.
This paper is organized as follows. In Section 2, we recall the concept of coarse
equivalence and quasi-isometry. In Section 3, we discuss Gromov’s concept of finite
asymptotic dimension. In Section 4, we introduce the notion of decomposition
complexity. In Section 5, we disuss Property A and coarse embeddings. Finally, in
Section 6, we give an overview of the applications of these large scale properties to
the topology and geometry of manifolds.
I would like to thank Henri for his encouragement and great friendship.
2. Coarse equivalence and quasi-isometry

In this section, we recall the concept of coarse equivalence and quasi-isometry.
Throughout this paper, we focus on proper metric spaces. Recall that a metric
space is proper if every closed ball is compact. For application purpose, the two
most interesting classes of metric spaces are countable groups with proper length
metrics and Riemannian manifolds.
Definition 2.1. Let X and Y be two proper metric spaces.
2010 Mathematics Subject Classification. Primary: 53C24. Secondary: 53C23, 58J22.

The author is partially supported by a grant from the U.S. National Science Foundation.
2011
c 0000
c Mathematical
holder)
1
305
306
2 GUOLIANG YU
(1) A map f : X → Y is said to be coarse if f is proper in the sense that

the inverse image of every compact set is compact and there exists a
non-decreasing function ρ on R+ = [0, ∞) such that d(f (x1 ), f (x2 )) ≤
ρ(d(x1 , x2 )) for all x1 and x2 in X (such a ρ is called the control function
of f );
(2) X is said to be coarsely equivalent to Y if there exist coarse maps f :
X → Y and g : Y → X and a non-negative constant C such that f ◦ g
and g ◦ f are C-close to the identity maps (two maps are called C-close
if the images of each point in X under the two maps are at most distance
C from each other).
(3) X and Y are said to be quasi-isometric if the above f and g can be chosen
to have linear control functions.
Large scale scale geometry is the study of geometric properties invariant under
coarse equivalences.
A non-negative function l on a countable group G is called a length function
if (1) l(g −1 ) = l(g) for all g ∈ G; (2) l(gh) ≤ l(g) + l(h) for all g and h in G;
(3) l(g) = 0 if and only if g = e, the identity element of G. We can associate a
left-invariant length metric dl to l: dl (g, h) = l(g −1 h) for all g, h ∈ G. A length
metric is called proper if the length function is a proper map (i.e. the inverse image
of every compact set is finite in this case). It is not difficult to show that every
countable group G has a proper length metric. If l and l are two proper length
functions on G, then their associated length metrics are coarsely equivalent. If G
is a finitely generated group and S is a finite symmetric generating set (symmetric
in the sense that if an element is in S, then its inverse is also in S), then we can
define the word length lS on G by
lS (g) = min{n : g = s1 · · · sn , si ∈ S}.

If S and S are two finite symmetric generating sets of G, then their associated
proper length metrics are quasi-isometric.
If M is a compact Riemannian manifold, then its universal cover M (with a
Riemannian metric lifted from M ) is quasi-isometric to the fundamental group of
M with a word length metric.
A metric space X is said to be quasi-geodesic if there exist δ > 0, C0 ≥ 0 and
C1 > 0 such that for each pair of points x and y in X, there exist a sequence of
points x0 = x, x1 , · · · , xn = y in X satisfying d(xi , xi+1 ) ≤ δ and C1−1 n − C0 ≤
d(x, y) ≤ C1 n + C0 . A finitely generated group with a word length metric is quasi-
geodesic. Two quasi-geodesic metric spaces are coarsely equivalent if and only they
are quasi-isometric.
3. Asymptotic dimension
In this section, we discuss Gromov’s concept of asymptotic dimension. This
concept is a large scale analogue of the covering dimension in topology.
Definition 3.1. The asymptotic dimension of a proper metric space X is the
smallest integer n such that for every r > 0, there exists a uniformly bounded cover
{Ui } for which the number of Ui intersecting each r ball B(x, r) is at most n + 1.
As examples the asymptotic dimension of Zn is n and the asymptotic dimen-
sion of the free group Fn with n generators is 1. The asymptotic dimension is
LARGE SCALE GEOMETRY AND ITS APPLICATIONS 307
3
invariant under coarse equivalence. The Lie group GL(n, R) with a left invariant
Riemannian metric is quasi-isometric to T (n, R), the subgroup of invertile upper
triangular matrices. By permanence properties of asymptotic dimension [BD1],
we know that the solvable group T (n, R) has finite asymptotic dimension. As a
consequence, every countable discrete subgroup of GL(n, R) has finite asymptotic
dimension (as a metric space with a proper length metric). More generally one
can prove that every discrete subgroup of an almost connected Lie group has finite
asymptotic dimension (a Lie group is said to be almost connected if the number
of its connected components is finite). Gromov’s hyperbolic groups also have finite
asymptotic dimension [R3].
IfG ⊂ GL(2, R) is the finitely generated group consisting of all matrices of the
π n p(π)
form for any n ∈ Z and Laurent polynomial p with Z coefficients,
0 π −n
then G has infinite asymptotic dimension. This follows from the observation that
G contains the abelian group ⊕+∞ k=−∞ Z as a subgroup.
It is an open question whether the fundamental group of a compact aspherical
manifold has finite asymptotic dimension (a manifold is called aspherical if its
universal cover is contractible).
4. A notion of geometric complexity

In this section, we discuss the concept of decomposition complexity introduced
by Guentner-Tessera-Yu [GTY].
For any r > 0, a collection of subspaces { Zi } of a metric space Z is said to be
r-disjoint if for all i
= j we have d(Zi , Zj ) ≥ r. To express the idea that Z is the
union of subspaces Zi , and that the collection of these subspaces is r-disjoint we
write
Z= Zi .
r−disjoint
A family of of metric spaces { Zi } is called bounded if there is a uniform bound on

the diameter of the individual Zi :
sup diameter(Zi ) < ∞.
Definition 4.1. A family of metric spaces {X} is r-decomposable over another
family of metric spaces {Y } if every X ∈ {X} admits an r-decomposition

X = X 0 ∪ X1 , Xi = Xij ,
r−disjoint
where each Xij ∈ {Y }.

Definition 4.2. Let Ω be a collection of families of metric spaces. A family
of metric spaces {X} is said to be decomposable over Ω if, for every r > 0, there
exists a family of metric spaces {Y } ∈ Ω and an r-decomposition of {X} over {Y }.
The collection Ω is said to be stable under decomposition if every family of metric
spaces which decomposes over Ω actually belongs to Ω.
Definition 4.3. The collection D of families of metric spaces with finite de-
composition complexity is the minimal collection of families of metric spaces con-
taining all bounded families of metric spaces and stable under decomposition. We
abbreviate membership in D by saying that a family of metric spaces in D has finite
308
4 GUOLIANG YU
decomposition complexity. A metric space X is said to have finite decomposition

complexity if the family consisting of only X has finite decomposition complexity.
Observe that finite decomposition complexity is invariant under coarse equiv-
alence.
By the definition of asymptotic dimension, any proper metric space with as-
ymptotic dimension at most 1 has finite decomposition complexity. More generally
a metric space with finite asymptotic dimension has finite decomposition complex-
ity. This fact follows from a theorem of Dranishnikov and Zarichnyi stating that
a proper metric space with finite asymptotic dimension is coarsely equivalent to a
subspace of the product of finitely many trees.
Let G = ⊕∞ k=1 Z be the countable group with the proper length metric associ-
ated to the length funtcion l:
∞

∞
l(⊕k=1 nk ) = k|nk |
k=1
for each ⊕∞
k=1 nk ∈ G. For each r > 0, let k0 be the smallest integer greater than r.
For each α ∈ ⊕∞k=k0 Z, let
0 −1 0 −1 0 −1
Gα = {⊕kk=1 nk ⊕ α : ⊕kk=1 nk ∈ ⊕kk=1 Z}.
0 −1
Notice that G = r−disjoint Gα and Gα is (uniformly) coarsely equivalent to ⊕kk=1 Z.
This implies that {Gα } ∈ D. It follows that G has finite decomposition complex-
ity despite the fact that G has infinite asymptotic dimension. If G⊂ GL(2, R) is
π n p(π)
the finitely generated group consisting of all matrices of the form
0 π −n
as in the previous section, with a little extra work we can show that G has finite
decomposition complexity.
More generally, we have the following result from [GTY].
Theorem 4.4. Any countable subgroup of GL(n, k) has finite decomposition
complexity (as a metric space with a proper length metric), where k is a field.
The same result is true for any countable subgroup of an almost connected Lie
group and any countable elementary amenable group [GTY].
It is an open question whether any countable amenable group has finite decom-
position complexity.
5. Property A and coarse embeddability

In this section, we discuss the concept of Property A and and coarse embed-
dability into Banach spaces.
Any proper metric space is coarsely equivalent to a locally finite metric space
(a metric space is called locally finite if every ball contains only finitely many
elements).
The following definition was introduced in [Y2] to study coarse embeddability
into Hilbert space.
Definition 5.1. Let Γ be a locally finite metric space. Γ is said to have Prop-
erty A if ∀ r > 0, ε > 0, there exists a family of finite subsets {Aγ }γ∈Γ of Γ × N
such that
(1) (γ, 1) ∈ Aγ for all γ ∈ Γ;
5
#(Aγ − Aγ ) + #(Aγ − Aγ )
(2) <ε
#(Aγ ∩ Aγ )
if d(γ, γ ) ≤ r;
(3) there exists R > 0 such that if (x, m) ∈ Aγ and (y, n) ∈ Aγ for some γ, then
d(x, y) ≤ R.
Observe that Property A is invariant under coarse equivalence.
With the help of Folner sets, one can easily verify that any countable amenable
group has Property A.
Let Γ = Fn , the free group on n generators. Fix a geodesic ray σe on Fn starting
from the identity e (a geodesic ray in a finitely generated group with a word metric
is a subspace of Fn isometric to the metric space of non-negative integers with the
standard metric). For each γ ∈ Fn , there exists a geodesic ray σγ starting from γ
such that σγ coincides with σe outside a compact set. Define
Aγ = {(x, 1) ∈ Fn × N, x ∈ σγ , d(x, γ) ≤ N }
for some large natural number N . It is not difficult to see that {Aγ } satisfies the
conditions in the definition of Property A.
By a theorem in [GTY], any metric space with finite decomposition complex-
ity implies Property A. Previously, Higson and Roe proved that finite asymptotic
dimension implies Property A [HR] and Higson-Guentner-Weinberger proved that
all linear groups have Property A [HGW].
It is an open problem whether every Cat(0) space has Property A. Important
special cases have been obtained in [BCGNW] and [C].
Definition 5.2. (Gromov): Let Γ be a metric space and X be a Banach space.
A map f : Γ → X is said to be a coarse embedding if there exist non-decreasing
functions ρ1 and ρ2 on [0, ∞) such that
(1) ρ1 (d(x, y)) ≤ dX (f (x), f (y)) ≤ ρ2 (d(x, y)) for all x, y ∈ Γ;
(2) limr→+∞ ρ1 (r) = +∞.
Roughly speaking, coarse embeddability of a metric space Γ in a Banach space
X means that one can draw a good picture of Γ in X which reflects the large
scale geometry of Γ. A characterization of metric spaces coarsely embeddable into
Hilbert space is given by Tessera [T].
The following result is inspired by a theorem of Bekka-Cherix-Valette [BCV].
Theorem 5.3. If a discrete metric space Γ has Property A, then Γ admits a
coarse embedding into Hilbert space.
Proof. Let
∞

H= 2 (Γ × N).
k=1
(k)
By the definition of property A, there exists a family of finite subsets {Aγ }γ∈Γ
such that
(k)
(1) (γ, 1) ∈ Aγ for all γ ∈ Γ;
(k) (k)
(2) ∃ Rk > 0 such that if (x, m) ∈ Aγ , (y, n) ∈ Aγ for some k and γ ∈ Γ,
then d(x, y) ≤ Rk ;
(3)
χ (k) χA(k)
Aγ γ 1
− < k
(#Aγ )1/2
(k)
(#Aγ )1/2 l2 (Γ×N)
(k) 2
310
6 GUOLIANG YU
if d(γ, γ ) ≤ k and k ∈ N, where χA(k) is the characteristic function of Aγ .

(k)
γ
Fix γ0 ∈ Γ.
Define f : Γ → H by
∞

χA(k) χA(k)
−
γ γ 0
f (γ) = (k) 1/2 (k) 1/2
.
k=1 (#A γ ) (#A γ 0 )
We have
∞

χA(k) χA(k)
f (γ) − f (β) = (k)
γ
− (k)
β
.
k=1 (#Aγ )1/2 (#Aβ )1/2
It is not diffcult to prove that f is a coarse embedding.
Examples of groups coarsely embeddable into Hilbert space include countable
subgroups of connected Lie groups [HGW], hyperbolic groups [S], amenable groups
[BCV], Coxeter groups [DJ], mapping class groups [Ki] [Ha], and semi-direct prod-
ucts of groups of the above types.
It is an open question whether every countable subgroup of the diffeomorphism
group of the circle is coarsely embeddable into Hilbert space. The same question
for Out(Fn ) is also open.
The first example of a metric space not coarsely embeddable into Hilbert space
was given in [DGLY]. Gromov showed that any sequence of expanding graphs does
not coarsely embed into Hilbert space [G3]. The only known examples of groups not
coarsely embeddable into Hilbert spaces are Gromov’s random group constructed
with the help of expanders [G4] [AD]. Nowak constructed the first example of met-
ric spaces coarsely embeddable into Hilbert space but without Property A. Very
recently Arzhantseva, Guentner and Spakula constructed the first bounded geom-
etry space coarsely embeddable into Hilbert space but without Property A (recall
that a metric space is said to have bounded geometry if it is coarsely equivalent to a
locally finite metric space such that for each r > 0, there exists a positive integer N
for which every ball with radius r has at most N elements). It is an open question
to construct a countable group coarsely embeddable into Hilbert space but without
Property A.
Let Γ be a countable group with a proper length metric, let X = l∞ (Γ). Fix
γ0 ∈ Γ. We define a map f : Γ → X by:
γ → (f (γ))(x) = d(x, γ) − d(x, γ0 )
for every γ ∈ Γ. It is not difficult to check that f is an isometric embedding. This
construction tells us that l∞ (Γ) is universal. Brown and Guentner proved that Γ
can be coarsely embedded into a reflexive and strictly convex Banach space [BG].
For the purpose of applications, we need to impose a certain mild condition on
the Banach space.
Definition 5.4. A Banach space X is said to have Property (H) if there exist
an increasing sequence of finite dimensional subspaces {Vn } of X and an increasing
sequence of finite dimensional subspaces {Wn } of a Hilbert space such that
(1) V = ∪n Vn is dense in X,
(2) there exists a uniformly continuous map ψ : S(V ) → S(W ) such that the
restriction of ψ to S(Vn ) is a homeomorphism (or more generally a degree
one map) onto S(Wn ) for each n, where W = ∪n Wn , S(V ) and S(W )
are respectively the unit spheres of V and W .
7
As an example, let X be the Banach space lp (N) for some p ≥ 1. Let Vn and
Wn be respectively the subspaces of lp (N) and l2 (N) consisting of all sequences
whose coordinates are zero after the n-th terms. We define a map ψ from S(V ) to
S(W ) by
ψ(c1 , · · · , ck , · · · ) = (c1 |c1 |p/2−1 , · · · , ck |ck |p/2−1 , · · · ).
ψ is called the Mazur map. It is not difficult to verify that ψ satisfies the conditions
in the definition of Property (H). For each p ≥ 1, we can similarly prove that Cp ,
the Banach space of all Schatten p-class operators on a Hilbert space, has Property
(H).
Let c0 be the Banach space consisting of all sequences of real numbers that are
convergent to 0. It is an open question whether c0 has Property (H). As we shall
see, a positive answer to this question would imply the Novikov conjecture since
every countable group admits a coarse embedding into c0 [BG].
It is conjectured that every countable subgroup of the diffeomorphism group
of a compact smooth manifold is coarsely embeddable into Cp for some p ≥ 1. For
each p > q ≥ 2, it is also an open question to construct a bounded geometry space
which is coarsely embeddable into lp (N) but not lq (N). Beautiful results in [JR]
and [MN] indicate that such a construction should be possible.
6. Applications to geometry and topology

In this section, we discuss applications of large scale geometry to the geometry
and topology of manifolds.
One could ask the question whether coarse equivalence of two Riemannian
manifolds would imply that the two manifolds are homeomorphic. In general the
answer is no. The obstruction is the local non-triviality of topological data. The
following concept of Gromov removes the non-triviality of topological data.
Definition 6.1. A Riemannian manifold M is said to be uniformly contractible
if for each r > 0, there exists R ≥ r such that each ball B(x, r) can be contracted
to a point within B(x, R) for every x ∈ M .
An example of a uniformly contractible manifold is the universal cover of a com-
pact aspherical manifold. Davis constructed beautiful aspherical manifolds whose
universal covers are not homeomorphic to Rn [D].
Conjecture 6.1. (Geometric Borel Conjecture) Let M n and N n be two uni-
formly contractible complete Riemannian manifolds with bounded geometry. If M n
and N n are coarsely equivalent, then there exists a homeomorphism between M and
N implementing the given coarse equivalence.
The following rigidity result is proved in [GTY].
Theorem 6.2. If n ≥ 5, then the geometric Borel conjecture holds for uniformly
contractible complete Riemannian manifolds with finite decomposition complexity.
Conjecture 6.2. (Stable Borel Conjecture) If two compact aspherical mani-
folds M and N are homotopy equivalent, then M × Rn is homeomorphic to M × Rn
for some non-negative integer n.
Observe that the homotopy equivalence between two compact manifolds implies
that their universal covers are coarsely equivalent. The Borel conjecture claims
that the above conjecture is true for n = 0. Farrell-Hsiang proved the stable Borel
312
8 GUOLIANG YU
conjecture for non-positively curved manifolds [FH]. Important results on the Borel
conjecture were obtained by Farrell-Jones [FJ1] [FJ2] [FJ3] [FJ4] and Bartels-Lück
[BL].
The following result is proved in [GTY].
Theorem 6.3. The stable Borel conjecture holds for aspherical manifolds whose
fundamental groups have finite decomposition complexity.
By a deep theorem of Novikov [Nov], the stable Borel conjecture implies the
following conjecture.
Conjecture 6.3. (The Novikov conjecture for aspherical manifolds) The ratio-
nal Pontryagin classes of compact smooth aspherical manifolds are invariant under
orientation preserving homotopy equivalence.
The Novikov conjecture for aspherical manifolds should be viewed as an infin-
itesimal version (or linearization) of the (stable) Borel conjecture. The Novikov
conjecture for general manifolds states that higher signatures are homotopy invari-
ant. The Novikov conjecture for any manifold follows from the strong Novikov
conjecture for its fundamental group, which provides an algorithm for determining
nonvanishing of higher indices of elliptic operators.
The following result on the strong Novikov conjecture is proved in [KY].
Theorem 6.4. Let Γ be a countable discrete group. If Γ admits a coarse em-
bedding into a Banach space with Property (H), then the Strong Novikov conjecture
holds for Γ, i.e. the Baum-Connes assembly map
μ : K∗Γ (EΓ) → K∗ (Cr∗ Γ)
is injective, where EΓ is the universal space for proper Γ-actions and Cr∗ Γ is the
reduced group C ∗ -algebra.
Corollary 6.5. The Novikov conjecture holds for manifolds whose fundamen-
tal groups are coarsely embeddable into a Banach spaces with Property (H).
In the case of aspherical manifolds, fundamental groups completely decide the
homotopy types of aspherical manifolds since their higher homotopy groups are
trivial. Roughly speaking, the above corollary says that if one can draw a good
picture of the fundamental group on a reasonably “good” blackboard, then the
manifold can be recognized at the infinitesimal level.
The special case when the Banach space is the Hilbert space is proved in [Y2]
and [STY]. The important cases of hyperbolic groups, discrete groups of almost
connected Lie groups and amenable groups are respectively due to [CM] [K] [HK].
We also remark that the Novikov conjecture for Gromov’s random groups remains
true since they are inductive limits of hyperbolic groups.
We should mention another important application of the strong Novikov con-
jecture to differential geometry. With the help of the higher index theory of the
Dirac operator, Rosenberg proved that the strong Novikov conjecture implies the
Gromov-Lawson-Rosenberg conjecture that a compact aspherical manifold does not
admit a Riemannian metric with positive scalar curvature [Ro].
References
[AD] G.N. Arzhantseva, T. Delzant. Examples of random groups. Preprint, 2008.
[B] A. Bartels. Squeezing and higher algebraic K-theory. K-Theory 28 (2003), no. 1, 19–37.
9
[BL] A. Bartels, W. Lück. The Borel Conjecture for hyperbolic and CAT(0)-groups.
arXiv:0901.0442v1, 2009.
[BR] A. Bartels, D. Rosenthal. On the K-theory of groups with finite asymptotic dimension.
J. Reine Angew. Math. 612 (2007), 35–57.
[BC] P. Baum, A. Connes, K-theory for discrete groups, Operator Algebras and Applications. (D.
Evans and M. Takesaki, editors), Cambridge University Press (1989), 1–20.
[BCH] P. Baum, A. Connes, N. Higson, Classifying space for proper actions and K-theory of
group C ∗ -algebras. C ∗ -algebras: 1943–1993 (San Antonio, TX, 1993), 240–291, Contemp.
Math., 167, Amer. Math. Soc., Providence, RI, 1994.
[BCV] M. E. B. Bekka, P. A. Cherix, A. Valette. Proper affine isometric actions of amenable
groups. Novikov conjectures, index theorems and rigidity, Vol. 2 (Oberwolfach, 1993), 1–4,
London Math. Soc. Lecture Note Ser., 227, Cambridge Univ. Press, Cambridge, 1995.
[BD1] G. Bell, A. Dranishnikov. On asymptotic dimension of groups. Algebr. Geom. Topol. 1
(2001), 57–71 (electronic).
[BD2] G. Bell, A. Dranishnikov. On asymptotic dimension of groups acting on trees. Geom.
Dedicata 103 (2004), 89–101.
[BD3] G. Bell, A. Dranishnikov. A Hurewicz-type theorem for asymptotic dimension and ap-
plications to geometric group theory. Trans. Amer. Math. Soc. 358 (2006), no. 11, 47494764.
[BCGNW] J. Brodzki, S. Campbell, E. Guentner, N. Wright. Property A and CAT(0) cube
complexes. J. Funct. Anal. 256 (2009), no. 5, 1408–1431.
[BG] N. Brown, E. Guentner. Uniform embeddings of bounded geometry spaces into reflexive
Banach space. Proc. Amer. Math. Soc. 133 (2005), no. 7, 2045–2050.
[C] S. Campbell. Property A and affine buildings. J. Funct. Anal. 256 (2009), no. 2, 409–431.
[CG] G. Carlsson, B. Goldfarb. The integral K-theoretic Novikov conjecture for groups with
finite asymptotic dimension. Invent. Math. 157 (2004), no. 2, 405–418.
[CP] G. Carlsson, E. Pedersen. Controlled algebra and the Novikov conjectures for K- and
L-theory. Topology 34 (1995), no. 3, 731–758.
[CFY] S. Chang, S. Ferry, G. Yu. Bounded rigidity of manifolds and asymptotic dimension
growth. J. K-Theory 1 (2008), no. 1, 129–144.
[CTWY] X. Chen, R. Tessera, X. Wang, G. Yu. Metric sparsification and operator norm local-
ization. Adv. Math. 218 (2008), no. 5, 1496–1511.
[C1] A. Connes. Noncommutative differential geometry. Inst. Hautes Études Sci. Publ. Math.
No. 62 (1985), 257–360.
[C2] A. Connes. Noncommutative Geometry. Academic Press, 1994.
[C3] A. Connes. Cyclic cohomology and the transverse fundamental class of a foliation. Geo-
metric methods in operator algebras (Kyoto, 1983), 52–144, Pitman Res. Notes Math. Ser.,
123, Longman Sci. Tech., Harlow, 1986.
[CM] A. Connes, H. Moscovici. Cyclic cohomology, the Novikov conjecture and hyperbolic
groups. Topology 29 (1990), no. 3, 345–388.
[CGM] A. Connes, M. Gromov, H. Moscovici. Group cohomology with Lipschitz control and
higher signatures. Geom. Funct. Anal. 3 (1993), no. 1, 1–78.
[DG] M. Dadarlat, E. Guentner. Constructions preserving Hilbert space uniform embeddability
of discrete groups. Trans. Amer. Math. Soc. 355 (2003), no. 8, 3253–3275.
[D] M. Davis. Groups generated by reflections and aspherical manifolds not covered by Euclidean
space. Ann. of Math. (2) 117 (1983), no. 2, 293–324.
[DGLY] A. Dranishnikov, G. Gong, V. Lafforgue, G. Yu. Uniform embeddings into Hilbert
space and a question of Gromov. Canad. Math. Bull. 45 (2002), no. 1, 60–70.
[DJ] A. Dranishnikov, T. Januszkiewicz. Every Coxeter group acts amenably on a compact
space. Proceedings of the 1999 Topology and Dynamics Conference (Salt Lake City, UT).
Topology Proc. 24 (1999), Spring, 135–141.
[DS] A. Dranishnikov, J. Smith. Asymptotic dimension of discrete groups. Fund. Math. 189
(2006), no. 1, 27–34.
[DFW] A. Dranishnikov, S. Ferry, S. Weinberger. Large Riemannian manifolds which are
flexible. Ann. of Math. (2) 157 (2003), no. 3, 919–938.
[DZ] A. Dranishnikov, M. Zarichnyi. Universal spaces for asymptotic dimension. Topology
Appl. 140 (2004), no. 2-3, 203–225.
[FH] F. T. Farrell, W. C. Hsiang. On Novikov’s conjecture for nonpositively curved manifolds.
Ann. of Math. (2) 113 (1981), no. 1, 199–209.
314
10 GUOLIANG YU
[FJ1] T. Farrell, L. Jones. A topological analogue of Mostow’s rigidity theorem. J. Amer. Math.
Soc. 2 (1989), no. 2, 257–370.
[FJ2] T. Farrell, L. Jones. Classical aspherical manifolds. CBMS Regional Conference Series in
Mathematics, 75. Published for the Conference Board of the Mathematical Sciences, Wash-
ington, DC; by the American Mathematical Society, Providence, RI, 1990.
[FJ3] T. Farrell, L. Jones. Topological rigidity for compact non-positively curved manifolds. Dif-
ferential geometry: Riemannian geometry (Los Angeles, CA, 1990), 229–274, Proc. Sympos.
Pure Math., 54, Part 3, Amer. Math. Soc., Providence, RI, 1993.
[FJ4] T. Farrell, L. Jones. Rigidity for aspherical manifolds with π1 ⊂ GLm (R). Asian J.
Math. 2 (1998), no. 2, 215–262.
[FP] S. Ferry, E. Pedersen. Epsilon surgery theory. Novikov conjectures, index theorems and
rigidity, Vol. 2 (Oberwolfach, 1993), 167–226, London Math. Soc. Lecture Note Ser., 227,
Cambridge Univ. Press, Cambridge, 1995.
[FW1] S. Ferry, S. Weinberger. Curvature, tangentiality, and controlled topology. Invent. Math.
105 (1991), no. 2, 401–414.
[FW2] S. Ferry, S. Weinberger. A coarse approach to the Novikov conjecture. Novikov conjec-
tures, index theorems and rigidity, Vol. 1 (Oberwolfach, 1993), 147–163, London Math. Soc.
Lecture Note Ser., 226, Cambridge Univ. Press, Cambridge, 1995.
[G1] M. Gromov. Hyperbolic groups. Essays in group theory, 75–263, Math. Sci. Res. Inst. Publ.,
8, Springer, New York, 1987.
[G2] M. Gromov. Asymptotic invariants of infinite groups. Geometric group theory, Vol. 2 (Sus-
sex, 1991), 1–295, London Math. Soc. Lecture Note Ser., 182, Cambridge Univ. Press, Cam-
bridge, 1993.
[G3] M. Gromov. Spaces and questions. GAFA 2000 (Tel Aviv, 1999). Geom. Funct. Anal. 2000,
Special Volume, Part I, 118–161.
[G4] M. Gromov. Random walks in random groups. Geom. Funct. Anal. 13 (2003), no. 1, 73–146.
[G] E. Guentner, Permanence in coarse geometry. Preprint 2010.
[GHW] E. Guentner, N. Higson, S. Weinberger. The Novikov Conjecture for Linear Groups.
Publ. Math. Inst. Hautes tudes Sci. No. 101 (2005), 243–268.
[GTY] E. Guentner, R. Tessera, E. Guentner. A notion of geometric complexity and its ap-
plications to topological rigidity. Preprint, 2008.
[H] N. Higson. Bivariant K-theory and the Novikov conjecture. Geom. Funct. Anal. 10 (2000),
no. 3, 563–581.
[HP] U. Haagerup, A. Przybyszewska. Proper metrics on locally compact groups, and proper
affine isometric actions on Banach spaces. Preprint.
[Ha] U. Hamenstaedt . Geometry of the mapping class groups. I. Boundary amenability. Invent.
Math. 175 (2009), no. 3, 545–609.
[HK] N. Higson, G. G. Kasparov. E-theory and KK-theory for groups which act properly and
isometrically on Hilbert space. Invent. Math. 144 (2001), no. 1, 23–74.
[HR] N. Higson, J. Roe. Amenable group actions and the Novikov conjecture. J. Reine Angew.
Math. 519 (2000), 143–153.
[J] L. Ji. The integral Novikov conjectures for linear groups containing torsion elements. J.
Topol. 1 (2008), no. 2, 306–316.
[JR] W. B. Johnson, N. L. Randrianarivony. lp (p > 2) does not coarsely embed into a Hilbert
space. Proc. Amer. Math. Soc. 134 (2006), 1045–1050 (electronic).
[Ka] N. Kalton. The nonlinear geometry of Banach spaces. Rev. Mat. Complut. 21 (2008), no.
1, 7–60.
[K] G. G. Kasparov. Equivariant KK-theory and the Novikov conjecture. Invent. Math. 91
(1988), 147–201.
[Ki] Y. Kida. The mapping class group from the viewpoint of measure equivalence theory. Mem.
Amer. Math. Soc. 196 (2008), no. 916, viii+190 pp.
[KY] G. Kasparov, G. Yu. The Novikov conjecture and geometry of Banach spaces. Preprint,
2010.
[1] L V. Lafforgue. K-théorie bivariante pour les algébres de Banach et conjecture de Baum-
Connes. Invent. Math. 149 (2002), no. 1, 1–95.
[Ma] G. Margulis. Discrete subgroups of semisimple Lie groups. Ergebnisse der Mathematik und
ihrer Grenzgebiete (3), 17. Springer-Verlag, Berlin, 1991. x+388 pp.
[MN] M. Manor, A. Naor. Metric cotype. Ann. of Math. (2) 168 (2008), no. 1, 247–298.
11
[MY] I. Mineyev, G. Yu. The Baum-Connes conjecture for hyperbolic groups. Invent. Math. 149
(2002), 97–122.
[Mo] G. D. Mostow. Strong rigidity of locally symmetric spaces. Annals of Mathematics Studies,
No. 78. Princeton University Press, Princeton, N.J.; University of Tokyo Press, Tokyo, 1973.
v+195 pp.
[Nov] S. P. Novikov. Topological invariance of rational Pontrjagin classes. Soviet Math. Dokl. ,
6 : 4 (1965) pp. 921923 Dokl. Akad. Nauk SSSR , 163 (1965) 298300.
[N] P. Nowak. Coarsely embeddable metric spaces without Property A. J. Funct. Anal. 252
(2007), no. 1, 126–136.
[NY1] P. Nowak, G. Yu. What is . . . property A? Notices Amer. Math. Soc. 55 (2008), no. 4,
474–475.
[NY2] P. Nowak, G. Yu. Large scale geometry.
[R1] J. Roe, Coarse cohomology and index theory on complete Riemannian manifolds. Mem.
Amer. Math. Soc. 104 (1993), no.497.
[R2] J. Roe, Index theory, coarse geometry, and topology of manifolds. CBMS Regional Confer-
ence Series in Mathematics, the American Mathematical Society, Providence, RI, 1996.
[R3] J. Roe, Hyperbolic groups have finite asymptotic dimension. Proc. Amer. Math. Soc. 133
(2005), no. 9, 2489–2490.
[R4] J. Roe. Lectures on coarse geometry. University Lecture Series, 31. American Mathematical
Society, Providence, RI, 2003.
[Ro] J. Rosenberg, C ∗ -algebras, positive scalar curvature and the Novikov Conjecture, Publ.
I.H.E.S., No. 58 (1983), 197–212.
[P] M. Puschnigg. The Kadison-Kaplansky conjecture for word-hyperbolic groups. Invent. Math.
149 (2002), no. 1, 153–194.
[S] Z. Sela. Uniform embeddings of hyperbolic groups in Hilbert spaces. Israel J. Math. 80
(1992), no. 1-2, 171–181.
[STY] G. Skandalis, J. L. Tu, G. Yu. The coarse Baum-Connes conjecture and groupoids. Topol-
ogy 41 (2002), no. 4, 807–834.
[T] R. Tessera. Coarse embeddings into a Hilbert space, Haagerup property and Poincaré in-
equalities. J. Topol. Anal. 1 (2009), no. 1, 87–100.
[Tu] J. L. Tu. Remarks on Yu’s “property A” for discrete metric spaces and groups. Bull. Soc.
Math. France 129 (2001), no. 1, 115–139.
[W] R. Willett. Some notes on property A. Limits of graphs in group theory and computer
science, 191–281, EPFL Press, Lausanne, 2009.
[Y1] G. Yu. The Novikov conjecture for groups with finite asymptotic dimension. Ann. of Math.
(2) 147 (1998), no. 2, 325–355.
[Y2] G. Yu. The coarse Baum-Connes conjecture for spaces which admit a uniform embedding
into Hilbert space. Invent. Math. 139 (2000), no. 1, 201–240.
Department of Mathematics, 1326 Stevenson Center, Vanderbilt University, Nashville,

TN 37240, USA
E-mail address: guoliang.yu@vanderbilt.edu
This volume represents the proceedings of the conference on Noncommutative Geometric
Methods in Global Analysis, held in honor of Henri Moscovici, from June 29–July 4,
2009, in Bonn, Germany.
Henri Moscovici has made a number of major contributions to noncommutative geometry,
global analysis, and representation theory. This volume, which includes articles by some
of the leading experts in these fields, provides a panoramic view of the interactions of
noncommutative geometry with a variety of areas of mathematics. It focuses on geometry,
analysis and topology of manifolds and singular spaces, index theory, group representa-
tion theory, connections of noncommutative geometry with number theory and arithmetic
geometry, Hopf algebras and their cyclic cohomology.
CONM/546 AMS on the Web

www.ams.org

Noncommutative Geometry and Global Analysis

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Noncommutative Geometry and Global Analysis

Uploaded by

Copyright:

Available Formats

CONTEMPORARY

American Mathematical Society

American Mathematical Society

Frontispiece photo courtesy of John R. Copeland.

Library of Congress Cataloging-in-Publication Data

Algebras of p-symbols, noncommutative p-residue, and the Brauer group

This volume represents the proceedings of the conference in honor of Henri

1. D. Barbasch and P. Pandžić “Dirac cohomology, unipotent representa-

2. P. Bressler, A. Gorokhovsky, R. Nest and B. Tsygan “Algebraic index the-

15. G. Yu “Large scale geometry and its applications.”

The signature operator on Riemannian pseudomanifolds

The lost Riemann-Roch index problem

C ∗ -algebras associated with integral domains

Hopf-cyclic cohomology and Connes-Moscovici characteristic map

Noncommutative geometry and cosmology: a progress report

A topological index theorem for manifolds with corners

free product groups Γ = Γ1 ∗Σ Γ2 , where Γ1 is Kazhdan and Σ inﬁnite amenable

Noncommutative residues and projections associated to boundary

Fine structure of special symplectic spaces

Operations on Hochschild complexes

The index of projective families of elliptic operators

Aspects of free analysis

Aldana Dominguez, Clara Lucia Himpel, Benjamin

Dirac cohomology and unipotent representations of complex

Dan Barbasch and Pavle Pandžić

This paper is dedicated to Henri Moscovici.

Abstract. This paper studies unitary representations with Dirac cohomology

Conjecture 1.1. Let G be a complex reductive Lie group viewed as a real

2010 Mathematics Subject Classiﬁcation. Primary 22E47 ; Secondary 22E46.

Let Γ ⊂ G be a discrete cocompact subgroup. A question of interest is the compu-

A more general problem where the trivial representation C of Γ is replaced by

A major role in providing an answer to the above problem is played by the

In this formula, due to Parthasarathy [P1],

H(g, K; π ⊗ F ∗ ) = HomK (HD (F ), HD (π)),

Let B = HN be a Borel subgroup of G. We identify the complexiﬁcation

where the sum is over all K-types E(μ) of π.

∗ 2.4. The restriction of the g-module E(ρ) to m is isomorphic to F (ρm )⊗

Since π is unitary, the computation for its Dirac cohomology is

Note that wm ρ +ρ is a sum of roots in Δ(n), and antidominant for Δm , because

To show that (2.15) holds, it is enough to show that

The parameter of the metaplectic representation is

(2.22) λm2 = (1/2, 5/2, 3/2), −sλm2 = (−1/2, 5/2, 3/2)

(2.23) λm2 = (5/2, 3, 2, 1/2), −sλm2 = (5/2, 3/2, −1/2)

The LHS contains the representation with lowest weight conjugate to

wm ρ + ρ = (2, 4, −5, −3, −1) + (5, 4, 3, 2, 1) = (7, 8, −2, −1, 0).

This is the sum of the following roots in Δ(n) :

This set of roots is stable under the operation of adding negative

3. Inﬁnitesimal character ρ/2

so τ = 0. Then the multiplicity in (2.6) becomes

(3.2) [L(ρ/2, −sρ/2) : E(ρ)].

3.1. Induced from a unitary character, inﬁnitesimal character ρ/2.

Proof. We can write wm ρ = xρ for a unique x ∈ W. By the last statement of

The metaplectic representations will be analyzed in Section 5. In particular,

Table 1. Hermitian parameters at ρ/2 for F4 .

4. The case of GL(n, C)

Langlands parameter given by a minimal principal series of G. The Levi compo-

5. Unipotent representations with Dirac cohomology

where Γ := Gal(C/R) is the Galois group. There is a canonical map WR −→ Γ that

with annihilator in U (g) maximal containing the ideal in Z := U (g)G corresponding

5.4. Type B. Let G = SO(2n + 1, C). We use the standard coordinates.

occurring with multiplicity 1. We sketch a proof. By the results of [AB], π occurs

To be conjugate to τ , this expression must have 2a + 1 components equal to ±1/2.

This should be equal to τ up to W , and this happens precisely when k = n. (Recall

5.6. Type D. Let G = SO(2n, C). We use the usual coordinates.

g [1]-modules and g 2

(4.6) GF : C•Lie g 2 g [1] , h; L• → Ω• (M, L• )

be reduction modulo . This is a map of (g g[1], H)-modules (where the module

H• g 2 g[1], H; C ((u, )) .

gh on the total space of a Riemannian submersion is the pullback