Grundlehfen def

mathematischen Wissenschaften 290

A Series of Comprehensive Studies in Mathematics

Editors
M. Artin S.S. Chern J.M. Frohlich E. Heinz H. Hironaka
F. Hirzebruch L. Hormander S. Mac Lane C.C. Moore
J.K. Moser M. Nagata W. Schmidt D.S. Scott Ya.G. Sinai
J. Tits B.L. van der Waerden M. Waldschmidt S. Watanabe

Managing Editors
M. Berger B. Eckmann S.R.S. Varadhan
Grundlehren der mathematischen Wissenschaften
A Series of Comprehensive Studies in Mathematics
A Selection

200. Dold: Lectures on Algebraic Topology

20 I . Beck: Continuous Flows in the Plane
202. Schmetterer: Introduction to Mathematical Statistics
203. Schoeneberg: Elliptic Modular Functions
204. Popov: Hyperstability of Control Systems
205. NikoI'skii: Approximation of Functions of Several Variables and Imbedding
Theorems
206. Andre: Homologie des Algebres Commutatives
207. Donoghue: Monotone Matrix Functions and Analytic Continuation
208. Lacey: The Isometric Theory of Classical Banach Spaces
209. Ringel: Map Color Theorem
210. Gihman/Skorohod: The Theory of Stochastic Processes I
211. ComfortiNegrepontis: The Theory of Ultrafilters
212. Switzer: Algebraic Topology~Homotopy and Homology
213. Shafarevich: Basic Algebraic Geometry
214. van der Waerden: Group Theory and Quantum Mechanics
215. Schaefer: Banach Lattices and Positive Operators
216. P6lyaiSzego: Problems and Theorems in Analysis II
217. Stenstrom: Rings of Quotients
218. Gihman/Skorohod: The Theory of Stochastic Process II
219. DuvantiLions: Inequalities in Mechanics and Physics
220. Kirillov: Elements of the Theory of Representations
221. Mumford: Algebraic Geometry I: Complex Projective Varieties
222. Lang: Introduction to Modular Forms
223. Bergh/Liifstrom: Interpolation Spaces. An Introduction
224. Gilbarg/Trudinger: Elliptic Partial Differential Equations of Second Order
225. Schiitte: Proof Theory
226. Karoubi: K-Theory, An Introduction
227. Grauert/Remmert: Theorie der Steinschen Riiume
228. Segal/Kunze: Integrals and Operators
229. Hasse: Number Theory
230. Klingenberg: Lectures on Closed Geodesics
231. Lang: Elliptic Curves: Diophantine Analysis
232. Gihman/Skorohod: The Theory of Stochastic Processes III
233. StroocklVaradhan: Multi-dimensional Diffusion Processes
234. Aigner: Combinatorial Theory
235. Dynkin/Yushkevich: Markov Control Processes and Their Applications
236. Grauert/Remmert: Theory of Stein Spaces
237. Kothe: Topological Vector-Spaces II
238. Graham/McGehee: b,says in Commutative Harmonic Analysis
239. Elliott: Probabilistic Number Theory I
240. Elliott: Probabilistic Number Theory II
241. Rudin: Function Theory in the Unit Ball of C"
242. Huppert/Blackburn: Finite Groups I
243. Huppert/Blackburn: Finite Groups II
244. Kubert/Lang: Modular Units
245. Cornfeld/Fomin/Sinai: Ergodic Theory
246. NaimarkiStern: Theory of Group Representations
247. Suzuki: Group Theory I
('ontinued q/fer /11ljt.\
J.H. Conway N.J.A. Sloane

Sphere Packings,
Lattices and Groups
With Additional Contributions by
E. Bannai, J. Leech, S.P. Norton,
A.M. Odlyzko, R.A. Parker, L. Queen
and B.B. Venkov

With 112 Illustrations

Springer Science+Business Media, LLC

J.H. Conway
Mathematics Department
Princeton University
Princeton, New Jersey 08540, USA

N.J.A. Sloane
Mathematical Science Department
AT&T Bell Laboratories
Murray Hill, New Jersey 07974, USA

AMS Classifications: 05B40, IIH06, 20E32. 11T7!, I1EI2

Library of Congress Cataloging-in-Publication Data

Conway. John Horton.
Sphere packings, lattices and groups.
(Grundlehren der mathematischen Wissenschaften; 290)
Bibliography: p.
Includes index.
I. Combinatorial packing and covering. 2. Sphere.
3. Lattice theory. 4. Finite groups. I. Sloane,
N.J .A. (Neil James Alexander) II. Title.
III. Series.
QAI66.7.C66 1985 511'.6 S7-23303

© 1988 by Springer Science+Business Media New York

Originally published by Springer-Verlag New York, Inc. in 1988
Softcover reprint of the hardcover 1st edition 1988
All rights reserved. This work may not be translated or copied in whole or in part
without the written permission of the publisher Springer Science+Business Media, LLC.
except for brief excerpts in connection with reviews
or scholarly analysis. Use in connection with any form of information storage and
retrieval, electronic adaptation. computer software, or by similar or dissimilar
methodology now known or hereafter developed is forbidden.
The use of general descriptive names, trade names, trademarks, etc. in this pUblication.
even if the former are not especially identified, is not to be taken as a sign that such
names, as understood by the Trade Marks and Merchandise Marks Act, may accordingly
be used freely by anyone.
Permission to photocopy for internal or personal use, or the internal or personal usc
of specific clients. is granted by Springer Science+Business Media, LLC. for libraries and other
users registered with the Copyright Clearance Center (CCC), provided that the base
fee of $0.00 per copy, plus $0.20 per page is paid directly to CCC, 21 Congress Street.
Salem, MA 01970, USA. Special requests should be addressed directly to
Springer Science+Business Media, LLC.

Text prepared by the authors using TROFF.

9 S7 6 54 3 2 I

ISBN 978-1-4757-2018-1 ISBN 978-1-4757-2016-7 (eBook)

DOI 10.1007/978-1-4757-2016-7
 Preface

The main themes. This book is mainly concerned with the problem of
packing spheres in Euclidean space of dimensions 1,2,3,4,5, .... Given a
large number of equal spheres, what is the most efficient (or densest) way
to pack them together? We also study several closely related problems: the
kissing number problem, which asks how many spheres can be arranged so
that they all touch one central sphere of the same size; the covering
problem, which asks for the least dense way to cover n-dimensional space
with equal overlapping spheres; and the quantizing problem, important for
applications to analog-to-digital conversion (or data compression), which
asks how to place points in space so that the average second moment of
their Voronoi cells is as small as possible. Attacks on these problems
usually arrange the spheres so their centers form a lattice. Lattices are
described by quadratic forms, and we study the classification of quadratic
forms. Most of the book is devoted to these five problems.
The miraculous enters: the E 8 and Leech lattices. When we investigate
those problems, some fantastic things happen! There are two sphere
packings, one in eight dimensions, the E 8 lattice, and one in twenty-four
dimensions, the Leech lattice A24 , which are unexpectedly good and very
symmetrical packings, and have a number of remarkable and mysterious
properties, not all of which are completely understood even today. In a
certain sense we could say that the book is devoted to studying these two
lattices and their properties.
At one point while working on this book we even considered adopting a
special abbreviation for "It is a remarkable fact that", since this phrase
seemed to occur so often. But in fact we have tried to avoid such phrases
and to maintain a scholarly decorum of language.
Nevertheless there are a number of astonishing results in the book, and
perhaps this is a good place to mention some of the most miraculous. (The
technical terms used here are all defined later in the book.)
vi Preface

- The occurrence of the Leech lattice as the unique laminated lattice in

24 dimensions (Fig. 6.1).
- Gleason's theorem describing the weight enumerators of doubly-even
self-dual codes, and Hecke's theorem describing the theta series of even
unimodular lattices (Theorems 16 and 17 of Chap. 7).
- The one-to-one correspondence between the 23 deep holes in the Leech
lattice and the even unimodular 24-dimensional lattices of minimal norm 2
(Chapters 16,23,26).
- The construction of the Leech lattice in 11 25 ,1 as wJ../w , where
w = (0,1,2,3 .. " .24170)

is a vector of zero norm (Theorem 3 of Chap. 26). (We remind the reader
that in Lorentzian space R 25 ,1, which is 26-dimensional space equipped
with the norm x'x=xr+'" +Xi5-xi6, there is a unique even
unimodular lattice 1125 •1 - see §1 of Chap. 26.)
- The occurrence of the Leech lattice as the Coxeter diagram of the
reflections in the automorphism group of 11 25 •1 (Chap. 27).
Some further themes. Besides the five problems that we mentioned in the
first paragraph, there are some other topics that are always in our minds:
error-correcting codes. Steiner systems and t-designs, and the theory of
finite groups. The E8 and Leech lattices are intimately involved with these
topics, and we investigate these connections in considerable detail. Many
of the sporadic simple groups are encountered, and we devote a whole
chapter to the Monster or Friendly Giant simple group, whose construction
makes heavy use of the Leech lattice.
The main applications. There are many connections between the
geometrical problems mentioned in the first paragraph and other areas of
mathematics, chiefly with number theory (especially quadratic forms and
the geometry of numbers).
The main application outside mathematics is to the channel coding
problem, the design of signals for data transmission and storage, i.e. the
design of codes for a band-limited channel with white Gaussian noise. It
has been known since the work of Nyquist and Shannon that the design of
optimal codes for such a channel is equivalent to the sphere packing
problem. Theoretical investigations into the information-carrying capacity
of these channels requires knowledge of the best sphere packings in large
numbers of dimensions. On the other hand practical signalling systems
(notably the recently developed Trellis Coded Modulation schemes) have
been designed using properties of sphere packings in low dimensions, and
certain modems now on the market use codes consisting of points taken
from the E 8 lattice!
There are beautiful applications of these lattices to the numerical
evaluation of n-dimensional integrals (see § 4.2 of Chap. 3).
Of course there are connections with chemistry, since crystallographers
have studied three-dimensional latti.::es since the beginning of the subject.
We sometimes think of this book as being a kind of higher-dimensional
Preface VII

analog of Wells' "Structural Inorganic Chemistry" [Wel4]. Recent work

on quasi-crystals has made use of six- and eight-dimensional lattices.
Higher-dimensional lattices are of current interest in physics. Recent
developments in dual theory and superstring theory have involved the E8
and Leech lattices, and there are also connections with the Monster group.
Other applications and references for these topics will be found in Chap. 1.
Who should buy this book. Anyone interested in sphere packings, in
lattices in n-dimensional space, or in the Leech lattice. Mathematicians
interested in finite groups, quadratic forms, the geometry of numbers, or
combinatorics. Engineers who wish to construct n-dimensional codes for a
band-limited channel, or to design n-dimensional vector quantizers.
Chemists and physicists interested in n-dimensional crystallography.
Prerequisites. In the early chapters we have tried to address (perhaps not
always successfully) any well-educated undergraduate. The technical level
fluctuates from chapter to chapter, and generally increases towards the end
of the book.
What's new here. A number of things in this book have not appeared in
print before. We mention just a few.
- Tables of the densest sphere packings known in dimensions up to one
million (Tables 1.2, 1.3).
- Tables and graphs of the best coverings and quantizers known in up to
24 dimensions (Chap. 2).
- A table of the best coding gain of lattices in up to 128 dimensions
(Table 3.2>-
- Explicit arrangements of n-dimensional spheres with kissing numbers
that grow like 20.003n (Eq. (56) of Chap. 1), the best currently known.
- Explicit constructions of good spherical codes (§2.5 of Chap. 1).
- Explicit constructions of n-dimensional lattice coverings with density
that grows like 20.084n (Eq. (13) of Chap. 2), again the best known.
- New formulae for the theta series of many lattices (e.g. for the Leech
lattice with respect to an octahedral deep hole, Eq. (144) of Chap. 4).
- Tables giving the numbers of points in successive shells of these lattices
(e.g. the first 50 shells of the Leech lattice, Table 4.14>-
- A new general construction for sphere packings (Construction AI' §3 of
Chap. 8).
- New descriptions of the lattice packings recently discovered by McKay,
Quebbemann, Craig, etc. (Chap. 8).
- Borcherds' classification of the 24-dimensional odd unimodular lattices
and his neighborhood graph of the 24-dimensional even unimodular lattices
(Chap. 24).
- A list of the 284 types of shallow hole in the Leech lattice (Chap. 25).
- Some easy ways of computing with rational and integral quadradic
forms, including elementary systems of invariants for such forms, and a
handy notation for the genus of a form (Chap. 15).
Tables of binary quadratic forms with -1 00 ~ det ~ 50,
indecomposable ternary forms with Idetl ~ 50, genera of forms with
Idetl ~ 11, genera of p-elementary forms for all p, positive definite forms
VIII Preface

with determinant 2 up to dimension 18 and determinant 3 up to dimension

17 (Chap. 15).
- A simple description of a construction for the Monster simple group
{Chap. 29).
Other tables, which up to now could only be found in journal articles or
conference proceedings, include:
- Bounds for kissing numbers in dimensions up to 24 (Table 1.5).
- The Minkowski-Siegel mass constants for even and odd unimodular
lattices in dimensions up to 32 (Chap. 16).
- The even and odd unimodular lattices in dimensions up to 24 (Table 2.2
and Chaps. 16, 17).
- Vectors in the first eight shells of the E 8 lattice (Table 4.10) and the
first three shells of the Leech lattice (Table 4.13).
- Best's codes of length 10 and 11, that produce the densest packing
known (P lOc ) in 10 dimensions and the highest kissing number known
(Pll) in 11 dimensions (Chap. 5).
- Improved tables giving the best known codes of length 2m for m ",; 8
(Table SA) and of all lengths up to 24 {Table 9.0.
- Laminated lattices in dimensions up to 48 (Tables 6.1, 6.3).
- The best integral lattices of minimal norms 2, 3 and 4, in dimensions up
to 24 (Table 604).
- A description of E 8 lattice vectors in terms of icosians {Table 8.0.
- Minimal vectors in McKay's 40-dimensional lattice M 40 (Table 8.6).
- The classification of subsets of 24 objects under the action of the
Mathieu group M24 {Fig. 10.0.
- Groups associated with the Leech lattice (Table lOA).
- Simple groups that arise from centralizers in the Monster (Chap. 10).
- Second moments of polyhedra in 3 and 4 dimensions (Chap. 21).
- The deep holes in the Leech lattice (Table 23.1).
- An extensive table of Leech roots, in both hyperbolic and Euclidean
coordinates (Chap. 28).
- Coxeter-Vinberg diagrams for the automorphism groups of the lattices
In,l for n ",; 20 (Chap. 28) and IIn,l for n ",; 24 (Chap. 27).
The contents of the chapters. Chapters 1-3 form an extended introduction
to the whole book. In these chapters we survey what is presently known
about the packing, kissing number, covering and quantizing problems.
There are sections on quadratic forms and their classification, the
connections with number theory, the channel coding problem, spherical
codes, error-correcting codes, Steiner systems, t-designs, and the
connections with group theory. These chapters also introduce definitions
and terminology that will be used throughout the book.
Chapter 4 describes a number of important lattices, including the cubic
lattice zn, the root lattices An' Dn , E 6 , E 7 , E 8 , the Coxeter-Todd lattice
K 12 , the Barnes-Wall lattice A l6 , the Leech lattice A 24 , and their duals.
Among other things we give their minimal vectors, densities, covering radii,
glue vectors, automorphism groups, expressions for their theta series, and
tables of the numbers of points in the first fifty shells. We also include a
Preface ix

brief discussion of reflection groups and of the technique of gluing lattices

together.
Chapters 5-8 are devoted to techniques for constructing sphere
packings. Many of the constructions in Chaps. 5 and 7 are based on
error-correcting codes; other constructions in Chapter 5 build up packings
by layers. Layered packings are studied in greater detail in Chap. 6,
where the formal concept of a laminated lattice An is introduced. Chapter
8 uses a number of more sophisticated algebraic techniques to construct
lattices.
Chapter 9 introduces analytical methods for finding bounds on the best
codes, sphere packings and related problems. The methods use techniques
from harmonic analysis and linear programming. We give a simplified
account of Kabatiansky and Levenshtein's recent sphere packing bounds.
Chapters 10 and 11 study the Golay codes of length 12 and 24, the
associated Steiner systems S (5,6,12) and S (5, 8, 24), and their
automorphism groups MI2 and M 24 • The MINIMOG and MOG (or
Miracle Octad Generator> and the Tetracode and Hexacode are
computational tools that make it easy to perform calculations with these
objects. These two chapters also study a number of related groups, in
particular the automorphism group Coo (or' 0) of the Leech lattice. The
Appendix to Chapter 10 describes all the sporadic simple groups.
Chapter 12 gives a short proof that the Leech lattice is the unique even
unimodular lattice with no vectors of norm 2. Chapter 13 solves the
kissing number problem in 8 and 24 dimensions - the E 8 and Leech
lattices have the highest possible kissing numbers in these dimensions.
Chapter 14 shows that these arrangements of spheres are essentially
unique.
Chapters 15-19 deal with the classification of integral quadratic forms.
Chapters 16 and 18 together give three proofs that Niemeier's enumeration
of the 24-dimensional even unimodular lattices is correct. In Chap. 19 we
find all the extremal odd unimodular lattices in any dimension.
Chapters 20 and 21 are concerned with geometric properties of lattices.
In Chap. 20 we di~cuss algorithms which, given an arbitrary point of the
space, find the closest lattice point. These algorithms can be used for
vector quantizing or for encoding and decoding lattice codes for a
band limited channel. Chapter 21 studies the Voronoi cells of lattices and
their second moments.
Soon after discovering his lattice, John Leech conjectured that its
covering radius was equal to ..fi times its packing radius, but was unable to
find a proof. In 1980 Simon Norton found an ingenious argument which
shows that the covering radius is no more than 1.452... times the packing
radius (Chap. 22), and shortly afterwards Richard Parker and the authors
managed to prove Leech's conjecture (Chap. 23).
Our method of proof involves finding all the "deep holes" in the Leech
lattice, i.e. all points of 24-dimensional space that are maximally distant
x Preface

from the lattice. We were astonished to discover that there are precisely
23 distinct types of deep hole, and that they are in one-to-one
correspondence with the Niemeier lattices (the 24-dimensional even
unimodular lattices of minimal norm 2) - see Theorem 2 of Chap. 23.
Chapter 23, or the Deep Holes paper, as it is usually called, has turned out
to be extremely fruitful, having stimulated the remaining chapters in the
book, also Chap. 6, and several journal articles.
In Chap. 24 we give 23 constructions for the Leech lattice, one for each
of the deep holes or Niemeier lattices. Two of these are the familiar
constructions based on the Golay codes. In the second half of Chap. 24 we
introduce the hole diagram of a deep hole, which describes the environs of
the hole. Chapter 25 (the Shallow Holes paper) uses the results of
Chap. 23 and 24 to classify all the holes in the Leech lattice.
Considerable light is thrown on these mysteries by the realization that
the Leech lattice and the Niemeier lattices can all be obtained very easily
from a single lattice, namely 11 25 ,1> the unique even unimodular lattice in
Lorentzian space R 25,). For any vector w E R25 ,), let
wl- = Ix EII25 ,):X·w=0}.
Then if w is the special vector
W25 = (0,1,2,3, .... 23,24170),

wl-/w is the Leech lattice, and other choices for w lead to the 23 Niemeier
lattices.
The properties of the Leech lattice are closely related to the geometry
of the lattice 1125,). The automorphism groups of the Lorentzian lattice
In,) for n ~ 19 and lIn,) for n = 1, 9 and 17 were found by Vinberg,
Kaplinskaja and Meyer. Chapter 27 finds the automorphism group of
11 25 ,). This remarkable group has a reflection subgroup with a Coxeter
diagram that is, speaking loosely, isomorphic to the Leech lattice. More
precisely, a set of fundamental roots for 11 25 ,) consists of the vectors
r E 1125 ,) satisfying
r . r = 2, r . W25 = -I ,

and we call these the Leech roots. Chapter 26 shows that there is an
isometry between the set of Leech roots and the points of the Leech lattice.
Then the Coxeter part of the automorphism group of Il 25 ,) is just the
Coxeter group generated by the Leech roots (Theorem I of Chap. 27).
Since 11 25 ,) is a natural quadratic form to study, whose definition
certainly does not mention the Leech lattice, it is surprising that the Leech
lattice essentially determines the automorphism group of the form.
The Leech roots also provide a better understanding of the
automorphism groups of the other lattices In,) and lIn,), as we see in
Chap. 28. This chapter also contains an extensive table of Leech roots.
Chapter 29 describes a construction for the Monster simple group, and the
Preface XI

final chapter describes an infinite-dimensional Lie algebra that is obtained

from the Leech roots, and conjectures that it may be related to the
Monster.
The book concludes with a bibliography of about 1550 items.
The structure of this book. Our original plan was simply for a collection of
reprints. But over the past two years the book has been completely
transformed: many new chapters have been added, and the original
chapters have been extensively rewritten to bring them up to date, to
reduce duplicated material, to adopt a uniform notation and terminology,
and to eliminate errors.
We have however allowed a certain amount of duplication to remain, to
make for easier reading. Because some chapters were written at different
times and by different authors, the reader will occasionally notice
differences in style from chapter to chapter.
The arrangement of the chapters presented us with a difficult problem.
We feel that readers are best served by grouping them in the present
arrangement, even though it means that one or two chapters are not in
strict logical order. The worst flaw is that the higher-dimensional part of
the laminated lattices chapter (Chap. 6) depends on knowledge of the deep
holes in the Leech lattice given in Chap. 23. But the preceding part of
Chapter 6, which includes all the best lattice packings known in small
dimensions, had to appear as early as possible.
Chapters 1-4, 7, 8, 11, 15, 17, 20, 25 and 29 are new. Chapter 5 is
based on J. Leech and N.J.A.S., Canad. J. Math. 23 (1971) (see
reference [LeeIO] in the bibliography for full details); Chap. 6 on J.H.C.
and N.J.A.S., Annals of Math. 116 (1982) [Con32]; Chap. 9 is based on
N.J.A.S., Contemp. Math. 9 (1982), published by the American
Mathematical Society [Slo13]; Chap. 10 on J.H.C., in Finite Simple
Groups, edited by M. B. Powell and G. Higman, Academic Press, N.Y.
1971 [ConS]; Chap. 12 on J.H.C., Invent. Math. 7 (1969), published by
Springer-Verlag, N.Y. [Con4]; Chap. 13 on A. M. Odlyzko and N.J.A.S.,
J. Combin. Theory A26 (1979), published by Academic Press, N.Y.
[OdIS]; Chap. 14 on E. Bannai and N.J.A.S., Canad. J. Math. 33 (1981)
[Banl3); Chap. 16 on J.H.C. and N.J.A.S., J. Number Theory 15 (1982)
[Con27], and Europ. J. Combin. 3 (1982) [Con34], both published by
Academic Press, N.Y.; Chapter 18 on B. B. Venkov, Trudy Mat. Inst.
Steklov 148 (1978), English translation in Proc. Steklov Inst. Math.,
published by the American Mathematical Society [Venl); Chap. 19 on
J.H.C., A. M. Odlyzko and N.J.A.S., Mathematika, 25 (1978) [ConI9];
Chap. 21 on J.H.C. and N.J.A.S., IEEE Trans. Information Theory, 28
(1982) [Con28]; Chap. 22 on S. P. Norton, Proc. Royal Society London,
A380 (1982) [Nor4]; Chap. 23 on J.H.C., R. A. Parker and N.J.A.S.,
Proc. Royal Society London, A380 (1982) [Con20]; Chap. 24 on J.H.C.
and N.J.A.S., Proc. Royal Society London, A381 (1982) [Con30];
Chap. 26 on J.H.C. and N.J.A.S., Bulletin American Mathematical
Society, 6 (1982) [Con311; Chap. 27 on J.H.C., J. Algebra, 80 (1983),
XII Preface

published by Academic Press, N.Y. [ConI3]; Chap. 28 on J.H.C. and

N.J.A.S., Proc. Royal Society London, A384 (1982) [Con33]; Chap. 30 on
R. E. Borcherds, J.H.C., L. Queen and N.J.A.S., Advances in Math. 53
(1984), published by Academic Press, N.Y. [Bor5].
Our collaborators mentioned above are:
Eiichi Bannai, Math. Dept., Ohio State University, Columbus, Ohio 43210;
Richard E. Borcherds, Dept. of Pure Math. and Math. Statistics,
Cambridge University, Cambridge CB2 ISB, England;
John Leech, Computing Science Dept., University of Stirling, Stirling FK9
4LA, Scotland;
Simon P. Norton, Dept. of Pure Math. and Math. Statistics, Cambridge
University, Cambridge CB2 ISB, England;
Andrew M. Odlyzko, Math. Sciences Research Center, AT&T Bell
Laboratories, Murray Hill, New Jersey 07974;
Richard A. Parker, Dept. of Pure Math. and Math. Statistics, Cambridge
University, Cambridge CB2 ISB, England;
Larissa Queen, Dept. of Pure Math. and Math. Statistics, Cambridge
University, Cambridge CB2 ISB, England;
B. B. Venkov, Leningrad Division of the Math. Institute of the USSR
Academy of Sciences, Leningrad, USSR.
Acknowledgements. We thank all our collaborators and the publishers of
these articles for allowing us to make use of this material.
We should like to express our thanks to E. S. Barnes, H. S. M.
Coxeter, Susanna Cuyler, G. D. Forney, Jr., W. M. Kantor, J. J. Seidel,
J.-P. Serre, P. N. de Souza, and above all John Leech, for their comments
on the manuscript. Further acknowledgements appear at the end of the
individual chapters. Any errors that remain are our own responsibility:
please notify N. J. A. Sloane, Mathematical Sciences Research Center,
Bell Laboratories, Murray Hill, New Jersey 07974, USA. We would also
like to hear of any improvements to the tables.
We thank Ann Marie McGowan, Gisele Wallace and Cynthia Martin,
who typed the original versions of many of the chapters, and especially
Mary Flannelly and Susan Tarczynski, who produced the final manuscript.
B. L. English and R. A. Matula of the Bell Laboratories library staff
helped locate obscure references.
N.J.A.S. thanks Bell Laboratories (and especially R. L. Graham and
A. M. Odlyzko) for support and encouragement during this work, and
J.H.C. thanks Bell Laboratories for support and hospitality during various
visits to Murray Hill.
Preface xiii

We remark that in two dimensions the familiar hexagonal lattice

o o o

o o o o

o o o o o

o o o o

o o o

solves the packing, kissing, covering and quantizing problems. In a sense

this whole book is simply a search for similar nice patterns in higher
dimensions.
XIV Preface

List of Symbols

page
an ordinary Coxeter-Dynkin diagram 481
A Gram matrix of lattice 5
An n-dimensional zero-sum root lattice 108
A(m) Craig's lattice 223
n
An alternating group of degree n 82
A.B, AB, A :B, A'B extensions of groups 92
Aut (A) automorphism group of lattice 90
Bn lattice constructed from BCH codes 235
BWn n-dimensional Barnes-Wall lattice 234
en cyclic group of order n 109
~4 [4, 2, 31 tetracode over F3 81.320
~6 [6, 3, 41 hexacode over F4 82. 300
~12 [12,6,61 Golay code over F3 85. 321
~24 or ~ [24, 12,81 Golay code over F2 84. 303
c.w.e. complete weight enumerator 78
d minimal distance of code (usually) 76
dn ordinary Coxeter-Dynkin diagram 481
Dn n-dimensional checkerboard root lattice 117
det square of volume of fundamental region 5
e6, e7, e8 ordinary Coxeter-Dynkin diagrams 481
E 6, E 7, E8 Gosset's root lattices 120
8 Eisenstein integers 52
Fq Galois field of order q 75
<!} Gaussian integers 52
go, gJ, g2 orders of glue groups 100
Go, GJ, G 2 glue groups 100
G(p) second moment of polytope 34
G(A) second moment of Voronoi cell of lattice 35
;Yf Hurwitzian integral quaternions 53
HiS Higman-Sims group 292
HJ or J 2 Hall-lanko group 293
I identity matrix 77

In The Euclidean integer lattice (often called Zn);

also the genus to which this belongs 42
Preface xv

lIn the genus of even unimodular Euclidean

lattices in n dimensions 48
In,b lIn ,1 odd and even unimodular Lorentzian
lattices, or the genera to which they belong 385
J b J 2,··· lanka groups 273, 298
K12 12-dimensional Coxeter-Todd lattice 127
M generator matrix for code or lattice; 4,77
also the Monster simple group 297, 554
M 40 McKay's 40-dimensional lattice 221
n dimension of lattice or length of code 3,75
N(x) = x'x norm of vector 41
N m or N(m) number of vectors of norm m 102
0 23 shorter Leech lattice 179
0 24 odd Leech lattice 179
P48p' P 48q extremal 48-dimensional lattices 195
q = e1<iz variables for theta series 45
Q the rational numbers 366
Q32, Q64 Quebbemann's lattices 220,213
R covering radius 33
R the real numbers 3
Rn n -dimensional Euclidean space 3
Rn ,1 Lorentzian space 522
Sn symmetric group of degree n 82
S{t,k,v) Steiner system 88
Suz Suzuki group 293
V4 Klein 4-group (rare) 118
Vn volume of n-dimensional unit sphere 9
V(u) Voronoi cell around u 33
We weight enumerator of code 78
Z the integers 8
Zn n-dimensional cubic lattice 106
'Yn Hermite's constant 20
0= 6./Vn center density 13
6. density of packing 6
6. 24 generating function for Ramanujan numbers 51
1] (A) lattice obtained from Construction E 238
(J = 0/Vn normalized thickness 32
XVI Preface

()=w-w=.g an Eisenstein integer 55

()J,()2,()3,()4 Jacobi theta functions 102
0 thickness of covering 31
0A theta series of lattice 45
An n -dimensional laminated lattice 163
AI6 16-dimensional Barnes-Wall lattice 129
A24 24-dimensional Leech lattice 131, 286
A(C) packing obtained from Construction A I~Q

/.L minimal norm of lattice 42

p packing radius of lattice 10
rr k (n) sum of k-th powers of divisors of n 51
T kissing number 21
T(m) Ramanujan number 51
1>J, 1>2 theta series of hexagonal lattice 1m
-.Jik theta series of Z + k- 1 1m
w a cube root of unity 53
-0 = Coo automorphism group of Leech lattice 287
-I = CoJ, -2 = Co 2 ,--- simple groups 290
-00 = Co= automorphism group of Leech
lattice including translations 290
om, -lmn, subgroups of -0 290
A* dual lattice 10
[n, k, d] parameters of a code 76
[x] integer part of x 154
[0], [1], - - - glue vectors 100
 Contents

Preface............................................................ v

Chapter 1
Sphere Packings and Kissing Numbers
J.H. Conway and N.J.A. Sloane .............................. .
I. The Sphere Packing Problem ............................... I
1.1 Packing Ball Bearings ...................................... I
1.2 Lattice Packings ............................................ 3
1.3 Nonlattice Packings ........................................ 7
1.4 n-Dimensional Packings .................................... 8
1.5 Sphere Packing Problem-Summary of Results............. 12
2. The Kissing Number Problem .............................. 21
2.1 The Problem of the Thirteen Spheres ....................... 21
2.2 Kissing Numbers in Other Dimensions ..................... 21
2.3 Spherical Codes ............................................ 24
2.4 The Construction of Spherical Codes from Sphere
Pac kings .................................................... 26
2.5 The Construction of Spherical Codes from Binary Codes ... 26
2.6 Bounds on A(n,lj» .......................................... 27
Appendix: Planetary Perturbations. . . . . . . . . . . . . . . . . . . . . . . . . . 29

Chapter 2
Coverings, Lattices and Quantizers
J.H. Conway and N.J.A. Sloane ............................... 31
I. The Covering Problem ...................................... 31
1.1 Covering Space with Overlapping Spheres .................. 31
1.2 The Covering Radius and the Voronoi Cells ................ 33
1.3 Covering Problem-Summary of Results ................... 36
1.4 Computational Difficulties in Packings and Coverings ....... 40
xviii Contents

2. Lattices, Quadratic Forms and Number Theory ............ 41

2.1 The Norm of a Vector...................................... 41
2.2 Quadratic Forms Associated with a Lattice................. 42
2.3 Theta Series and Connections with Number Theory ........ 44
2.4 Integral Lattices and Quadratic Forms. . . . . . .. . . . . . . . . . . . . . . 47
2.5 Modular Forms............................................. 50
2.6 Complex and Quaternionic Lattices ........................ 52
3. Quantizers . .... .. .. .. . . . .. .. .. ... .. .. .. .. . .. ... .. . . .. .. . . . . . 56
3.1 Quantization, Analog-to-Digital Conversion and Data
Compression ............................................... 56
3.2 The Quantizer Problem. .. .. .. .. .. .. .. . . . . .. .. .. . . . . .. .. .. . . 59
3.3 Quantizer Problem-Summary of Results. . . . . . . . . . . . . . . . . . . 59

Chapter 3
Codes, Designs and Groups
J.R. Conway and N.J.A. Sloane 63
I. The Channel Coding Problem .............................. . 63
1.1 The Sampling Theorem .................................... . 63
1.2 Shannon's Theorem ....................................... . 66
1.3 Error Probability .......................................... . 69
1.4 Lattice Codes for the Gaussian Channel ................... . 71
2. Error-Correcting Codes .................................... . 75
2.1 The Error-Correcting Code Problem ....................... . 75
2.2 Further Definitions from Coding Theory ................... . 77
2.3 Repetition, Even Weight and Other Simple Codes ......... . 79
2.4 Cyclic Codes .............................................. . 79
2.5 BCH and Reed-Solomon Codes ............................ . 81
2.6 Justesen Codes ............................................ . 82
2.7 Reed-Muller Codes ........................................ . 83
2.8 Quadratic Residue Codes .................................. . 84
2.9 Perfect Codes ............................................. . 85
2.10 The Pless Double Circulant Codes ......................... . 86
2.11 Goppa Codes and Codes from Algebraic Curves ........... . 87
2.12 Nonlinear Codes ........................................... . 87
2.13 Hadamard Matrices ........................................ . 87
3. t-Designs, Steiner Systems and Spherical t-Designs ........ . 88
3.1 t-Designs and Steiner Systems ............................. . 88
3.2 Spherical I-Designs ........................................ . 89
4. The Connections with Group Theory ...................... . 90
4.1 The Automorphism Group of a Lattice .................... . 90
4.2 Constructing Lattices and Codes from Groups ............. . 92

Chapter 4
Certain Important Lattices and Their Properties
J.R. Conway and N.J.A. Sloane ............................... 94
I. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
2. Reflection Groups and Root Lattices ....................... 95
3. Gluing Theory.............................................. 99
Contents XIX

4. Notation; Theta Functions.................................. 101

4.1 Jacobi Theta Functions ..................................... 102
5. The n-Dimensional Cubic Lattice Z" ........................ 106
6. The n-Dimensional Lattices An and A! ..................... 108
6.1 The Lattice An .............................................. 108
6.2 The Hexagonal Lattice ..................................... 110
6.3 The Face-Centered Cubic Lattice ......... " ., .......... " .. 112
6.4 The Tetrahedral or Diamond Packing....................... 113
6.5 The Hexagonal Close-Packing .............................. 113
6.6 The Dual Lattice A! ........................................ 115
6.7 The Body-Centered Cubic Lattice .......................... 116
7. The n-Dimensional Lattices D nand D! ...................... 117
7.1 The Lattice Dn ............................................. 117
7.2 The Four-Dimensional Lattice D4 ........................... 118
7.3 The Packing D: ............................................ 119
7.4 The Dual Lattice D! ........................................ 120
8. The Lattices E•• E7 and Ex .................................. 120
8.1 The 8-Dimensional Lattice E. ............................... 120
8.2 The 7-Dimensional Lattices E7 and E~ ...................... 124
8.3 The 6-Dimensional Lattices E.and E! ...................... 125
9. The 12-Dimensional Coxeter-Todd Lattice KI2 .............. 127
10. The 16-Dimensional Barnes-Wall Lattice AI6 ••••••••.••••••• 129
11. The 24-Dimensional Leech Lattice A24 . . . . . . . . . . . . . . . . . . . . . . 131

Chapter 5
Sphere Packing and Error-Correcting Codes
J. Leech and N.J.A. Sloane .................................... 136
1. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
1.1 The Coordinate Array of a Point.. .. . .. .. .. .. .. .. .. .. .. .. . .. 137
2. Construction A ............................................. 137
2.1 The Construction ........................................... 137
2.2 Center Density ............................................. 137
2.3 Kissing Numbers ........................................... 138
2.4 Dimensions 3 to 6 .......................................... 138
2.5 Dimensions 7 and 8 ......................................... 138
2.6 Dimensions 9 to 12 ......................................... 139
2.7 Comparison of Lattice and Nonlattice Packings ............. 140
3. Construction B ............................................. 141
3.1 The Construction ........................................... 141
3.2 Center Density and Kissing Numbers....................... 141
3.3 Dimensions 8, 9 and 12 ..................................... 142
3.4 Dimensions 15 to 24 ........................................ 142
4. Pac kings Built Up by Layers ............................... 142
4.1 Packing by Layers. .. .. ..... ....... .. .. ... .. ... .. .. ... .. .. . . 142
4.2 Dimensions 4 to 7 .......................................... 144
4.3 Dimensions 11 and 13 to 15 ................................. 144
4.4 Density Doubling and the Leech Lattice A24 •••••••••••••••• 145
4.5 Cross Sections of A24 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
5. Other Constructions from Codes............................ 146
xx Contents

5.1 A Code of Length 40 ....................................... 146

5.2 A Lattice Packing in R40 .................................... 147
5.3 Cross Sections of A40 . . . . . . • . . . . . . . . • . . . . . . . . . . . . . . . . . . . . . . . 148
5.4 Packings Based on Ternary Codes.......................... 148
5.5 Packings Obtained from the Pless Codes............... ..... 148
5.6 Packings Obtained from Quadratic Residue Codes.......... 149
5.7 Density Doubling in R24 and R 48 • • . . • . • • . • • • • . • • • • . . . . . . . . . . • 149
6. Construction C ............................................. 150
6.1 The Construction ........................................... 150
6.2 Distance Between Centers .................................. 150
6.3 Center Density ............................................. 150
6.4 Kissing Numbers........................................... 151
6.5 Packings Obtained from Reed-Muller Codes................ 151
6.6 Packings Obtained from BCH and Other Codes. .. .. . . . . . .. . 152
6.7 Density of BCH Packings ................................... 153
6.8 Packings Obtained from Justesen Codes .................... 155

Chapter 6
Laminated Lattices
J.B. Conway and N.J.A. Sloane 157
1. Introduction.. .. .. .. ... . .. .. .. .. .. .. . .. . .. .. .. . . . . .. .. .. .. . . 157
2. The Main Results........................................... 163
3. Properties of A" to A. ....................................... 168
4. Dimensions 9 to 16 ......................................... 170
5. The Deep Holes in AI6 ...................................... 174
6. Dimensions 17 to 24 ........................................ 176
7. Dimensions 25 to 48 ........................................ 177
Appendix: The Best Integral Lattices Known ............... 179

Chapter 7
Further Connections Between Codes and Lattices
N.J.A. Sloane.................................................... 181
I. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
2. Construction A ............................................. 182
3. Self-Dual (or Type I) Codes and Lattices ................... 185
4. Extremal Type I Codes and Lattices........................ 189
5. Construction B ............................................. 191
6. Type II Codes and Lattices................................. 191
7. Extremal Type II Codes and Lattices ....................... 193
8. Constructions A and B for Complex Lattices ............... 197
9. Self-Dual Nonbinary Codes and Complex Lattices..... .. ... 202
10. Extremal Nonbinary Codes and Complex Lattices .......... 205

Chapter 8
Algebraic Constructions for Lattices
J.B. Conway and N.J.A. Sloane ............................... 206
1. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
2. The Icosians and the Leech Lattice. . . . . . . . . . . . . . . . . . . . . . . . . 207
Contents xxi

2.1 The Icosian Group......... .. ..... .. ............ ....... .. .. . 207

2.2 The Icosian and Turyn-Type Constructions for the Leech
Lattice ..................................................... 210
3. A General Setting for Construction A, and Quebbemann's
64-Dimensional Lattice ..................................... 211
4. Lattices Over Z[e"i/4] , and Quebbemann's 32-Dimensional
Lattice ..................................................... 215
5. McKay's 40-Dimensional Extremal Lattice ................. 221
6. Repeated Differences and Craig's Lattices.................. 222
7. Lattices from Algebraic Number Theory.................... 224
7.1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
7.2 Lattices from the Trace Norm.............................. 224
7.3 Examples from Cyclotomic Fields .......................... 227
7.4 Lattices from Class Field Towers........................... 227
7.5 Unimodular Lattices with an Automorphism of Prime
Order....................................................... 229
8. Constructions D and D' .................................... 232
8.1 Construction D ............................................. 232
8.2 Examples................................................... 233
8.3 Construction D' ............................................ 235
9. Construction E ............................................. 236
10. Examples of Construction E ................................ 238

Chapter 9
Bounds for Codes and Sphere Pac kings
N.l.A. Sloane .................................................... 245
I. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
2. Zonal Spherical Functions .................................. 249
2.1 The 2-Point-Homogeneous Spaces .......................... 250
2.2 Representations of G ....................................... 252
2.3 Zonal Spherical Functions.................................. 253
2.4 Positive-Definite Degenerate Kernels ....................... 256
3. The Linear Programming Bounds ........................... 257
3.1 Codes and Their Distance Distributions. . . . . . . . . . . . . . . . . . . . . 257
3.2 The Linear Programming Bounds ........................... 258
3.3 Bounds for Error-Correcting Codes......................... 260
3.4 Bounds for Constant-Weight Codes......................... 263
3.5 Bounds for Spherical Codes and Sphere Packings .......... 263
4. Other Bounds .............................................. 265

Chapter 10
Three Lectures on Exceptional Groups
l.R. Conway ..................................................... 267
I. First Lecture ............................................... 267
1.1 Some Exceptional Behavior of the Groups Ln(q) ............ 267
1.2 The Case p = 3 ............................................ 269
1.3 The Case p = 5 ............................................ 269
1.4 The Case p = 7 ............................................ 269
xxii Contents

1.5 The Case p = 11 ........................................... 271

1.6 A Presentation for MI2 ...................................... 273
1.7 Janko's Group of Order 175560 ............................. 273
2. Second Lecture....... ...... .......... ...................... 274
2.1 The Mathieu Group M24 .................................... 274
2.2 The Stabilizer of an Octad .................................. 276
2.3 The Structure of the Golay Code ((;24 . . . . . . . . . . . . . . . . . . . . . . . 278
2.4 The Structure of P (!l)/'f". .................................. 278
2.5 The Maximal Subgroups of M24 ............................. 279
2.6 The Structure of P(n) ...................................... 283
3. Third Lecture .............................................. 286
3.1 The Group Coo = ·0 and Some of its Subgroups........ .... 286
3.2 The Geometry of the Leech Lattice ........................ 286
3.3 The Group ·0 and its Subgroup N .......................... 287
3.4 Subgroups of ·0 ............ ................................. 290
3.5 The Higman-Sims and McLaughlin Groups................. 292
3.6 The Group Co, = ·3 ........................................ 293
3.7 Involutions in ·0 ............................................ 294
3.8 Congruences for Theta Series........... .................... 294
3.9 A Connection Between ·0 and Fischer's Group Fi24 ......... 295
Appendix: On the Exceptional Simple Groups .............. 296

Chapter 11
The Golay Codes and the Mathieu Groups
J.H. Conway ..................................................... 299
I. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
2. Definitions of the Hexacode ................................ 300
3. Justification of a Hexacodeword ............................ 302
4. Completing a Hexacodeword ............................... 302
5. The Golay Code 't;24 and the MOG .......................... 303
6. Completing Octads from 5 of their Points ................... 305
7. The Maximal Subgroups of M 2• • • • • • • • • • • • • • • • • • • • • • • • • • • • • • 307
8. The Projective Subgroup L 2(23) ............................. 308
9. The Sextet Group 26 :3'S6 •• •• •• • • • •• •• • • • •• • •• •• • • • • • • • •• • • •• 309
10. The Octad Group 24: AR ...•................................. 311
II. The Triad Group and the Projective Plane of Order 4 ....... 314
12. The Trio Group 26: (S, x L/7)) ............................. 316
13. The Octern Group .......................................... 318
14. The Mathieu Group M 2 , • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • 319
15. The Group M,,:2 ........................................... 319
16. The Group M'2' the Tetracode and the MINIMOG ......... 320
17. Playing Cards and Other Games ............................ 323
1.8. Further Constructions for M'2 .............................. 327

Chapter 12
A Characterization of the Leech Lattice
J.H. Conway ..................................................... 331
Contents xxiii

Chapter 13
Bounds on Kissing Numbers
A.M. Odlyzko and N.J.A. Sloane .............................. 337
1. A General Upper Bound.................................... 337
2. Numerical Results .......................................... 338

Chapter 14
Uniqueness of Certain Spherical Codes
E. Bannai and N.J.A. Sloane................................... 340
I. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340
2. Uniqueness of the Code of Size 240 in 0 8 • • • • • • • • • • • • • • • • • • 342
3. Uniqueness of the Code of Size 56 in 0 7 • • • • • • • • • • • • • • • • • • • • 344
4. Uniqueness of the Code of Size 196560 in 0 24 • • • • • • • • • • • • • • 345
5. Uniqueness of the Code of Size 4600 in 0 23 ••••••••••••••••• 349

Chapter 15
On the Classification of Integral Quadratic Forms
J.H. Conway and N.J.A. Sloane ............................... 352
1. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
2. Definitions ... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 354
2.1 Quadratic Forms ........................................... 354
2.2 Forms and Lattices; Integral Equivalence .................. 355
3. The Classification of Binary Quadratic Forms .............. 356
3.1 Cycles of Reduced Forms .................................. 356
3.2 Definite Binary Forms...................................... 357
3.3 Indefinite Binary Forms .................................... 359
3.4 Composition of Binary Forms .............................. 364
3.5 Genera and Spinor Genera for Binary Forms ............... 366
4. The p-Adic Numbers ....................................... 366
4.1 The p-Adic Numbers ....................................... 367
4.2 p-Adic Square Classes ...................................... 367
4.3 An Extended Jacobi-Legendre Symbol ..................... 368
4.4 Diagonalization of Quadratic Forms ........................ 369
5. Rational Invariants of Quadratic Forms. . . . . . . . . . . . . . . . . . . . . 370
5.1 Invariants and the Oddity Formula ......................... 370
5.2 Existence of Rational Forms with Prescribed Invariants .... 372
5.3 The Conventional Form of the Hasse-Minkowski
Invariant ................................................... 373
6. The Invariance and Completeness of the Rational
Invariants .................................................. 373
6.1 The p-Adic Invariants for Binary Forms .................... 373
6.2 The p-Adic Invariants for n-Ary Forms ..................... 375
6.3 The Proof of Theorem 7 .................................... 377
7. The Genus and its Invariants ............................... 378
7.1 p-Adic Invariants........................................... 378
7.2 The p-Adic Symbol for a Form ............................. 379
 xxiv Contents

7.3 2-Adic Invariants ........................................... 380

7.4 The 2-Adic Symbol ......................................... 380
7.5 Equivalences Between Jordan Decompositions ............. 381
7.6 A Canonical 2-Adic Symbol ................................ 382
7.7 Existence of Forms with Prescribed Invariants ............. 382
7.8 A Symbol for the Genus.................................... 384
8. Classification of Forms of Small Determinant and of
p-Elementary Forms. ................ .. ...... .......... .. .. . 385
8.1 Forms of Small Determinant. .. .. . . . . . . . . . . . . . . . . . . .. . . . . .. . 385
8.2 p-Elementary Forms. . . . . . . . . .. .. . . . . . . . . . . . . . . . . . . . . . . . . .. . 386
9. The Spinor Genus .......................................... 388
9.1 Introduction. . . . . . . . . . .. . . . . . . . . . . . . . . .. . . . . .. . . . . . . . . . . . . . . 388
9.2 The Spinor Genus .......................................... 389
9.3 Identifying the Spinor Kernel......... ..... . ......... .. .. .. . 390
9.4 Naming the Spinor Operators for the Genus of f .. ....... .. . 390
9.5 Computing the Spinor Kernel from the p-Adic Symbols .... 391
9.6 Tractable and Irrelevant Primes ............................ 392
9.7 When is There Only One Class in the Genus'? .............. 393
10. The Classification of Positive Definite Forms ............... 396
10.1 Minkowski Reduction ...................................... 396
10.2 The Kneser Gluing Method ................................. 399
10.3 Positive Definite Forms of Determinant 2 and 3 ............ 399
II. Computational Complexity ................................. 402

Chapter 16
Enumeration of Unimodular Lattices
i.H. Conway and N.i.A. Sloane ............................... 406
I. The Niemeier Lattices and the Leech Lattice. . . . . . . . . . . . . . . 406
2. The Mass Formulae for Lattices ............................ 408
3. Verifications of Niemeier's List ............................ 410
4. The Enumeration of Unimodular Lattices in Dimensions
n ~ 23 ...................................................... 413

Chapter 17
The 24-Dimensional Odd Unimodular Lattices
R.E. Borcherds .................................................. 421

Chapter 18
Even Unimodular 24-Dimensional Lattices
B.B. Venkov ...................................................... 427
1. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427
2. Possible Configurations of Minimal Vectors ................ 428
3. On Lattices with Root Systems of Maximal Rank........... 431
4. Construction of the Niemeier Lattices ...................... 434
5. A Characterization of the Leech Lattice .................... 437
Contents xxv

Chapter 19
Enumeration of Extremal Self-Dual Lattices
J.H. Conway, A.M. Odlyzko and N.J.A. Sloane.............. 439
I. Dimensions 1-16 ............................................ 439
2. Dimensions 17-47 ........................................... 439
3. Dimensions n ;;: 48 ......................................... 441

Chapter 20
Finding the Closest Lattice Point
J.H. Conway and N.J.A. Sloane ............................... 443
l. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 443
2. The Lattices Z", D" and A" . .. .. .. ... .. .. ..... .. .. ... .. .. . .. . 444
3. Decoding Unions of Cosets ................................. 446
4. "Soft Decision" Decoding for Binary Codes................ 447
5. Decoding Lattices Obtained from Construction A ........... 448
6. Decoding E8 ................................................ 448

Chapter 21
Voronoi Cells of Lattices and Quantization Errors
J.H. Conway and N.J.A. Sloane ............................... 449
I. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 449
2. Second Moments of Polytopes.............................. 451
2.A Dirichlet's Integral. .. .. .. . ... .. .. .. .. .. ... ... .. .. .. .. .. .. . .. 451
2.B Generalized Octahedron or Cross polytope .................. 452
2.C The n-Sphere ............................................... 452
2.D n-Dimensional Simplices.................................... 452
2.E Regular Simplex............................................ 453
2.F Volume and Second Moment of a Polytope in Terms
of its Faces ................................................. 453
2.G Truncated Octahedron ...................................... 454
2.R Second Moment of Regular Polytopes...................... 454
2.1 Regular Polygons ........................................... 455
2.1 Icosahedron and Dodecahedron ............................ 455
2.K The Exceptional 4-Dimensional Polytopes .................. 455
3. Voronoi Cells and the Mean Squared Error of Lattice
Quantizers .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 456
3.A The Voronoi Cell of a Root Lattice......................... 456
3.B Voronoi Cell for An ......................................... 459
3.C Voronoi Cell for D" (n ;;: 4) ................................. 462
3.D Voronoi Cells for E6 ,E7 ,E. .................................. 462
3.E Voronoi Cell for D~ ........................................ 463
3.F Voronoi Cell for A~ ........................................ 472
3.G The Walls of the Voronoi Cell.............................. 474

Chapter 22
A Bound for the Covering Radius of the Leech Lattice
S.P. Norton...................................................... 476
XXVI Contents

Chapter 23
The Covering Radius of the Leech Lattice
J.H. Conway, R.A. Parker and N.l.A. Sloane 478
I. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 478
2. The Coxeter-Dynkin Diagram of a Hole .................... 480
3. Holes Whose Diagram Contains an An Subgraph ............ 484
4. Holes Whose Diagram Contains a Dn Subgraph ............. 495
5. Holes Whose Diagram Contains an En Subgraph ............ 502

Chapter 24
Twenty-Three Constructions for the Leech Lattice
J.H. Conway and N.J.A. Sloane............................... 506
1. The "Holy Constructions" ................................. 506
2. The Environs of a Deep Hole. .. .. ........ .. ....... ......... 510

Chapter 25
The Cellular Structure of the Leech Lattice
R.E. Borcherds, J.H. Conway and L. Queen 513
1. Introduction................................................ 513
2. Names for the Holes ....................................... 513
3. The Volume Formula....................................... 514
4. The Enumeration of the Small Holes 519

Chapter 26
Lorentzian Forms for the Leech Lattice
J.H. Conway and N.1.A. Sloane............................... 522
1. The Unimodular Lorentzian Lattices ...................... . 522
2. Lorentzian Constructions for the Leech Latticc 523

Chapter 27
The Automorphism Group of the 26-Dimensional Even
Unimodular Lorentzian Lattice
J.H. Conway ..................................................... 527
1. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 527
2. The Main Theorem ......................................... 528

Chapter 28
Leech Roots and Vinberg Groups
J.H. Conway and N.l.A. Sloane ............................... 532
1. The Leech Roots ........................................... 532
2. Enumeration of the Leech Roots ........................... 541
3. The Lattices In.Jor n ~ 19 ................................. .547
4. Vinberg's Algorithm and thc Initial Batches
of Fundamental Roots ...................................... .547
5. The Later Batches of Fundamental Roots .................. .5.50
Contents xxvii

Chapter 29
The Monster Group and its 196884-Dimensional Space
l.R. Conway ..................................................... 554
1. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 554
2. The Golay Code C(5 and the Parker Loop!!J> ................. 556
3. The Mathieu Group M 24 ; the Standard Automorphisms
of!!J> ........................................................ 556
4. The Golay Cocode C(5* and the Diagonal Automorphisms ... 556
5. The Group N of Triple Maps ............................... 557
6. The Kernel K and the Homomorphism g ~ II .............. 557
7. The Structures of Various Subgroups of N ................. 557
8. The Leech Lattice A24 and the Group Qx .................... 55R
9. Short Elements.................................... ......... 559
10. The Basic Representations of N, ........................... 559
11. The Dictionary ............................................. 560
12. The Algebra ................................................ 561
13. The Definition of the Monster Group G. and its
Finiteness .................................................. 561
14. Identifying the Monster ......... '" .. .. .. . .. .. ... .. .. .. . .. .. 562
Appendix 1. Computing in?P ............................... 563
Appendix 2. A Construction for?P .......................... 563
Appendix 3. Some Relations in Q, .......................... 564
Appendix 4. Constructing Representations for N, ........... 566
Appendix 5. Building the Group G, ......................... 567

Chapter 30
A Monster Lie Algebra?
R.E. Borcherds. l.R. Conway. L. Queen and
N.l.A. Sloane.................................................... 568

Bibliography ...................................................... 572

Index ............................................................. 641

 1
Sphere Packings and Kissing Numbers
J. H. Conway and N. J. A. Sloane

The first three chapters describe some of the questions that motivated this
book. In this chapter we discuss the problems of packing spheres in
Euclidean space and of packing points on the surface of a sphere. The
kissing number problem is an important special case of the latter, and asks
how many spheres can just touch another sphere of the same size. We
summarize what is known about these topics and also introduce
terminology that will be used throughout the book.

1. The Sphere Packing Problem

1.1 Packing ball bearings. The classical sphere packing problem, still
unsolved even today, is to find out how densely a large number of identical
spheres (ball bearings, (I) for example) can be packed together. To state
this another way, consider a large empty region, such as an aircraft
hangar, and ask what is the greatest number of ball bearings that can be
packed into this region. If instead of ball bearings we try to pack identical
wooden cubes, children's building blocks, for example, the answer becomes
easy. The cubes fit together with no waste space in between, we can fill
essentially one hundred percent of the space (ignoring the small amount of
space that may be left over around the walls and at the ceiling), and so the
number of cubes we can pack is very nearly to the volume of the hangar
divided by the volume of one of the cubes.
But spheres do not fit together so well as cubes, and there is always
some wasted space in between. No matter how cleverly the ball bearings
are arranged, about one quarter of the space will not be used. One
familiar arrangement is that shown in Figs. l.1a and b, in which the
centers of the spheres form the face-centered cubic (or fcc) lattice. (This
lattice is discussed further in Chap. 4, where there is a detailed account of

(!) Or more pedantically, "bearing balls".

2 Chapter 1

the most important packings in low dimensions.) In this packing the

spheres occupy 7r 1M = .7405... of the total space. By arranging the ball
bearings in this way the number that can be packed into the hangar is
about .7405 ... times the volume of the hangar, divided by the volume of
one ball bearing. We therefore say that the face-centered cubic lattice
packing has density .7405 ....

(a)

Figure 1.1. Two views of the most familiar sphere packing, in which the
centers form the face-centered cubic (or fcc) lattice. This is the packing
usually found in fruit stands, in piles of cannon balls on war memorials, or
in crystalline argon. The spheres occupy 0.7405 ... of the space, and each
sphere touches 12 others. (b) Shows only the centers of the spheres (the
open circles). These centers are obtained by taking those points of the
cubic lattice Z3 whose coordinates add up to an even number.
Sphere Packings and Kissing Numbers 3

The classical version of the sphere packing problem now asks: is this
the greatest density that can be attained? Unfortunately this is an
unsolved problem, one of the famous open problems in mathematics. For
many years the best bound known was Rogers' 1958 result (Eq. (39)
below, [Rog2], [Rog7]) that no packing of ball bearings can have density
greater than .7796... . Recently Lindsey [Lin9] has tightened Rogers'
bound, replacing .7796 ... by .7784... . (Muder [Mudl] and Lindsey
[Lin9a] have obtained further small improvements to the upper bound.
See also [Fej9, p. 298].) As Rogers [Rog2] remarks, "many
mathematicians believe, and all physicists know" that the correct answer is
.7405 .... The situation is complicated, however, by the fact that there are
partial packings that are denser than the face-centered cubic lattice over
larger regions of space than one might have supposed [Boell.
The general sphere-packing problem, to which we return in §IA, asks
for the densest packing of equal spheres in n-dimensional space. This
problem has considerable practical importance even for dimensions greater
than 3.
We pause to assure the reader that there is nothing mysterious about
n -dimensional space. A point in real n -dimensional space Rn is simply a
string of n real numbers

A sphere in Rn with center u = (u) • ...• un) and radius p consists of all
the points x = (x J,X2 • . . . • xn) satisfying
(X)-U)2 + (X2- U 2)2 + ... + (xn- u n)2 = p2.

We can describe a sphere packing in Rn just by specifying the centers u

and the radius. Everything is done with coordinates, and there is no need
to draw pictures. There has been a great deal of nonsense written in
science fiction about the mysterious fourth dimension. One should
certainly not think that the fourth dimension represents time. In
mathematics 4-dimensional space just consists of points with four
coordinates instead of three (and similarly for any number of dimensions).
For a concrete model the reader may imagine a telegraph wire over which
numbers are being sent in sets of four: (I.8, 2.9, -1.3, 2.0), (I.l, -0.8,
0.5, 3.1), .... Each set of four numbers is a point in 4-dimensional space.
Later in this chapter we'll do some easy calculations in 4-dimensional
space.
1.2 Lattice packings. The packing shown in Fig. 1.1 is called a lattice
packing because it has the properties that 0 is a center and that if there
are spheres with centers u and v then there are also spheres with centers
u + v and u - v. In other words the set of centers forms an additive
group. In crystallography these lattices are usually called Bravais lattices
([Hahll, [Kit4], [Plel], [Rys5]). We can find three centers VJ, V2, V3 (or
in general n centers VJ, V2 • . . . • Vn for an n-dimensional lattice) such that
the set of all centers consists of the sums ~ k i Vi where the k i are integers.
4 Chapter 1

The vectors v I • . . . • vn are then called a basis for the lattice. The
parallelotope consisting of the points
0IVI+"'+Onvn (O~Oi<l)

is a fundamental parallelotope. Figure l.2 shows a two-dimensional lattice

and the fundamental parallelotope determined by the basis v], V2. A
fundamental parallelotope is an example of a fundamental region for the
lattice, that is, a building block which when repeated many times fills the
whole space with just one lattice point in each copy. Figure l.3c below
shows a hexagonal fundamental region for a two-dimensional lattice.
There are many different ways of choosing a basis and a fundamental
region for a lattice A. But the volume of the fundamental region is
uniquely determined by A, and the square of this volume is called the
determinant or discriminant of the lattice. (Some authors use this term for
the unsquared volume.) There is a simple formula for the determinant.
Let the coordinates of the basis vectors be
VI = (VIl, V!2 . .... VIm),
V2 = (V2], V22 . .... V2rn),

where m ~ n. (Sometimes it is convenient to use m > n coordinates to

describe an n -dimensional lattice.) The matrix

M=

is called a generator matrix for the lattice, and the lattice vectors consist
of all the vectors
~M,

where ~ = (~I ..... ~n) is an arbitrary vector with integer components ~i'
The matrix

A = MM t"

where tr denotes transpose, is called a Gram matrix for the lattice. The
(j ,jhh entry of A is the inner product Vi . Vj. (Given two vectors
u = (UI . .... urn), v = (VI . .... vrn), their inner or scalar product
u I VI + ... + Urn Vrn will be denoted in this book either by u . v or
(u , v ).} The determinant of A is then the determinant of the matrix A,

det A = det A.
Sphere Packings and Kissing Numbers 5

FUNDAMENTAL
LATTICE PARALLELOTOPE
Figure 1.2. The plane divided into fundamental parallelotopes of a 2-
dimensional lattice.

If M is a square matrix this reads

det A = (det M)2. (5)

Consider for example the familiar planar hexagonal lattice shown in

Fig. 1.3a. An obvious generator matrix is

M = [1~2 1/2~) ' (6)

for which the Gram matrix is

A = MMtr = [I1/2 1/2)

1 '
(7)

and det A = det A = 3/4. But for many purposes it is better to use the
generator matrix

(8)

with Gram matrix

A'=M'M'tr= [ 2-I)
-I 2 ' (9)

and now det A = det A' = 3. Eq. (8) uses three coordinates to define a
two-dimensional lattice, but the points all lie in the plane x + y + z "" O.
Eqs. (6) and (8) both describe the hexagonal lattice in Fig. 1.3a, but in
different coordinates and on a different scale. Formally we say that the
lattices defined by (6) and (8) are equivalent (see §1.4). The advantages
of (8) over (6) are that the coordinates are nicer (the J3's have
disappeared), and the symmetries of the lattice are more easily seen. It is
 6 Ch a pte r 1

0 0 0

0 0 0 0

0 0 0 0 0

0 0 0 0

0 0 0

(0) (b)

(C)

Figure 1.3. (a) The hexagonal lattice in the plane. (b) The
corresponding sphere-packing (in this case a circle-packing). The points
labeled a are the lattice points; the points labeled band c are the "deep
holes" in this lattice. the points of the plane furthest from the lattice.
(c) The Voronoi cells are regular hexagons. (d) An enlargement of (b)
and (c). showing the packing radius p of the lattice (the radius of the
circles in (b». and the covering radius R. the distance from a lattice
point to a deep hole. For this lattice R - 2p/.J3.

clear from (8) that the three coordinates can be permuted in any way
without changing the lattice. (8) is the two-dimensional representative
(A 2) of the infinite sequence of root lattices A I> A 2. A 3.... (see Chap. 4.
§6.1).
We can now give a precise definition of the density .:l of a lattice
packing:

.:l = proportion of the space that is (I 0)

occupied by the spheres
Sphere Packings and Kissing Numbers 7

A DEEP HOLE

o o

(d)

Figure 1.3 (con!.)

volume of one sphere

volume of fundamental region
volume of one sphere
(detA) 1/2

If the hexagonal lattice is defined by (6) we may take the spheres

(actually circles in this case) to have radius p - 1/2, obtaining the packing
shown in Fig. 1.3b, with density

.:l = .Ji2
7r
= 0.9069 .... (] 2)

Using (8) leads to the same value: the density is independent of the choice
of coordinates, as we should expect it to be.
1.3 Nonlattice packings. Of course most packings are not lattices. In
three dimensions, for example, there are non lattice packings that are just
as dense as the face-centered cubic lattice. This happens because the fcc
can be built up by layers, beginning with one layer of spheres placed in the
hexagonal lattice arrangement of Fig. 1.3b, with the centers at the points
marked 'a'. There are then two (equivalent) ways to place the second
8 Chapter 1

layer: the spheres can be placed above the positions marked 'b' or above
those marked 'c'. Suppose we place them at 'b'. The third layer can now
be placed either above the positions marked 'a', i.e. directly above the first
layer, or above the positions marked 'c'. Similarly there are two choices
for every remaining layer. Choosing the layers in the order
... abcabcabc ... or .. .acbacbacb ...

produces the face-centered cubic lattice. But there are uncountably many
other (nonlattice) possibilities, such as ... acbabacbca ... , all with the same
density. Taking the layers in the order ... ababab... produces the
hexagonal close packing or hcp (as found for example in helium crystals
[Kit4], [WeI4]). We shall give a more detailed study of such laminating
constructions in Chaps. 5 and 6.
The hexagonal close packing, like most of the packings encountered in
this book, is a periodic packing, one that is obtained by placing a fixed
configuration of say s spheres in each fundamental region of a lattice A.
Another example is the tetrahedral or diamond packing (Chap. 4, §6.4)
that is obtained by placing s = 8 spheres in each fundamental region of
the simple cubic lattice. The density of a periodic packing is given by
.:1 = s . volume of one sphere (I3)
(detA) 1/2

In the diamond packing the spheres have radius .J3/8, so the density is
only
7r.J3
.:1 = 16 = 0.3401....

The density of an arbitrary (not necessarily periodic) packing is found

by taking a very large spherical region, of radius R say, calculating the
proportion of this region that is occupied by the spheres of the packing,
and letting R - 00 (see [Rog7, Chap. 1]).
1.4 n-dimensional packings. The first main problem of the book, then, is
to find the densest (lattice or non lattice) packing of equal nonoverIapping,
solid spheres (or balls) in n-dimensional space. In one dimension this is
trivial, for a one-dimensional ball is a line segment, a rod for example, and
we can obtain a density .:1 = 1 by centering these rods at the integer points
on a line (Fig. 1.4). These points form the one-dimensional lattice z.

The answer is also known in two dimensions: the highest density that
can be attained is .:1 = 7r /m = 0.9069 ... , as in the hexagonal lattice
packing of Fig. 1.3b. This result has a long history - see especially
Thue's 1910 paper [Thul1 and the proof given by Fejes T6th in 1940
[Fej31. For a concise and elegant proof see [FejlO, pp.58-611. Other
references are [Foil], [Gra5], [Rog7, p. 11], [Seg11.

In three dimensions, as we remarked in §1.1, the answer is unknown.

But if we consider only lattice packings then the answer is known: Gauss
Sphere Pac kings and Kissing Numbers 9

"SPHERE"
~
---0 ! 0 ! 0>-----'--<0)--'---0---
-I 0 1 2 3

Figure 104. The I-dimensional lattice Z, the integers.

showed in 1831 [Gau2] that the face-centered cubic lattice is the densest
lattice packing in three dimensions. Proofs of this result may be found in
[Cas2, Chap. II, Th. IIIl, [Con42], [Cox14, §18.4], [Demll or [MorS].
What about dimensions higher than 3? To demonstrate how easy it is
work in 4-dimensional space, we describe the densest sphere packing known
in four dimensions. This is the checkerboard lattice D4 (Chap. 4, §7.2), in
which the centers of the spheres are all the points (u b U 2, U 3, U4) where the
Uj are integers that add to an even number. Thus (0,0,0,0) is allowed,
and (I, 1,0,0), but not (1,0,0,0). The sphere centered at (0,0,0,0) has 24
spheres around it, centered at the points (± 1, ± 1,0,0), with any choice of
signs and any ordering of the coordinates. (There are V2 4' 3 - 6 positions
for the O's and then 4 choices for the signs, for a total of 24.) Any two
distinct centers must differ by at least 1 in at least two coordinates, or by
at least 2 in one coordinate, so the minimal distance between the centers is
.Ji. Then if the spheres are taken to have radius p = .Ji12 = 1/.Ji, they
will not overlap, and each sphere will touch 24 others.
It is clear from the definition that D 4 contains the centers (2,0,0,0),
(0,2,0,0) ..... (0,0,0,2), (1,1,0,0), (1,0,1,0) ..... (0,0,1,1), and
conversely that each center of D4 is an integer combination of these
vectors. Actually it is enough to use (2,0,0,0), (1,1,0,0), (1,0, 1,0) and
(1,0,0,0, and
2 0 0 0
1 1 0 0
M= o 0
(Is)
o 0
is a generator matrix for D 4. [E.g. (0,2,0,0) = 2 (1,1,0,0) - (2,0,0,0)']
Therefore D4 has determinant det D4 = (detM)2 = 4.
In order to compute the density of D4 we need to know the volume of
an n -dimensional sphere of radius p: this is

where Vn , the volume of a sphere of radius 1, is given by

7rn/2 2 n7r(n-O/2«n-I)/2)!
(17)
Vn = (nI2)! n!

(The second form avoids the use of (nI2)! when n is odd. See §2C of
Chap. 21, or [Soml, p. 1361.) Also

(18)
10 Chapter 1

where 0 < E < (Iog2e)/(6n). The surface area of a sphere of radius p is

(19)

From (11) and (16) the density of a lattice packing A is given by

V. pn
Ll = _....;;nc...:....,,-,:- (20)
(detA) 1/2 '

where p is the radius of the spheres. We take p to be half the minimal

distance between lattice points; p is called the packing radius of A.
For the D4 lattice we have V4 = 71' 2/2, p = 1/.J2, so the density is
71'2
Ll = 16 = 0.6169... . (21)

This is the highest density known in four dimensions, and Korkine and
Zolotareff proved in 1872 [KorI] that this is the highest possible density
for a 4-dimensional lattice packing. By using Mordell's inequality
(Eq. (I9) of Chap. 6) Korkine and Zolotareff's result can be deduced
immediately from the optimality of the fcc lattice packing ([Mor41,
[Oppl]).
D4 may be generalized in an obvious way to an n-dimensional packing
Dn (take n integer coordinates with an even sum). In Chap. 4 we
summarize the properties of the lattices Dn, and of other important lattices
that occur throughout this book, including the n-dimensional lattices An
(for n ~ 1), the 6-, 7- and 8-dimensional lattices E 6, E7 and E 8, and the
24-dimensional Leech lattice A24 . (The subscript usually gives the
dimension,)
If one lattice can be obtained from another by (possibly) a rotation,
reflection and change of scale we say they are equivalent, or similar,
written =. Two generator matrices M and M' define equivalent lattices if
and only if they are related by
M' = cUMB, (22)

where c is a nonzero constant, U is a matrix with integer entries and

determinant ± 1, and B is a real orthogonal matrix (with BBI' = The n.
corresponding Gram matrices are related by
A' = c 2 U A U I' . (23)

If U has determinant ± 1 and c = 1 then M and M' are congruent

lattices (they are directly congruent if det U = + 1). For example the fcc
lattice occurs in both the An and Dn sequences as A3 = D 3 •
Any n-dimensional lattice Ln has a dual (or reciprocal, or polar)
lattice, L;, given by
L; = {x E Rn : x . u E Z for all u E Ln}. (24)
Sphere Packings and Kissing Numbers 11

For example the dual of the fcc is the body-centered cubic or bcc lattice
A; ;;;; D; (see Chap. 4, §6.8). If A is a Gram matrix for L n , L: has Gram
matrix A-I, and detL:- (detLn)-I. If M is a square generator matrix
for L n , then (M-I)'r is a generator matrix for L:.
Why do we care about finding dense packings of n-dimensional
spheres? There are many reasons.
(j) This is an interesting problem in pure geometry. Hilbert mentioned it
in 1900 in his list of open problems [Hill], [MilS]. There is an enormous
literature (see the bibliography), from which we select a few items of
particular interest: [BarIl, [CoxI8], [Fejl], [Fej9], [FejIO], [GruIl,
[Hahl], [Ham2], [Hil2], [Mil7], [Rog7], [Sch16], [SigIl, [SI014], [We14].
Furthermore the best packings turn out to have connections, sometimes
totally unexpected, with other branches of mathematics. For example the
best lattice packings in up to eight dimensions belong to the families An,
Dn and En, and the corresponding Coxeter-Dynkin diagrams (see Chap. 4)
turn up in several apparently unrelated areas - see Hazewinkel et al.'s
survey article "The ubiquity of Coxeter-Dynkin diagrams (an introduction
to the A-D-E problem)" [HazIl. Similarly in 24 dimensions the Leech
lattice A24 has mysterious connections with hyperbolic geometry, Lie
algebras, and the Monster simple group - see Chaps. 23-30 of this book,
and the "Monstrous Moonshine" articles [Conll], [ConI7], [FonIl,
[Kac2]-[Kac6], [Koil], [Konl], [Kon2], [Mas2], [Mas3], [NorS], [Smil3],
[ThoS], [Th06]. We expect that one day someone will write an article on
"The ubiquity of the Leech lattice".
OJ) There are direct applications of lattice packings to number theory,
for example in solving Diophantine equations, and to "the geometry of
numbers" - see [Cas2], [Grul], [Grula], [Han3], [HlaIl, [Hla3], [Kell],
[Min4]-[Min6]. We shall say more about this in Chap. 2, §2.
(iij) There are important practical and theoretical applications of sphere
packings to problems arising in digital communications, as we shall see in
Chap. 3. Just to give the flavor, here is a typical question from the design
of spread-spectrum communications for mobile radio (cf. [CooIl, [Mazl)):
how many spheres of radius 0.43 can be packed into a sphere of radius I in
100-dimensional space?
Ov) Two- and three-dimensional packings have many applications. For
example the circles in a two-dimensional packing may represent optical
fibers as seen in the cross-section of a cable [KinIl. Three-dimensional
packings have applications in chemistry and physics [Berll], [HoaIl,
[Hoa2], [Kit4], [O'KeIl, [SI017], [SI019], [Teol]-[Te03], [WeI2]-[WeIS],
[ZimIl; biology [RiU], [TamS]; antenna design [Strl]; choosing directions
for X-ray tomography [She3], [She4], [Smi7]; and in performing
statistical analysis on spheres [Wat24].
(v) n-dimensional packings may be used in the numerical evaluation of
integrals, either on the surface of a sphere in Rn or in its interior. See
12 Chapter 1

[Bab2], [Bou2], [DeI6], [De116], [FroI], [Hla2], [Hual], [KeaI], [Niel],

[Ro02], [ShaO], [Sob3], [Str3], [Zarl] and especially [Goe5], [McLO],
[Slnl], [Sln2], [Sob21. We shall discuss this topic further in §3.2 of
Chap. 3. A related application that has not yet received much attention is
the use of these packings for solving n -dimensional search or
approximation problems - cf. [Airl], [DobI], [RenI], [Sobll
(vi) Recent developments in physics (in dual theory and superstring
theory) have involved the E g and A24 lattices and the related Lorentzian
lattices in dimensions 10 and 26 discussed in Chaps. 26 and 27 ([Cha3],
[Godl], [Grel], [Jac2], [Sch3], [SchI5], [ThiI], [Thi2]).
1.5 Sphere packing problem - summary of results. In this section we
summarize what is presently known about the sphere packing problem.
Table 1.1 shows the results in dimensions 1 to 8, 12, 16 and 24, comparing
the densest known packings with the answers to some of the other questions
discussed in the first two chapters.
The entries in Table 1.1 enclosed in boxes are known to be optimal.
The boxes in the first row indicate that the densest possible sphere packings
are known only in 1 and 2 dimensions! The entries to the left of the double
lines are known to be optimal among lattices. In particular the densest
possible lattice packings are known in dimensions n ~ 8. We have already
mentioned dimensions 1-4. Korkine and Zolotareff [Kor3] showed that D5
is the densest lattice packing in 5 dimensions, and guessed E 6 , E 7 , E g, A9,
A 10 in the next five dimensions. Blichfeldt [Bli2]-[Bli4] established the
optimality of E 6, E 7 and E g. Blichfeldt's proof is quite difficult, but has
been confirmed by Watson [Wat5] and Vetchinkin [Vet21 Mordell's
inequality enables the optimality of Eg to be deduced from that of E 7 , but
no simple proof of the optimality of E 7 is known, nor has Blichfeldt's work
been extended to higher dimensions. The optimality of E 6 also follows

Table 1.1. Records for packings, kissing numbers, coverings and

quantizers. (Box: optimal. To left of double line: known to be optimal
among lattices.} For n ~ 8 the entry in the first row is =- An.

DIMENSION I 2 3 4 5 6 7 8 12 16 24

DENSEST
PACKING ~~A3 D4 ~ E6 E7 E8 K!2 A!6 A24

rn~~ E7~
~
HIGHEST 04 D!5 E6 K!2 A!6
KISSING
2 6 12 24 40 72 126 240 756 4320 196560
NUMBER

THINNEST
COVERING 0~A; A~ A; A: A~ A; A~2 Ar6 A24

BEST
QUANTIZER 0~A; D4 0; E: E~ E8 K!2 A!6 A24
Sphere Packings and Kissing Numbers 13

from Barnes's classification of all 6-dimensional lattices whose density is a

local maximum [Bar6], [Bar71. It is known that the densest lattice
packings in dimensions 1 to S are unique. (See Barnes [Bar6], [Bar7] and
Vetchinkin [Vet2] for the uniqueness of E 6, E 7, Es.)
The symbol An in Table 1.1 represents a laminated lattice. These will
be defined and studied in detail in Chap. 6. In particular the laminated
lattices are the densest lattices in up to S dimensions, and we have the
following equivalences:
AI ~z ~AI ~A;, A2-A2~A;,
A3~A3~D3, A4 ~ D4 ~ D:,
(25)
As ~ D s, A6 ;; E 6 ,
A7 ~ E7, As ~ Es ~ E;.

AI6 ~ BW l6 is the Barnes-Wall lattice (§10 of Chap. 4, [BarIS]), A24 is

the Leech lattice (§ll of Chap. 4, [Lee5]) and K 12 , the only entry in
Table 1.1 not yet mentioned, is the Coxeter-Todd lattice (§9 of Chap. 4,
[Cox29]). It seems likely that K\2, AI6 and A24 are the densest lattices in
dimensions 12, 16 and 24 respectively, although this has not been proved.
In fact it is a reasonable guess that all the entries in Table 1.1 are optimal.
The densities of these packings are shown in Fig. 1.5 and in Table 1.2,
which is an updated version of a table in [Leel0], and includes most of the
best packings known in up to 24 dimensions. As well as the density ~
itself, Table 1.2 shows the center density 0, given by
~
0=- (26)
v.'
n

which is a much simpler number. For A24 for example (an extreme case),
~ = 11"12/479001600 = 0.001930 ... , whereas 0 = 1. The table gives the
exact values only for o. If the spheres have radius 1, 0 is the number of
centers per unit volume. For a lattice packing, from (20),
o= pn (detA)-1/2. (27)

The center densities in Fig. 1.5 have been rescaled to make them easier
to see. The vertical axis gives
1
log20 +% n(24-n).

Fig. 1.5 and Table 1.2 also give the best upper bound on the center density
presently known. For n = 3 this is Lindsey's bound mentioned in §1.1,
while for n ~ 4 it is Rogers' bound [Rog2] as calculated by Leech [Lee5].
Leech's figures are truncated to 5 decimal places, but apart from this the
last digits of all decimal expansions in this book have been rounded off to
the closest digit. In the column headed Type in Table 1.2, B indicates that
both a lattice and a nonlattice packing with this density and kissing
number are known, L indicates that only a lattice packing is known, N
that only a nonlattice packing is known and A indicates a local
14 Ch a pte r 1

2 DENSITY
10928 + ~ n (24-n)

o A~

-1

-2

-3
DIMENSION n
o 4 8 12 16 20 24 28 32 36 40 44 48
Figure 1.5. The densest sphere packings known in dimensions n ~ 48.
The vertical axis gives log20 + n (24-n)/96, where 0 is the center density.
The An are laminated lattices, the Kn are described in Chap. 6, K 12 is the
Coxeter-Todd lattice, the crosses are non lattice pac kings (Chap. 5,
§§2.6,4.3), and Q32, B36 and P48q are described in Table 1.3a. The upper
bound is Rogers' bound (39), (40).

arrangement (or cluster) of spheres touching one sphere. The previous

column indicates where in this book these packings are described.
We see from Fig. 1.5 and Table 1.2 that the laminated lattices An are
the densest packings known in dimensions n ~ 29, except for dimensions
10-13. In dimension 12, as already mentioned, the Coxeter-Todd lattice
K 12 is the densest known, and in dimensions 10, 11 and 13 there are non-
lattice packings, to be described in Chap. 5, that are denser than any
known lattices. In dimensions 30-32 Quebbemann's lattices (§4 of
Chap. 8, [Que51, [Que6]) surpass An.
Table 1.3 gives a selection of the densest packings known from
dimensions 24 to about 106 . All these packings are lattices. The lattices in
Table 1.3 are all quite explicit: without too much difficulty one could write
down generator matrices (the amount of effort needed being polynomial in
the dimension n) .
In contrast, Minkowski gave a nonconstructive proof in 1905 that there
exist lattices with density satisfying
~ ~ ~(n) (28)
:7 2n - I '

where ~(n) = :2::k'-1 k- n is the Riemann zeta-function . Thus

log2 ~ ~ - n + I, as n --+ 00 . (29)

This implies (from (27» that there are lattices with center density at least
as large as the following :
n 128 256 512 1024 4096 65536 1048576
log2o 63 249 750 2006 12102 324603 7290660
Sphere Packings and Kissing Numbers 15

Table 1.2. Sphere packings in up to 24 dimensions.

Name of Density Center density 0 Kissing number
n packing 6. Attained Bound Max Ave Ch .• § Type

o Ao I I 6,2 L
AI ;A I 1/2 - 0.5 0.5 4,6.1 L
2 A 2 ;;;; A2 0.90690 1/2JJ - 0.28868 0.28868 6 6 4,6.2 L
3 AJ ; DJ 0.74048 1/4..fi. -0.17678 0.1847 12 12 4,6.3 B
4 A4; D4 0.61685 1/8 - 0.125 0.13127 24 24 4,7.2 L
A5;;;; D5 0.46526 l/8..fi. - 0.08839 0.09987 40 40 4,7.1 B
6 A6 "" £6 0.37295 l/8JJ - 0.07217 0.08112 72 72 4,8.3 B
A7; £7 0.29530 1/16 - 0.0625 0.06981 126 126 4,8.2 B
Ag "" £S 0.25367 1/16 - 0.0625 0.06326 240 240 4,8.1 L
9 A9 0.14577 1/16..fi. - 0.04419 0.06007 272 272 6,4 B
P 9a 0.12885 5/128 - 0.03906 306 235 3/5 5,2.6 N
10 AIO 0.09202 1/16JJ - 0.03608 0.05953 336 336 6,4 B
P lOa 0.09463 19/512 - 0.03711 372 353 9/19 5,2.6 N
P lOb 0.08965 9/256 - 0.03516 500 340 1/3 5,2.6 N
PICk 0.09962 5/128 - 0.03906 372 372 5,2.6 N
II A'rrJ. 0.05888 1/32 - 0.03125 0.06136 438 438 6,4 B
KII 0.06043 1/18JJ - 0.03208 432 432 6,2 L
Plio 0.06624 9/256 - 0.03516 566 519 7/9 5,2.6 N
P llb 580 5,4.3 A
PIle 582 5,2.6 A
12 Ai2ax 0.04173 1/32 = 0.03125 0.06559 648 648 6,4 B
KI2 0.04945 1/27 - 0.03704 756 756 4,9 L
LI2 0.04456 3 7/216 - 0.03337 704 704 5,3.3 N
Pl2a 0.04694 9/256 -0.03516 840 770 213 5,2.6 N
13 Aj{1. 0.02846 1/32 - 0.03125 0.07253 906 906 6,4 B
K IJ 0.02921 1/18JJ - 0.03208 918 918 6,2 B
P'3a 0.03201 9/256 - 0.03516 1130 1060 2/3 5,4.3 N
Pm 1066 5,4.3 A
14 AI4 0.02162 1/16JJ - 0.03608 0.08278 1422 1422 6,4 B
Pl4a 1484 5,4.3 A
P I4b 1582 5,4.3 A
15 AI5 0.01686 1/16..fi. - 0.04419 0.09735 2340 2340 6,4 B

P'Sa 2564 5,4.3 A

16 AI6 0.01471 1/16 = 0.0625 0.11774 4320 4320 4,10 B
17 A 17 0.008811 1/16 - 0.0625 0.14624 5346 5346 6,6 B
18 A ls 0.005928 l/8JJ - 0.07217 0.18629 7398 7398 6,6 B
19 AI9 0.004121 l/8..fi. = 0.08839 0.24308 10668 10668 6,6 B
20 A 20 0.003226 1/8 - 0.125 0.32454 17400 17400 6,6 B
21 A2I 0.002466 1/4..fi. -0.17678 0.44289 27720 27720 6,6 B
22 A22 0.002128 1/2JJ - 0.28868 0.61722 49896 49896 6,6 L
23 A 2J 0.001905 1/2 - 0.5 0.87767 93150 93150 6,6 L
24 A24 0.001930 - 1.0 1.27241 196560 196560 4,11 L

A comparison with Table l.3 shows that in dimensions above 1000 these
lattices are denser than any presently known. Unfortunately the proof of
(28) (see [Cas2, Chap.61, [Lekl, §191, [Rog?, Chap.4]) uses an
averaging argument, and does not say how lattices satisfying (28) may be
16 Chapter 1

Table 1.3. Sphere packings in more than 24 dimensions.

n Name log2o Bound Kissing no. Ch., §

32 A32 0 5.52 208320 6,7
BW32 0 146880 8,8.2f
e 32 1 249280 8,8.2h
Q32 1.359 261120 8,4
36 P36p -1 8.63 42840 5,5.5
A36 1 234456 5,5.3
B36 2 8,8.2d
48 A4s 12 15.27 6,7
P 4Sp 14.039 52416000 5,5.7
P 4Sq 14.039 52416000 5,5.7
60 P 60p 16.548 27.85 3908160 5,5.5
64 BW64 16 31.14 9694080 8,8.2f
Q64 18.719 2611200 8,2
P 64c 22 8,8.2e
80 T/ (Es) 36 49.90 8,lOc
96 T/ (p 4Sq) 52.078 70.96 8, 109
104 T/ (Es) 60 80.20 8,lOc
128 BWI28 64 118.6 1260230400 8,8.2f
P\2Sh 85 5,6.6
T/ (Es) 88 8,lOc
136 T/ (Es) 100 129.4 8,lOc
150 A (14) 113.06 153.2 8.6
150
180 T/ (A 20) 133 206.7 8,lOc
A (16) 154.12 8,6
ISO

found. Many generalizations and extensions of (28) have been found,

although no essential improvement is known for large n. In its general
form this bound is known as the Minkowski-Hlawka theorem.
We still do not know how to construct packings that are as good as
(28). In 1959 Barnes and Wall found an infinite sequence of lattices BWn
in dimensions n = 2m with

1 1
-;; log2 ~ - - '4 log2 n, as n - 00 (30)

(§8.2f of Chap. 8, [BarI8]); in 1971 [LeelOl (see §6.7 of Chap. 5)

constructed nonlattice packings with

1 1
-;; log2 ~ - - '2 iog2 log2 n ;
Sphere Packings and Kissing Numbers 17

Table 1.3 (cont.)

n Name loU) Bound Ch., §

192 1] (A 24) 156 230.0 8,10e

A (17)
171.44 8,6
192
256 BW256 192 357.0 8,8.2f
B 256 250 8,8.2g
A (22) 270.89 8,6
256
508 A (40)
742.66 948.1 8,6
508
512 B512 698 957.4 8,8.2g
520 A (41) 767.46 980.1 8,6
520
1020 A (73)
1922 2406 8,6
1020
1030 A (73)
1947 2439 8,6
1030
2052 A (134)
4755 5871 8,6
2052
4096 1] (A I6 ) 11344 13750 8,lOc
4098 A (245)
11279 13758 8,6
4098
8184 1] (A 24 ) 26712 31547 8,10e
8190 A (453)
26154 31573 8,6
8190
8208 1] (A 24) 26808 31655 8,10e
16380 A (843)
59617 71325 8,6
16380
16392 1] (A 24) 61608 71387 8,10e
32784 1] (A 24 ) 139488 159154 8,10e
65520 1] (A 24 ) 311364 350788 8,10e
65544 1] (A 24) 311496 350932 8,10e
131088 1] 3(A 24) 664962 767395 8,10f
262152 1] 3(A 24) 1.475.106 1.666.106 8,10f
524304 1] 3(A 24) 3.178.106 3.594.107 8,10f
1048584 1] 3(A 24) 6.918.106 7.711.107 8,10f

and in 1973 [Slo11 (see §6.8 ot Chap. 5) gave nonlattice packings for
which
1
- log 2 Ll > - 6.
n -

Lattices with
1 1
- log 2 Ll - - - log2 log2 n
n 2

were found in [BariS] (see Chap. 8, §8.2g) and by Craig (see Chap. 8,
§6), and lattices with
1 •
- log2 Ll - - log2 n
n
18 Chapter 1

were given in [Bos3] (see Chap. 8, §10h). Here log; n is the smallest
value of k for which the kth iterated base 2 logarithm of n is less than 1.
The lattices constructed in Chap. 8 are however still reasonably dense in
quite large dimensions. For example there are lattices in dimensions
n ::s;; 98328 with
1
-log2 ~ = -1.2454 ... + E,
n

where lEI < 12 n- 1 log2(n/6), in dimensions n ::s;; 1051 with

.llog2 ~ = -2.0006 ... + E',

n

and in dimensions n ::s;; 109870 with

.llog2 ~ = -2.2005 ... + E". (35)

n

Again all these packings are quite explicit. Litsyn and Tsfasman [Lit3],
[Lit5] have recently announced that nonlattice packings with
1
-log2 ~ > -1.31, (36)
n -

and lattices with

1
- log2 ~ > -2.30
n -

may be constructed using the methods of Chap. 8 applied to codes obtained

from algebraic curves (cf. Chap. 3, §2.ll).
As we shall see in §7.4 of Chap. 8, the infinite class field towers found
by Golod and Shafarevich [GoI9], Martinet [Mar2] and others produce
infinite sequences of lattices satisfying, in the most favorable case known,

1 (38)
-log2 ~ - -2.218... as n -- 00.
n

(This was pointed out by Litsyn and Tsfasman [Lit3].) However there
does not seem to be any practicable method known for explicitly finding
these lattices.

Remark. Although we have described Minkowski's result and others as

"nonconstructive", there are algorithms that would theoretically enable one
to write down generating vectors in a bounded time, and so in the logical
sense these results are effective. However the typical such algorithm
involves a search through an extremely large population and would require
a super-exponential running time. In contrast our "explicit" constructions
can be used to write down generators very quickly. We deliberately refrain
from making precise definitions, since we feel that these distinctions are
best made informally.
Sphere Pac kings and Kissing Numbers 19

Very recently it has been shown in [Rusl1 that packings in dimension

n = p2, where p is a prime, with
1
- log2Ll > -1.000000007719 ...
n -

(the right-hand side is -0 + 2 (Jog2e)/e,,2+ ... » can be found by

searching over a specific finite set of error-correcting codes of length n.
Although this result is also "nonconstructive", the search is over a much
smaller population than in Minkowski's theorem.
Minkowski's result (28) guarantees that dense lattices exist. In the
other direction, Rogers showed· in [Rog2], [Rog7] that the density of any
n -dimensional packing satisfies
(39)

where (J n is defined as follows. Let S be a regular n -dimensional simplex

of edge length 2. Spheres of radius 1 centered at the vertices of S are
disjoint, and (J n is the ratio of the volume of the part of S covered by the
spheres to the volume of S. Eq. (39) implies that no two-dimensional
packing can be denser than the hexagonal lattice, and in three dimensions
it gives the bound 0.7796 ... mentioned in §1.1. Another form of (39) found
by Leech [Leel0] is
1 3
log2 0 ~ - n log2(--)
n
+ -log2 n -log2
e
r + -5.25
2 5 ' (40)
2 4e1l" 2 "11" n+ .

the last term being approximate. For large n (39) gives

1
-log 2 Ll < -0.5. (41)
n -

In 1979 Levenshtein [Lev7] showed that

where j (t) is the smallest positive zero of the Bessel function J/ (x). For
large n = 2t the approximate formula ([Bos2], [Olvl])
j{t) - ( + 1.8557571 + 1.033150 (-1/3
(1/3 - 0.003971 (-1

- 0.0908 t- 5 / 3 + 0.043 (-7/3 (43)

can be used in (42). Asymptotically (42) gives

1
- log2 Ll < -0.5573 ....
n -

In 1979 Kabatiansky and Levenshtein [Kabl1 obtained an even stronger

bound, which for large n gives
1
- \Og2 Ll < - 0.5990. (45)
n -
20 Chapter 1

The precise form of their bound and an outline of its proof are given in
Chap. 9.
Other bounds are given in [Lev4], [Lev5], [Lev9], [Rog7], [Sid 1],
[Sid3], [Ural]. Rogers' bound (39), (40) is the best for n ~ 42, and
above that the Kabatiansky-Levenshtein bound takes over. The upper
bounds in Table 1.3 are obtained from these two bounds.
Table 1.4 compares the various densities in dimension 65536, showing
the best packings known, the density that we know from Minkowski's
result is possible, and three upper bounds on the highest attainable density.
In summary, from (29) and (45) the densest (lattice or non lattice)
packings satisfy
1
-1 < -log2 Ll < -0.5990 (46)
n -

for large n. (Roughly speaking, when the dimension increases by 1, the

density of the best packing is divided by a number between 2 and 1.51.)
Many writers on number theory use Hermite's constant T'n, given by
(47)

where On is the center density of the densest lattice packing in Rn [Lek 1,

p. 294]. T'n is known for n ~ 8 (it can be obtained from Tables 1.1, l.2),
and for large n (29) and (45) imply

(48)

In view of Eq. (28)-(34) of Chap. 3, T'n measures the highest attainable

coding gain of an n-dimensional lattice.

Table 1.4. Comparison of center densities (, in dimension 65536.

Type Name log2o Chap.

Constructions BW65536 180224 8,§8.2f
B65536 290998 8,§8.2g
11 (\'32) 295120 8,§IOc
A (2954)
297740 8,§6
65536
11 (A 24)* 311364 8,§IOe
Existence Minkowski 324603 I, (28)
Upper Kab.-Lev. 350885 9
bounds Levenshtein 353768 1, (42)
Rogers 357385 1, (40)

* in dimension 65520
Sphere Packings and Kissing Numbers 21

Although this book only considers the problem of packing a very large
number of spheres, there is a considerable body of literature dealing with
the best packings of N spheres for small values of N. See for example
[Ben3], [Boel1, [Fej2a], [Fej2b], [GoI8], [Gra4a], [Gril2]·[Gril4], [GroO],
[Wegl1, [WilO).

2. The Kissing Number Problem

2.1 The problem of the thirteen spheres. The second main problem in this
book, closely related to the packing problem, is fondly known as the kissing
number problem. In three dimensions this asks how many billiard balls
can be arranged so that they all just touch, or "kiss", another billiard ball
of the same size. Don't imagine this on a billiard table - instead consider
one billiard ball fixed in space, and mentally try to arrange the others
around it. (For this problem billiard balls seem easier to "handle" than
ball bearings, and besides "kiss" is a billiards term!) This question was the
subject of a famous discussion between Isaac Newton and David Gregory
in 1694. Newton believed the answer was 12, as found in the fcc lattice
for example (see Fig. l.la), while Gregory thought that 13 might be
possible.
The correct answer to this question, which is often called the problem
of the thirteen spheres, is now known to be 12. Some proofs appeared in
the nineteenth century in [Benl1, [Hopl1 and [GiinH Schiitte and
van der Waerden gave a detailed proof in 1953 [Sch12). The best proof
now available is Leech's [Lee2], but although elementary and
straightforward, it cannot be called trivial. (See also [Boel1, [WasH)
The difficulty arises because the arrangement is not unique; in fact
there are infinitely many ways to arrange 12 billiard balls around another.
For example, if the 12 balls are placed at positions corresponding to the
vertices of a regular icosahedron concentric with the central ball, the
twelve outer balls do not touch each other and may all be moved freely.
(In fact any permutation of the twelve spheres can be achieved by rolling
the outer spheres around the inner one - see the Appendix). The
icosahedral arrangement of 12 atoms around a central atom occurs in
certain alloys and other compounds [Bril], [Teo2], [WeI4, p. 71], and in
the recently discovered non periodic structures known as quasilattices or
quasicrystals (see for example [ConI6a], lEis 11, [Els2], [Gra6], [Grii3],
[LevlO], [Levl11, [Nell], [SheO]), while the arrangement of 12 atoms
around one found in the fcc lattice (forming the vertices of a
cuboctahedron) is very common [WeI4, p. 711.
2.2 Kissing numbers in other dimensions. More generally we may define
the kissing number (usually denoted by T) of a sphere packing in any
dimension to be the number of spheres that touch one sphere. For a lattice
packing T is the same for every sphere, but for an arbitrary packing T may
vary from one sphere to another. Other names for T that have been used
are Newton number (after the originator of the problem), contact number,
coordination number or ligancy (the last two being chemist's terms).
 22 Chapter 1

The n -dimensional version of the kissing number problem then asks for
the greatest value of T attained by any packing of n-dimensional spheres.
In one-dimensional space the answer is 2, and in two dimensions it is 6 (it
is easy to show that although six pennies can be arranged around another
penny on a table-top without overlapping, as in Fig. 1.3b, seven can not.)
In three dimensions, as we have mentioned, the answer is 12.
It is somewhat surprising that we also know the answers in 8 and 24
dimensions ([Od15] == Chap. 13, [Lev7]), but in no other dimension above
3. In fact these numbers (240 and 196560 respectively) are technically
easier to establish than the result in three dimensions. This is partly
because in these dimensions the arrangements are unique: the only way to
place 240 8-dimensional spheres around a central sphere is the
arrangement found in the E 8 lattice, and similarly in 24 dimensions the
arrangement must be that found in the Leech lattice in one of its two
mirror-image forms ([Ban13] ;;;;; Chap. 14).
On the other hand in four dimensions, for example, the highest
attainable kissing number is at present only known to be either 24 or 25.
24 occurs in the D4 lattice, while the best bound known is 25. The kissing
numbers of various packings in dimensions up to 24 can be found in
Tables 1.1 and 1.2, while Table 1.3 includes the kissing numbers of some
lattices in dimensions 24 to 128. (Beyond that the kissing numbers in
Table 1.3 are not known.> The best bounds currently known on Tn, the
highest attainable kissing number in dimension n, are shown in Table 1.5
for 1 ~ n ~ 24. The lower bounds in this table are taken from Table 1.2,
and the upper bounds will be derived in Chap. 13.
So far we have considered the kissing numbers of arbitrary
arrangements of spheres. If we restrict ourselves to lattice packings rather
more is known: Watson [Wat7] showed that the laminated lattices An
have the highest possible kissing numbers for lattices in dimensions n ~ 9.
For n ~ 8 the highest kissing numbers presently known for arbitrary
packings coincide with what can be attained by lattices. However in
dimension 9 a nonlattice packing (p 9a ) exists in which some spheres touch
306 others (see Table 1.2 and §2.6 of Chap. 5), whereas the highest
possible kissing number for a lattice is 272 [Wat7] (see also [Wat6J,
[Wat8], [Wat9], [Wat13]). So as regards the maximal kissing number, 9
is the first dimension in which non lattice packings are known to be superior
to lattice packings.
It is worth commenting on our present state of knowledge about 11-
dimensional packings, which is summarized in the following table (part of
Table 1.2):
Name Type 0 Tmax

P lla nonlattice 0.03516 566

P llc nonlattice low 582
KII lattice 0.03208 432
A max
II lattice 0.03125 438
 Sphere Pac kings and Kissing Numbers 23

This suggests that in general we should expect the lattice and nonlattice
versions of the packing and kissing problems to have four different answers.
The moral is that the kissing number question is a local problem, while the
sphere packing question is a global problem! Of course the packings in
this table are not known to be optimal, and it is still possible that there is a
single lattice that is superior to all four.

Table 1.5. Range of possible values for T", the highest attainable kissing
number in dimension n. The third column gives the degree of the
polynomial used in Chap. 13 to obtain the upper bound.

n Tn Deg.
2
2 6
3 12
4 24-25 9
5 40-46 10
6 72-82 10
7 126-140 10
8 240 6
9 306-380 11
10 500-595 11
11 582-915 11
12 840-1416 11
13 1130-2233 12
14 1582-3492 12
15 2564-5431 12
16 4320-8313 13
17 5346-12215 13
18 7398-17877 13
19 10668-25901 13
20 17400-37974 13
21 27720-56852 13
22 49896-86537 14
23 93150-128096 14
24 196560 10

In higher dimensions (returning now to the general kissing number

problem) rather less is known. Kabatiansky and Levenshtein showed that
in n dimensions the kissing number is bounded above by
T ~ 2°.401n(l+o(l» (49)
24 Chapter 1

(see [Kabl1 and Chap. 9, Eq. (55», and Wyner [Wynl1 showed that
arrangements of spheres exist with
T ~ 2(J-O.5Iog23)n(J+o(I)
= 20.2075 ... n(J+o(I» . (50)

But Wyner's result, like Minkowski's Eq. (28), is nonconstructive. Up to

now the best constructive result was due to Leech, who showed that the
Barnes-Wall lattice BWn in dimension n = 2m has kissing number
m
T = II (2i +2)
i-I
_ 4.768 .... 2m (m+\)/2 = 4.768 .... 2o.5Iog2nOog2n+I)

(§6.5 of Chap. 2, §8.2f of Chap. 8, [BarIS], [BarI8], [Lee4], [Bos3]). The

Barnes-Wall lattices, already mentioned in Eq. (30), provide one possible
way to continue the sequence of lattices D 4, E g, . . . to all dimensions 2m .
However, as we shall see in Chaps. 5, 8, these lattices are based on Reed-
Muller codes and are therefore not especially good packings in high
dimensions. A16 , the third lattice in the sequence, has the highest density
and kissing number known in dimension 16 (see Chap. 4, §1O), but the
fourth lattice, which is equivalent to the laminated lattice A32 (Chap. 6) is
inferior to Quebbemann's 32-dimensional lattice Q32 (§4 of Chap. 8,
[Que51, [Que6D. For large n the kissing numbers of the Barnes-Wall
lattices are very small compared to (50). An explicit construction for
much higher kissing numbers will be given in §2.5.
2.3 Spherical codes. The kissing number problem can be stated in another
way: how many points can be placed on the surface of a sphere in n-
dimensional space Rn so that the angular separation between any two
points is at least 60 ? To see that this is exactly the same problem,
0

consider any arrangement of nonoverlapping spheres touching a central

sphere. Then the angular separation between the points where the outer
spheres touch the central sphere is at least 60 (Fig. 1.6). 0

Thus the kissing number problem can be regarded as a packing

problem, analogous to that considered in §1, but packing the points on the
surface of a sphere in Rn (rather than in Rn itself). This leads to an
important generalization. We denote the surface of the sphere by On:
(52)

We call a finite subset X of 0 n a spherical code (by analogy with binary

codes - see Chap. 3, §2), and say that X has minimal angle cP if cP is the
largest number for which
x .y ~ cos cP for any x, y E X, x ~ y.

(The inequality is ~ rather than ~, since the closer two points are on On,
the larger is their inner product. When x = y, x . y = 1. When y is the
antipodal point to x, i.e. y = -x, then x . y = -1.)
Sphe re Packings and Kissing Numbers 2S

Figure 1.6. If spheres A and B touch sphere C, then the angular

separation between the contact points P and Q, .J:. POQ , is at least 60· .
For if A touches B, the three centers form an equilateral triangle.

We now generalize the kissing number problem by asking, for given n

and <p, what is the maximal number A (n,<p) of points in a spherical code in
On of minimal angle <p? A (n, 7r /3) is the kissing number problem, and we
have seen that A (2, 7r /3) = 6, A (3, 7r /3) = 12, A (4, 7r /3) - 24 or 2S, etc.
The analogy with the sphere packing problem becomes clearer if we
consider equal spherical caps placed on the sphere and centered at the
points of the spherical code. Then A (n,<p) is the maximal number of
spherical caps of angular diameter <p that can be placed on On without
overlapping.
Spherical codes have many applications - see the references at the end
of §1.4, and also Chap. 3, §1.2. The three-dimensional problem, the
determination of A (3,<p), has been extensively studied. The inverse form is
often used: what is the largest angular separation that can be attained in a
spherical code on 0 3 containing M points? This is sometimes called
Tammes' problem, after the Dutch botanist who was led to this question
by studying the distribution of pores on pollen grains [TamSI .
Equivalently we can ask: where should M inimical dictators build their
palaces on a planet so as to be as far away from each other as possible?
Examples. The best code with M -= 4 points consists of the vertices of a
regular tetrahedron, with minimal angle <p = 7r - COS-I 1/ 3 :::: 109.47 · . For
M = 6, 8, 12 and 24 the best codes are given by a regular octahedron, a
square antiprism, a regular icosahedron and a snub cube, with <p = 90 · ,
74.8S9 · , tan-l 2 :::: 63.43S· and 43.7" respectively [FejlOl, [Robll,
[Sch Ill. But for M = 20 the answer is not the regular dodecahedron, for
which <p :::: 41.810 ·, since van der Waerden [Waell found a much less
symmetric code with <p :::: 47.431 · . The best codes known for larger values
of M are also often not the obvious (highly symmetric) candidates.
The best three-dimensional spherical codes presently known can be
found in [Bor3l, [Brull, [ClaIl, [CoxlS, p. 32SI, [Danll, [Fej21, [FejIO,
p. 1681, [Go13I-[GoI71, [Habll, [Leelll, [Mell I, [Robll, [Rob21, [SchIll,
[Str2l, [Sze2l, [Tar 11-[Tar4l, [Why II. Three-dimensional codes with
large numbers of points have been constructed in [Baull, [Lubll, [Lub21.
26 Chapter 1

See also [Arnl1, [Becl], [Ber9], [BerIO], [Blal1, [Gril2], [Lee3], [Linl],
[Mell], [Meyl], [Mohl], [Muel], [SI013a], [WeI4]. For four-dimensional
spherical codes see [Cox24], [MacO], [Nunl1, [War4].
There are two simple and general methods for constructing spherical
codes that are not restricted to three dimensions, one based on sphere
packings and the other on error-correcting codes. These constructions, like
the three- and four-dimensional codes just mentioned, all give lower bounds
on A (n,rjJ).
2.4 The construction of spherical codes from sphere packings. Let A be a
sphere packing in Rn, and let us place the origin of coordinates at a
convenient point P. (Usually P is taken to be the center of a sphere or a
point equidistant from a certain number of centers. For example, in
Fig. 1.3b we might choose P to be a point marked 'a' or 'b', or midway
between two points marked 'a'.) Suppose there are N spheres in A with
centers at distance u from P. Then these centers, when rescaled by
dividing them by u, form an n-dimensional spherical code of size N. In
other words we take a shell of points around P as the spherical code
[SI012].
The distance between the centers of the spheres is at least 2p (where p
is the radius of a sphere), so the minimal angle in this code is at least
2 sin-' (p/u). (53)

The number of points in this code is given by the theta series of the
packing with respect to P (Chap. 2, §2.3).
Examples. Consider the lattice D4 (§1.4), with the origin P at a lattice
point and p = 1/J2. The first shell has u = J2 and contains 24 centers
(the kissing number), and the corresponding spherical code consists of the
points T'/2(±I,±1,0,0), with minimal angle 2 sin-'1h = 60°. The
second shell has u = 2 and also contains 24 points, and the spherical code
consists of (± 1,0,0,0) and 1h (± 1, ± 1, ± 1, ± 1). Again the minimal angle
is 60 0, so in this case (53) does not give the exact value of rjJ. In fact this
second code is a rotation of the first. Both examples show that
A (4,60°) ~ 24.
On the other hand if we move P to 0,0,0,0) (a "deep hole" in D4 -
see Fig. 1.3d) we obtain a different sequence of spherical codes. The first
shell now contains 8 points at distance 1 from P, forming a code of
minimal angle 90 ° (the vertices of a regular 4-dimensional generalized
octahedron). Thus A (4,90 0) ~ 8. The numbers of points in these two
families of spherical codes are the coefficients of the theta series
Y2(8 3(2z)4 + 84(2z)4) and Y28 2 (2z)4 respectively (Chap. 4, §7).
2.5 The construction of spherical codes from binary codes. Let C be a
binary error-correcting code (Chap. 3, §2) of length n and minimal
distance d. A spherical code is obtained by changing the l's to -1's and
the O's to + l's in every codeword, and dividing by .In. The resulting
points lie on n n and the minimal angle is given by
Sphere Packings and Kissing Numbers 27

(S4)

d . 2 IjJ
-;; = sm "2' (SS)

Examples. The code containing all 2n binary vectors of length n produces

the spherical code consisting of the vertices n- I/2 (± 1, ... ,± 1) of an n-
dimensional cube. Any other spherical code obtained by the construction is
a subset of this. The Golay code C(J24 (Chap. 3, §2.8.2) with n -= 24,
d = 8 produces a spherical code containing 4096 points on 0 24 with
IjJ = COS-I 1/3 == 70.S29°. Thus A (24,coS-11/3) ~ 4096.

By using certain codes constructed by Justesen, Weldon, and Sugiyama

et al. (Chap. 3, §2.6), we may obtain infinite sequences of spherical codes,
with n --+ 00, having a fixed minimal angle IjJ and containing 2cn points
(where c depends only on 1jJ). Justesen's codes are the simplest, but have
dIn ~ 0.110 ... , and only produce spherical codes with IjJ ~ 38.73". For
larger angles the codes of Weldon and Sugiyama et al. must be used. The
kissing number problem for example requires IjJ = 60 0, and from the
Sugiyama et al. codes we obtain an explicit sequence of arrangements of
spheres in Rn with kissing number
T = 2o.003n {\+o{\)). (S6)

(This is not very good when compared with Wyner's existence result (SO),
but at least grows exponentially with n.)
2.6 Bounds on A (n, 1jJ). Again there is a nonconstructive lower bound
[Sha6], [Wynl]:
(S7)

(Eq. (SO) is the special case IjJ = 60°.) In the other direction there are a
number of bounds. We start with bounds that apply when IjJ is large, and
work down. Rankin [Ran4] found the exact value for IjJ ~ 90°:
A (n,ljJ) = 1, for 11" < IjJ ~ 211", (S8)

A (n,ljJ) - [1-secljJl, for sec-I(-n) ~ IjJ ~ 11", (S9)

A (n,ljJ) = n + 1, for 11"/2 < IjJ ~ sec-I (-n) , (60)

A (n, 11" /2) = 2n . (61)

The spherical codes corresponding to (60) and (61) are respectively the
vertices of a regular simplex and a regular cross-polytope or generalized
octahedron. Rankin also showed that
28 Chapter 1

Coxeter [Cox16] conjectured and Boroczky [Bor2] proved that

( ) 2Fn-l (a) (
A n,rP ~ Fn (a) , 63)

where sec 2a = sec rP + n - 2 and Fn (a) is Schliifli's function defined by

(64)

where U is the "area" of a regular spherical simplex of angle 2a contained

in fin. For large n, (63) is stronger than (62) by a factor approaching
2/e. For the kissing number problem, setting rP = 60· in (63) gives
1 1
-log2 T < ,
11 - 2

which is n.ot as good as the Kabatiansky-Levenshtein bound (49).

Kabatiansky and Levenshtein [KabI1 also show that for fixed rP with
o < rP < 7r /2, and large n,
l.- 1og2 A (n,,!,A.) < 1 + sin rP I
2 . og2
1 + sin rP 1 - sin rP 10g2 1 - sin rP (65)
n - SlllrP 2sinrP 2sinrP 2sinrP

and, for 0 < rP < rP',

l.-
n
log2 A (n, rP) < -
-
21 log2 (I - cos rP) - 0.0990, (66)

where rP' ::::: 63" is the root of a certain equation. When rP = 60", (66)
yields (49). Eqs. (65) and (66) follow from linear programming - see
Chap. 9, §3.5(iv). The linear programming method also produces many
tight estimates of A (n, rP) in particular cases - see for example Tables 1.4
and 9.2. For cos rP < l/.Jn, Delsarte et al. [DeI16] showed that
A (n,rP) ~ n(I-cosrP) (2+ (n+I)cosrP) , (67)
I-n cos 2rP

and Astola [Ast!] obtained

A (n,rP) ::s n (2.2 + loge (I + na» (68)

for cos rP = 0 (n- 2/ 3 ), and

A (n, rP) ::s V2 n loge (n cos rP) (69)

provided that n cos rP is unbounded as n - 00.

Bounds on A (n, rP) also lead to bounds on the density d of an n-

dimensional sphere packing via the inequality
(70)

(see Chap. 9, Th. 6). For example Eq. (45) was deduced from (66) and
(70).
Sphere Packings and Kissing Numbers 29

Sphere-packing problems in hyperbolic space have been studied in

[BezlJ, [Bez3], [Borl], [Coxl8, pp. 173-177], [Fejl, p. 325], [Fej4]-[Fej7],
[Fejl II.

Appendix. Planetary Perturbations

Consider a cluster of 12 unit spheres, the "planets", touching a given one,
the "sun", arranged as if they were in the fcc lattice. We shall show that
there is so much "play" that it is possible to achieve any permutation of
the 12 planets by cleverly rolling them around the sun in such a way that
they never overlap.
Figure 1.7 shows one way to prove this. Figure 1.7a shows the
12 points of contact when the planetary spheres are in the original
cuboctahedral arrangement of the fcc lattice. This cuboctahedron can be
rotated about a square face to achieve the permutation
11"1 = (0,10,3,11)(1,9,4,2)(5,6,7,8) .
If the planets are perturbed by rolling them in the directions indicated
by the arrows, they can be brought continuously into the icosahedral
arrangement of Fig. 1.7b. (A precise description of this perturbation can
be obtained from [Cox20, §8.4J.) The distances have now increased so that
the 12 planets do not touch each other. In this configuration call any
planet, say number I, the South pole, and its opposite (number 4), the
North pole, and regard each other planet as the "satellite" of the nearest
of these two polar ones, as indicated in Fig. 1.7c. Now move each satellite
towards its polar parent planet until they touch (as suggested by the
arrows in Fig. 1. 7c). The five satellites of the North polar planet can now
revolve around it to yield the permutation
11"2 - 0 , 6,5,9,10) .
The different choices for the poles give twelve 5-cycles like 11"2. We
shall show that these twelve 5-cycles generate the alternating group A12. In
Chap. II, §18 it is shown that their pairwise quotients generate the
Mathieu group M 12. Since M 12 is quintuply transitive there is an element
of it which transforms 11"2 into any desired 5-cycle, and the set of all 5-

W
10 10

9 3
9

~
2
8 2

7 7
(0 ) (b) (c)
Figure 1.7. How to permute the planets.
30 Chapter 1

cycles is well known to generate A12. Since we can also achieve the odd
permutation 1I'j, we can achieve the full symmetric group S12, as claimed.
By a slight perturbation of the satellite motions we can obtain the
twelve 5-cycles like 11'2 (which generate A12) even when the planets have
radius slightly larger than the sun. We do not know whether the whole
group S 12 can be achieved for such a radius.
 2
Coverings, Lattices and Quantizers
1. H. Conway and N. 1. A. Sloane

This second chapter continues the description of the questions motivating

this book. We first discuss the problem of finding the best covering of
space by overlapping spheres, a kind of dual to the packing problem. Then
we introduce the language of quadratic forms, show that lattices and
quadratic forms are really the same, and explain the connections with
number theory. One of the central issues is the classification of integral
quadratic forms or lattices. The last section describes the problem of
designing good quantizers or analog-to-digital converters. For each
problem we summarize what is presently known about its solution.

1. The Covering Problem

1.1 Covering space with overlapping spheres. We have already met the
packing and kissing number problems. The third main topic of this book is
a kind of dual to the packing problem, and asks for the most economical
way to cover n -dimensional Euclidean space with equal overlapping
spheres. Figure 2.1 shows two different ways to cover the plane with
overlapping circles. In (a) the centers of the circles belong to the square
lattice Z2, and in (b) to the hexagonal lattice. Clearly (b) is a more
efficient covering than (a), since there is less overlap among the circles.
To make this precise we define the thickness e of a covering in the
same way as the density A of a packing. Suppose an arrangement of
spheres of radius R covers Rn. If the centers form a lattice A then the
thickness is defined by formulae similar to Eqs. (10), (11), (20) of
Chap. I:
e= average number of spheres that contain a point of the space
volume of one sphere
(det A) 1/2

e is also called the density or the sparsity of the covering. None of these
32 Chapter 2

terms is completely satisfactory, and thickness seems the most descriptive.

The thickness of an arbitrary covering is defined in the same way as the
density of an arbitrary packing [Rog7, Chap. 11. We always have
A ~ 1 ~ 0. The normalized thickness (or center density) () is given by

(compare (26) of Chap. 1) . The lattice coverings in Fig. 2.1 have

thickness 0 = 11" /2 = 1.5708... and 0 = 211" /3.J3 = 1.2092... respectively.
Then the covering problem asks for the thinnest covering of n-
dimensional space by spheres, i.e. for the covering with minimal thickness.
Kershner showed in 1939 [Ker31 that no arrangement of circles can
cover the plane more efficiently than the hexagonal lattice arrangement
shown in Fig. 2.1 b. For a concise proof see [Fej 10, pp. 58-61l. (See also
[Aki 11, [Aki2J.) But, just as for packings, the optimal coverings are not
known in higher dimensions.
In three dimensions the best covering known, which Bambah [Bamll
showed is optimal among three-dimensional lattice coverings, is the body-
centered cubic lattice mentioned in §1.4 of Chap. I . This is at first sight
surprising, since the densest lattice packing is a different lattice, the fcc
lattice. In fact the bcc is the dual of the fcc lattice. To understand this
situation we must look further into the structure of these lattices, and
introduce some additional terminology.

(0)

(b)

Figure 2.1. Covering the plane with circles. In (a) the centers belong to
the square lattice Z2, in (b) they belong to the hexagonal lattice. (b) is a
more efficient or thinner covering.
Coverings, Lattices and Quantizers 33

1.2 The covering radius and the Voronoi cells. Consider any discrete
collection of points [lj> = {p b P 2 , . ..} in Rn. The least upper bound for the
distance from any point of Rn to the closest point Pi is called the covering
radius of [lj>, usually denoted by R. Thus
R = sup inf dist(x ,p). (3)
x ERn P E ilJ'

(If the upper bound does not exist we set R = 00.) Then spheres of radius
R centered at the points of [lj> will cover Rn, and no smaller radius will do.
Around each point Pi E .J}J is its Voronoi cell, V(Pi ), consisting of
those points of Rn that are at least as close to Pi as to any other Pj . Thus
V(Pi ) = { x ERn : dist(x ,Pi) ~ dist(x ,Pj ) for all j }. (4)

If P b P 2, ... represent schools, the Voronoi cells are the school districts!
Other names are nearest neighbor regions. Dirichlet regions. Brillouin
zones and Wigner-Seitz cells (the last two are physicists' terms). The
Voronoi cells of the hexagonal lattice, for example, are the regular
hexagons shown in Fig. 1.3c. The Voronoi cells of many other lattices are
described in Chaps. 4 and 21.
The interiors of the Voronoi cells are disjoint, although they have faces
in common. Each face lies in the hyperplane midway between two
neighboring points Pi' The Voronoi cells are convex polytopes whose union
is the whole of Rn. (It is true, but not entirely obvious, that the
intersection of two abutting Voronoi cells is an entire face of each of
them.) If [lj> is a lattice A, all the Voronoi cells are congruent and have
volume equal to (detA) 1/2.
The vertices of the Voronoi cells are especially interesting. They
include the points of Rn whose distance from [lj> is a local maximum: these
are called the holes in [lj> (cf. Fig. I.3d). If there is a point whose distance
from [lj> is an absolute maximum it is called a deep hole and its distance
from [lj> is the covering radius R. (The children who live at deep holes are
those that have the longest walk to schooL) Holes that are not deep are
called shallow. The deep holes in the hexagonal lattice are the points
marked band c in Fig. 1.3b. For this lattice R = 2pj.J3, where p is the
packing radius. If [lj> is a lattice, the vertices of the Voronoi cells are
exactly the holes, but in general they may include other points.
For a lattice packing A, with Voronoi cells congruent to a polytope V
say, the packing radius p is the inradius of V (the radius of the largest
inscribed sphere), while the covering radius R is the circumradius of V
(the radius of the smallest circumscribed sphere). We now see the
difference between the packing and covering problems. For a good packing
we try to maximize p, i.e. we wish to choose the centers of the spheres so
that the inradius of the Voronoi cells is as large as possible (for a
non lattice packing we consider the smallest inradius of any Voronoi cell).
On the other hand for a good covering we try to minimize R, i.e. we wish
to choose the centers so that the circumradius of the Voronoi cells is as
34 Chapter 2

small as possible (for a non lattice covering we consider the largest

circumradius of any Voronoi celt) .
Let us return to the three-dimensional question. The Voronoi cell for
the fcc lattice is a rhombic dodecahedron (Fig. 2.2a), one of the semi-
regular polyhedra [Cunl, p. 114], [Fej9], [Holl], [Loel, p. 42], [Wel4,
p. 731. If we choose the scale so that the lattice has determinant 1 (and
the Voronoi cell has volume 1) then the inradius and circumradius of the
r
Voronoi cell are p = 5/6 = 0.5612... and R = r
l /3 = 0.7937 ...
respectively. On the other hand the Voronoi cell for the bcc lattice is a
truncated octahedron (Fig. 2.2b) , one of the Archimedean polyhedra
[Cunl, p. 981, [Fej91, [Holll, [Loel, p. 1291, [WeI4, p. 731, [Wenl, p. 211.
When this lattice is scaled so as to have determinant 1 we find
p = r 5/3 3 1/2 = 0.5456 ... and R = r 5/ 3 5 1/2 = 0.7043 .... Thus although
the fcc lattice is the better packing, the bcc lattice is indeed a better
covering. There is another difference between these two lattices. In the
bcc lattice, as in the planar hexagonal lattice, there is only one kind of hole
(all holes are deep), but in the fcc lattice there are two kinds (shallow and
deep holes). See Fig. 2.2 and also Chap. 4, §§6.3, 6.7. This phenomenon
is particularly striking in the Leech lattice, where there are 23 kinds of
deep hole and 284 kinds of shallow hole (Chaps . 23 and 25) .
There are two other important concepts associated with the Voronoi cell
V(Pj) around a point Pj. The normalized second moment of V(Pj) is
defined to be

f
2
G = G(V(P)) = 1. Vol(V(p)) -1--;;- Ilx -PjIl 2dx .
n V(~)

(001)

(0) (b)
(-1/2 1/2 1/ 2)

(010)

(00-1)

Figure 2.2. (a) Rhombic dodecahedron (the Voronoi cell for fcc lattice)
centered at origin. There are 6 vertices (± 1,0,0), 8 vertices
(± V2, ± V2, ± V2) . (b) Truncated octahedron (the Voronoi cell for bcc
lattice) centered at origin. There are 24 vertices (± I, ± V2,0). For
example a = (O,lh,I), b=(0,1, V2), c =(V2 ,1 , 0), d-(I,V2 , 0) ,
e = (I,O,\l2),f = ( V2, 0,1) , g = (0,-V2 , 1), h = (-1/2, 0, I) .
Coverings, Lattices and Quantizers 3S

If the points Pi form a lattice A, then G will be denoted by G (A) . Apart

from the scale factors, G is the average squared distance that a child who
lives in V(Pi ) has to walk to school. (The second scale factor makes G a
dimensionless quantity,} G appears in Davenport's construction of efficient
coverings (§1.3), in studying quantizers (§3 .2), and in other applications.
The points Pi! • ... . Pir such that the hyperplane between Pi and Pij
contains a face of yep) with positive area are called Voronoi-relevant (or
just relevant) for Pi' These are the points actually needed to define the
Voronoi cell. For a lattice A the (Voronoi-)relevant vectors u E A are
those that are relevant for the origin, i.e. that are needed to define V(O) .
The following references discuss Voronoi cells: [Akill, [Aki2], [Ash3],
[Bar9], [Cha2], [Con28], [Con38], [Dirl], [Edel], [Fej9], [Howll, [1on4],
[Kell], [Kit4], [Kocll, [Lan2l, [Loel], [Maull, [Pre2], [Shu3], [YenS],
[Vorl], [Wanll, [Wigl] . Algorithms for computing Voronoi cells are
mentioned in §IA.
Before leaving this section we introduce the Delaunay cells associated
with the points f¥J. There is a Delaunay cell for each point that is the
vertex of a Voronoi cell; it is the polytope which is the convex hull of the
points of f¥J closest to that point [Cox26], [DelO], [Dell], [GalO], [Rog7].
(The early works of Boris Nikolaevich Delone were published under the
name Delaunay, and that spelling has come to be associated with these
polytopes,)
The Delaunay cells form a partition of Rn into convex regions that is a
kind of dual to the partition into Voronoi cells. Consider for example the
set f¥J indicated by the small circles in Fig. 2.3 . The Voronoi cells are
triangles such as XYZ, and there are two types of Delaunay cells: squares,

yl
-"'-
I

Figure 2.3. For the points fYJ indicated by small circles the Delaunay
cells are squares (A) and octagons (B); the Voronoi cells are triangles
(e.g. XYZ).
36 Chapter 2

centered at shallow holes such as A, and octagons centered at deep holes

such as B.
In the fcc lattice the Delaunay cells form a tessellation of alternating
tetrahedra and octahedra. For the bcc lattice the Delaunay cells are all
square bipyramids (i.e. squat octahedra).
1.3 Covering problem-summary of results. In this section we summarize
what is presently known about the covering problem (see Tables 1.1, 2.1
and Fig. 2.4). As mentioned in §l.l, the thinnest possible coverings are
only known in dimensions 1 and 2. The thinnest lattice coverings are
known in dimensions 1 through 5, and in each case the optimal lattice is
A; (the dual of An, and sometimes referred to as "Voronoi's principal
lattice of the first type" - see Chap. 4, §6.6). The results are due to
Bambah in 1954 for n = 3 [Baml], Delone (or Delaunay), Ryskov in 1963
for n = 4 [DeI4], and Ryskov and Baranovskii in 1975 for n = 5 [Rys12],
[Rys13]. (See also [Bam2], [Bar5], [Barl4], [Blel], [DeI2], [Dic4]-[Dic6],
[Fewl], [Gaml], [Gam21, [Kau1], [Wool].)
The covering associated with A; has thickness
e = v: .In+T { n (n +2) }n/2 (5)
n 12(n+I) '

and is in fact the thinnest covering known{[) in all dimensions n ~ 23. (It
is surprising that A;, with e = 3.6658 ... , is better than E 8, which has
e = 4.0587 ... .) But Ai4 has thickness 63.269 ... , and is inferior to the
Leech lattice A 24, for which e = 7.9035... (Chap. 4, §11, and [Con20] =
Chap. 23).
In three dimensions, A;
(the bcc lattice) is not just the thinnest lattice
covering, it is the only locally optimal lattice covering. In R4 Dickson
[Dic4] found that there are precisely three locally optimal lattice coverings,
namely A:
and the lattices with Gram matrices
2 a -1 -1 3-1' l' -1 -1
a 2 -1 -1 l' 3-1' -1 -1
Di 4a :
-1 -1 2 I-a
, Di 4b :
-1 -1 2+2{3 -{3 (6)
-1 -1 I-a 2 -1 -1 -{3 2+2{3

where a = (5 -.Jf3) /2 and {3 ::::: 0.544, l' ::::: 0.499 are the roots of certain
polynomials. In dimensions n = 5, 7, 9, ... Barnes and Trenerry
[Barl7] found locally optimal lattice coverings that are only slightly worse
than A;. Table 2.1 and Fig. 2.4 show the thinnest coverings known in
dimensions n ~ 24, together with the thicknesses of certain other lattices
of interest. The upper bound on the thickness is the Coxeter-Few-Rogers
bound (16), calculated via (17) and Eq. (40) of Chap. 1.

([) I t IS
. now known t h at A*·
23 IS a better covermg
. than A 23,
• and there are
other improvements to Table 2.1 and Fig. 2.4 in dimensions just below 24.
Cove rings, Lattices and Quantizers 37

:;

2 COXETEA
FEW
ROGERS
BOUND (16)

o 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
DIMENSION

Figure 2.4 . The thickness of various lattice coverings in dimensions

n ~ 24 . The values for AIJ to AI5 and AI7 to A 23 are lower bounds.
(They are computed using the subcovering radius of Chap. 6.)

In all dimensions n = 24£ + m ~ 24 (for 0 ~ m ~ 23) the lattice

A24 E& A24 E& ... E& A24 E& A~, with £ copies of A24 , has thickness
m/2 [ )n/2
0=V.Jm+l [m+2 ) ...!!.... (7)
n m +1 12'

and is a thinner covering than A;

[Bam3]. Even thinner coverings for
n ~ 25 are given in [Con43]. However all the preceding coverings have
the defect that, as n - 00,

-1 log2 0 - log2
n
-Jffe
--
12
= 0.2546 .... (8)

In 1952 Davenport [Dav2) found a method that yields thinner coverings

than (8) when n is large. Davenport's construction may be described as
follows. We start with a fixed k-dimensional lattice A having generator
matrix M say, and construct the km-dimensional lattice ff' with the
generator matrix shown in Fig. 2.5. Davenport showed that if £ and mare
large the thickness of ff' is bounded above by a quantity which is
essentially independent of £. For £ fixed (and large), and m, n - 00, the
thickness of ff'satisfies

I
- log2 0 < log2 .J27r eG (A) , (9)
n -

where G(A) is defined in §1.2. If A = Z, G(A) = 1/12 (see Eq. (88)

below), and (9) reduces to (8). Davenport took A - Ab calculated G(A k )
38 Chapter 2

Table 2.1. Coverings in up to 24 dimensions.

Name of Thickness Center density 8

n covering 8 Attained Bound
0 Ao 1 1 1
1 A; ;;= Z 1 0.5 0.5
2 A; ;;= A2 1.2092 0.3849 0.3849
3 A; ;;= D; 1.4635 0.3494 0.3419
A 3 ;;= D3 2.0944 0.5
4 A; 1.7655 0.3578 0.3360
Di 4b 1.89 0.382
Di 4a 1.93 0.391
D; ;;= D4 2.4674 0.5
A4 3.1780 0.644
5 A; 2.1243 0.4036 0.3581
BT5 2.2301 0.4237
D; 2.4982 0.4746
D5 4.5977 0.8735
A5 5.9218 1.125
6 A: 2.5511 0.4937 0.4087
D: 4.3603 0.8437
7 A; 3.0596 0.6476 0.4949
BT7 3.2441 0.6865
£; 4.1872 0.8862
D; 4.5687 0.9670
8 A; 3.6658 0.9032 0.6319
£8 4.0587 1
D; 8.1174 2
9 A; 4.3889 1.331 0.8460
BT9 4.6569 1.412
D; 8.6662 2.627
10 A;o 5.2517 2.059 1.183
11 A;, 6.2813 3.334 1.721
12 A;2 7.5101 5.624 2.597
KI2 17.7834 13.318
13 A;3 8.9768 9.858 4.055
14 A;4 10.727 17.90 6.537
15 A;5 12.817 33.60 10.86
16 A;6 15.311 65.06 18.56
17 A;7 18.288 129.7 32.57
18 A;8 21.841 265.9 58.63
19 A;9 26.082 559.4 108.1
20 A;o 31.143 1207 204.0
21 A;, 37.185 2666 393.5
22 A;2 44.395 6023 775.2
23 A;3 53.000 13908 1558
24 A24 7.9035 4096 3193
A;4 63.269 32789
Coverings, Lattices and Quantizers 39

0
M

I- 1M I- 1M I-1 M

Figure 2.5. Davenport's construction of thin lattice coverings. This

lattice contains the direct sum of m copies of the starting lattice A, whose
generator matrix is M.

(see Chap. 21, Eq. (26», finding that it is minimized when k = 8. This
produces lattices ff' with
1
- log2
n
e <- 0.2012 .... (10)

Ryskov [Rysll showed that lattices obtained from Davenport's construction

(taking A=A 2) are thinner coverings than A; for all n ~ 200.
Watson [Wat2] used A == E8 and obtained
1
- log2
n
e <- 0.1460... . (11)

(Watson's calculation of G (E 8) does not quite agree with the value

929112960 we find in Chap. 21, but the discrepancy is so small that it has
no effect on (11).)
Finally, by taking A to be the Leech lattice A24, and using our
numerical estimate [Con38]
G (A 24) - 0.065771 ± 0.000074, (12)

we obtain coverings with thickness satisfying

1
- log2
n
e <- 0.084.... (13)

These appear to be the most efficient coverings known.

On the other hand these are all extremely poor coverings when
compared with the existence results. For Rogers showed in [Rog3] that
lattice coverings exist with
40 Chapter 2

for some constant c, where a = V2 log2(21re), and in [RogIl that (possibly

non lattice) coverings exist with
8 ~ n loge n + n loge loge n + 5n , (15)

for n ~ 3 (see also [Erd21, [Grill], [Rog7]). As usual these results are
nonconstructive, in the sense explained in Chap. 1.
In the other direction Coxeter, Few and Rogers [Cox261 showed that
any n-dimensional covering satisfies
(16)

This is a companion to Eq. (39) of Chap. 1. Again let S be a regular

simplex of edge 2. Spheres of radius {2n/(n + 1) }l/2 centered at the
vertices of S just cover S, and Tn is the ratio of the sum of the volumes of
the intersections of these spheres with S to the volume of S. Thus
n/2
T = [~] (17)
n+1
0-
n n

as n - 00. (Is)

In summary, from (l5)-(lS), the thinnest coverings satisfy

n
,- ::S 8 ~ n loge n + n loge loge n + 5n . (19)
eve

Remarks. (j) These bounds are considerably closer together than the
corresponding bounds for packings (Chap. 1, Eq. (46)); on the other hand
the best constructions known are further from the bounds. (ij) There are
no known examples of nonlattice coverings that are thinner than the best
lattice coverings. (iii) One may also consider the covering problem on the
sphere n n (analogous to Chap. 1, §2.3). There are some results in [Fej 1,
p. 325], [Rog61, [Wyn21, but not much is known about this problem.
(iv) Davenport's construction can be adapted to produce binary codes with
low covering radius [Gra4].

1.4 Computational difficulties in packings and coverings. Given a generator

matrix for an n -dimensional lattice A we can find its determinant from
Eq. (4) of Chap. 1. But anything else is hard! For an arbitrary lattice A
the best algorithms known for finding either the packing radius p or the
covering radius R require a number of steps that grows exponentially with
n [Diel], [Finl], [He13], [Hel41, [Kan21, [PohIl The problem of finding
the covering radius is known to be in the class of NP-hard problems
[Emdl], and finding the packing radius is conjectured to be in this class.
Evidence for this conjecture is supplied by the fact that certain closely
related coding theory problems are NP-hard [BerS]. Finding the covering
radius of a code is also known to be NP-hard [McL2].
Coverings, Lattices and Quantizers 41

Lenstra, Lenstra and Lovasz [Lenl1 describe an algorithm (the "LLL

algorithm") which finds lattice vectors of moderately small length in a
number of steps that grows only as a polynomial function of n. The short
vectors in A have length 2p; the LLL algorithm finds lattice vectors that
are guaranteed to have length ~ 2(n+!)/2 p. For modest values of n the
algorithm does much better than this and very often finds a vector of
minimal length. It has been used on lattices of dimension as large as 100
in connection with breaking certain encryption schemes [Lag4], [Od14].
The LLL algorithm actually converts the given basis for the lattice into a
"reduced basis". One of the reduced basis vectors is then a short vector in
the lattice. Many other reduction algorithms are known; the classical
Minkowski reduction algorithm is briefly described in Chap. IS, §1O.1.
Reduction algorithms are discussed in [Affl1, [Babl1, [Barll]-[Barl3],
[Dic7], [Donl1, [GruI], [Hel3], [HeI4], [Lag3], [Minl1, [Novl1, [Rys2],
[Rys3], [RysS], [Rysl4], [Sch4], [Stol], [Taml1-[Tam4], [Wae3], [Wae5].
(For three-dimensional lattices see also the crystallographic references
mentioned at the end of this section.> Babai [Babl1 has shown that the
LLL algorithm can be used to find an upper bound on the covering radius
that is ~ 3n R.
There is of course no practicable algorithm known for finding the best
lattice packings or coverings of a given dimension. Other algorithms
related to the classification of lattices and quadratic forms are discussed in
Chap. IS, §11.
The situation improves when the lattices have some algebraic structure.
For the root lattices (Chap. 4, §2) and some related lattices we are able to
calculate the exact value of R and find the Voronoi cells in Chap. 21. In
some other special cases R can be found by Simon Norton's method
([Nor4] == Chap. 22, [Con37], [Con3S], [Con43]). Otherwise if n is small
one may try to find the Voronoi cells first and then obtain R as the
distance of the furthest vertex of a Voronoi cell from the lattice.
Algorithms for computing the Voronoi cells of arbitrary sets of points
are described in [Aki2], [Bow2], [Br03], [DevI], [Dev2], [FinS], [Leel],
[Pre2], [Sei41, [WaU]. References [Con29] ;;;; Chap. 20, [Con3S],
[Con3S], [Con40], [For2] describe algorithms for reasonably nice lattices
that, given an arbitrary point x E Rn , identify which Voronoi cell x
belongs to, i.e. find the closest lattice point to x. Such algorithms are
important for the applications described in §3 below and §I of Chap. 3.
There are many algorithms used by crystallographers for classifying
periodic structures in three dimensions - see for example [Ahml1, [BurI],
[Bur2], [Ferl], [HaIS], [Koc2], [PaU], [Sanl], [Say2], [Zim2]. See also
[Coh4a], [Spel1.

2. Lattices, Quadratic Forms and Number Theory

2.1 The norm of a vector. When studying lattice vectors
x = (X\>X2, . . . • x n ) E A it is generally easier to work, not with their
length, but with their squared length or norm. This is denoted by
N(x) =X·X = (x,x) = ~x?
42 Chapter 2

The minimal squared distance between distinct lattice vectors, or more

simply the minimal norm of A, is
min { N (x - y) : x,y E A, x ~ Y } (20)
= min ( N (x) : x E A, x ~ 0 }. (2 1)

If the minimal norm is IJ., the packing radius of A is given by

p = '/2 .J;.. (22)

Formulae (20) and (21) are equivalent definitions for the minimal norm of
a lattice, for if x and yare distinct vectors in A then x' = x - y is a
nonzero vector of A. (20) and (22) - but not necessarily (21) - are also
valid for nonlattice packings.
2.2 Quadratic forms associated with a lattice. Quadratic forms provide an
alternative language for studying lattices, especially useful for investigating
arithmetical properties. In this section we show how quadratic forms and
lattices are related. Some of the material is rather detailed, and the reader
may like to turn at once to §3 where we discuss the quantization problem.
The classification of quadratic forms and lattices is briefly discussed in
§2.4; we shall say much more in Chaps. 15-18.
Let A be a lattice in n -dimensional space Rn, having basis vectors
(forming the rows of a generator matrix M). The generic
VI • . . . • Vn
lattice vector x = (x I • . . . ,xn ) E A may be written (see Eq. (2) of
Chap. 1) as

where the ~i are integers and ~ = (~I' .... ~n). The norm of this vector is
n n
N(x) = N(~lvI+ ... +~nvn) = ~ ~ ~i~i Vi· Vi
i-I i-I

(24)

where A = MM 1r is a Gram matrix for A. Regarded as a function of the

n integer variables ~I • . . . • ~n' f(~) is a quadratic form associated with
the lattice.

For example, from the Gram matrix given in Eq. (7) of Chap. 1, one
quadratic form associated with the hexagonal lattice is
H + ~Ib + ~r (25)

The n-dimensional cubic lattice Zn (Chap. 4, §5) has generator matrix In

(the identity matrix), and the corresponding quadratic form is
~?+~1+ ... +~~. (26)
Coverings, Lattices and Quantizers 43

It is sometimes simpler to use the ~-coordinates (~l' ...• ~n) for lattice
vectors rather than the x-coordinates. For instance if we start with the
quadratic form f we may define the lattice by setting the norm of
(~l' ...• ~n) equal tof(~). (Or we may factor A as MM tr and use Mas
a generator matrix.)
We recall that congruent lattices are related by Eqs. (22) and (23) of
Chap. 1 with c '"' 1 and det U = ± 1. Quadratic forms corresponding to
congruent lattices are called integrally equivalent. Thus there is a one-to-
one correspondence between congruence classes of lattices and integral
equivalence classes of quadratic forms. For example the vectors VJ, V2 in
Fig. 2.6a and WJ, W2 in Fig. 2.6b also span lattices that are congruent to
the hexagonal lattice of Fig. 1.3a. The Gram matrices for these bases are

[_I/~ -II;] and [3~ 3/;] respectively,

and so the corresponding quadratic forms

~t - ~1~2 + ~i and 3~t + 3~lb + ~i (27)

are integrally equivalent to each other and to (25).

If A is a lattice of full rank, i.e. a lattice in n-dimensional space that is
spanned by n independent vectors, then M has rank n, A is a positive
definite matrix, and the associated quadratic form is called a positive
definite form.
Indefinite quadratic forms (those for which A is an indefinite matrix)
may also be studied via lattices, although these are no longer lattices in
Euclidean space Rn. One of the most interesting cases occurs when A has
signature (n - 1, 1), so that the corresponding lattice is in Lorentzian (or
hyperbolic) space. We shall say much more about this in Chaps. 15, 26
and 27.
In view of the equivalence between lattices and quadratic forms, all the
geometrical problems described in Chaps. 1 and 2 have corresponding
formulations in terms of quadratic forms. For example, finding the

C
0

0
VI
0
;> ·0
WI
0 0
0 0
0

(0) (b)

Figure 2.6. Two congruent versions of the hexagonal lattice.

44 Chapter 2

minimal norm of a lattice is equivalent to finding the minimal value of

f (~) over all integral ~ ~ 0, the so-called homogeneous minimum of the
quadratic form [Cas21 Similarly the inhomogeneous minimum of the
form is the square of the covering radius of the lattice. A locally optimal
lattice packing, i.e. a lattice A whose density Ll does not increase when A
is perturbed slightly, corresponds to an extreme form [Bac2], [Bar4],
[Bar6]-[BarlO], [Con42], [CoxlO], [Kne21, [RysI5]. We know all the
extreme forms in at most 6 variables [Bar6], [Bar71, [Hon]. Extreme
forms in 7 variables have been studied in [Bar9], [Con42], [Larl], [Scoll,
[Sc02], [Shu4], [Sta 1], [Sta2l. A quadratic form corresponding to a
densest lattice packing is called an absolutely extreme form. The results
described in Chap. 1, §1.5 mean that the absolutely extreme forms are
known for n ~ 8.
Much of the research into lattices has been carried out in the language
of quadratic forms. However in this book we usually prefer to study
lattices geometrically, since the choice of a particular set of basis vectors is
often arbitrary and conceals the symmetry of the situation. For example
the simplest quadratic form for the E 8 lattice is

~[ - ~1~2 + ~i - bb + ~}- ~3~4 + ~J - ~4b

+ ~~ - ~5~6 + ~l- ~6~7 + ~? - ~5~8 + ~j , (28)

obtained from the basis given in Chap. 4, Fig. 4.7. The definition given in
Chap. 4, Eq. (97), reveals much more of the symmetry.
Nevertheless quadratic forms are important for studying certain aspects
of lattices, and there is an extensive literature - see the references in §1 of
Chap. 15. Quadratic forms are particularly useful for investigating
arithmetical aspects of lattices, as we now show.
2.3 Theta series and connections with number theory. A very old problem
asks for the number of ways of writing an integer m as a sum of four
squares, or in other words for the number of quadruples of integers
(XJ,X2,X3,X4) such that

X[ +xi +x} +xJ =m. (29)

For example when m = 2 there are 24 solutions, conslstmg of all

permutations of (± 1, ± 1,0,0). (We agree to count 2 = 12 + 12 + 0 2 + 0 2,
2 = 12 + (_1)2 + 0 2 + 0 2, 2 = 12 + 0 2 + 12 + 0 2 etc. as distinct
solutions.)
There is a nice way to state this problem in terms of lattices. For any
lattice A, let N m be the number of vectors x E A of norm m, i.e. with
x . x = m. From (24), N m is also the number of integral vectors ~ such
that
(30)

or in other words the number of times the quadratic form associated with
A represents the number m. Eq. (30) is an example of a Diophantine
Coverings, Lattices and Quantizers 45

equation of degree 2 [Mor61. Now Xf + + + xi xi xi

is the quadratic
form associated with the four-dimensional cubic lattice Z4. So the number
of ways of writing m as a sum of four squares is equal to the number of
vectors of norm m in the lattice Z4.
The calculation of these numbers is facilitated by introducing the theta
series of a lattice A, which is
8 A (z) = ~ qX-X (31)
xEA

where q = e 1riz • For many purposes we can just think of 8 A as a formal

power series in an indeterminate q, although for deeper investigations we
must take q = e 7fiz , where z is a complex variable. In this case 8 A (z) is
easily seen to be a holomorphic function of z for Im(z) ~ 0 [GunI, p. 711,
[Serl, p. 1091.
We can also use (3 I) to define the theta series of a nonlattice packing
A. The commonest examples of this occur when A is a translate of a
lattice, or a union of translates (e.g. Eq. (35».
For a periodic packing @J, consisting of a union of say s translates
Uj + A, j = I, ... ,s,

of a lattice A (compare Eq. (13) of Chap. I), we define the average theta
series to be

= 8 A (z) + 1. ~ ~ qN(x+urUk)
S j<kxEA

[OdI61. If @J is distance invariant (cf. [Mac6, p. 40]), i.e. has the

property that the number of points of @J at any distance d from x E @J is
independent of x, then 8 r?,(z) reduces to

8 f?i(z) = ± ~
j-I xEA
qN(X +urUt) . (34)

For example, the theta series of the integers Z is

m-- oo

which is the Jacobi theta function (J3(Z) (Chap. 4, §4.1). The translate
Z + Ih = L, -lIh, -Ih, Ih, lIh ,.. .} has theta series

8 Z + I / 2(Z) = ~ q(m+I/2)2 = 2ql/4 + 2 q 9/4 + 2q25/4 + ... , (35)

m-- oo
46 Chapter 2

which is the Jacobi theta function 8 zCz).

We now see one of the advantages of using the norm N (x) = x . x in
Eq. (31). Because N( (Xi> ... ,xn )) = xl + ... + x; = N(Xl) +
+ N(x n ), we have
8 z n(z) = 8 z (z)n = 8 3(z)n.

So the answer to the problem at the start of this section is simply the
coefficient of qm in the expansion of 8 3 (z)4 in powers of q.
To obtain the theta series of the lattice Dn (which consists of the points
of Zn whose coordinates sum to an even number, Chap. 4, §7.1) we
introduce the theta function 8 4 • 8 3 and 8 2 may be regarded as assigning
unit masses to the integer and half-integer points on the real line
respectively. To get 84 we assign masses of +1 to the even integers and -I
to the odd integers, or formally

m-- oo

Then

(37)

We cannot resist giving two more examples. It is not difficult to show

that the translate (Yz, ... , Vz) + Dn has theta series V28 2(z)n. The
packing
(38)

is the diamond packing in three dimensions and the E 8 lattice in eight

dimensions. (See Chap. 4, §7.3. Dn+ is a lattice if and only if n is even,
and Dt is congruent to Z4.) Thus we have the appealing formulae
([SloI7])

8diamond(Z) = Vz (8 2(z)3 +8/z)3 + 84 (z)3 ) , (39)

8£8(Z) = V2 (82(z)8+83(z)8+84(z)8). (40)

Many other theta series will be given in later chapters. We calculate

the theta series of the hexagonal lattice in Chap. 4, §6.2, to illustrate one
general method of obtaining the theta series from the quadratic form. A
more general theta function that includes 8 2 , 8 3 and 8 4 as special cases is
the Jacobi theta function 8 3 (~ I z) (Chap. 4, Eq. (6)). However most of
the lattices we encounter can be handled using 8 2, 8 3 , 8 4 and other simpler
functions defined in §4.1 of Chap. 4.
We note that the theta series of a lattice tells us the packing radius p,
the kissing number T and the density Ll, since
8 A (z) = 1 + Tq4p2 + ... , (41)
Coverings, Lattices and Quantizers 47

(42)

We also obtain the theta series of the dual lattice A* by a simple

substitution:

(43)

- see Chap. 4, Eq. (19). Here for the first time we must use z rather
than q. There is also a generalization of (43) which applies to nonlattice
packings that are the union of translates of a lattice [OdI6]. Two
illustrations are given in Example 6 of Chap. 7.
The actual values of the theta series are also of importance in chemistry
(for example in calculating the Madelung constant, see [Bor6], [Gel2],
[Gla7], [ala8], [Zucl]), and in communication theory. The error
probability when a lattice is used as a code for a particular Gaussian
channel can be estimated by the value of e A (q) for a particular choice of
q - see Eq. (3S) of Chap. 3.
The theta series is determined by the lattice, but not conversely. Witt
[Wit4] observed that in 16 dimensions Dt6 and the direct sum E8 EB E8
are inequivalent lattices of determinant 1 with the same theta series. It is
easy to verify this, and the verification illustrates how identities between
theta functions have lattice-theoretic proofs. First, since Dt is congruent
to Z4, we have the identity 1f2(8 2 (z)4+8 3 (z)4+8 4(z)4) = 8 3 (z)4, i.e.
8 3 (z)4 = 8 2 (z)4 + 84(z)4. (44)

r
We must show that the theta series of Di6 and E8 EB E8 are equal, i.e. that

~ [8 2 (z) 16+8 3(z) 16+8 4(Z) 16) = ! [8 2 (z)8 +8 3 (z)8 +8 4(z)8

(from (37)-(40». This follows immediately by eliminating 8 3 (z) using

(44) and expanding both sides.
Similarly Kneser [KneS] observed that D12 and D4 EB E8 are
inequivalent l2-dimensional lattices of determinant 4 with the same theta
series. (This identity also follows immediately from (44).) Later Kitaoka
[Kit2] gave an example of two 8-dimensional lattices of determinant 81
with the same theta series, and examples are now known in dimension S.
In dimension 2, on the other hand, the theta series determines the lattice
uniquely (see for example [Wat14]). (Some related references are [HsiS],
[Hsi7], [Kitl], [Lil].)
2.4 Integral lattices and quadratic forms. The class of integral lattices
includes many important examples. In this book we shall call a lattice or
quadratic form integral if the inner product of any two lattice vectors is an
integer, or in other words if the Gram matrix A has integer entries. (The
term classically integral is sometimes used - see the discussion of
48 Chapter 2

integrality notions in Chap. ISJ Equivalently, a lattice A is integral if and

only if
(4S)

and for many geometrical purposes this is the best form of the definition.
Most of the lattices encountered so far in this chapter are integral when
suitably scaled. Our general policy (for example when selecting basis
vectors in Chap. 4) is to choose the scale so as to make the determinant as
small as possible while making the lattice integral. For example the
hexagonal lattice A2 as defined by Eq. (8) of Chap. I is integral, although
the equivalent lattice defined by (6) is not.
An important group associated with an integral lattice A is its dual
quotient group A* lA, which has order det A. For example A;IA 2 is a
cyclic group of order 3. The three cosets of A2 in A; are indicated by the
letters a, b, c in Fig. 1.3b.
Note that an integral lattice A has the property

A ~ A* c I A (46)
- det A .

(For if xEA*, x=~(M-l)tr= ~(M-l)trM-1M=(detA)-I~adj(A)M

where~, ( E zn and adj(A) is the adjoint of AJ
= (det A)-I(M,

An integral lattice with Idet A I = I, or equivalently with A = A *, is

called unimodular or self-dual. (We have written Idet AI rather than
det A so that the definition also applies to Lorentzian latticesJ If A is
integral then x . x is necessarily an integer for all x E A. If x . x is an
even integer for all x E A, then A is called even; otherwise odd. Even
unimodular lattices (also called Type II lattices) are especially interesting.
E8 and A24 are even unimodular lattices, while Z, Z2, Z3, ... are odd
unimodular (or Type 0 lattices.

It is known that if a unimodular lattice has the property that the norm
of every lattice vector is a multiple of some positive integer c, then c is
either I or 2 [O'Mel, p. 3241.

The classification of odd and even unimodular lattices is an important

problem in number theory and other parts of mathematics [Bri21, [00111,
[Hazll, [Hir41, [Nik21, and the known results are summarized in
Table 2.2. Even unimodular lattices exist if and only if the dimension is a
multiple of 8 (Cor. 18 of Chap. 7), while odd unimodular lattices exist in
all dimensions. E 8 is the unique even unimodular 8-dimensional lattice,
and E 8 ffi E 8 and D t6 are the only two such 16-dimensional lattices
[Wit41. The even unimodular 24-dimensional lattices were enumerated by
Niemeier [Nie21, who found that there are 24 such lattices, 23 with
minimal norm 2 and one, the Leech lattice A 24 , with minimal norm 4 -
see Table 16.1 of Chap. 16. Three separate verifications of Niemeier's
result will be given in Chaps. 16 and 18.
Coverings, Lattices and Quantizers 49

Table 2.2. The number of n-dimensional unimodular lattices. an is the

number of unimodular lattices of dimension n containing no vectors of
norm 1. If n == 0 (mod8), an is written as d n + en, where d n is the
number of odd lattices and en is the number of even lattices. Similarly bn
is the total number of unimodular lattices of dimension n (including those
with vectors of norm I).

n 0 1 2 3 4 5 6 7 8
an 0+1 0 0 0 0 0 0 0 0+1
bn 0+1 1 1 1 1 1 1 1 1+ 1

n 9 10 11 12 13 14 15 16 17
an 0 0 0 1 0 1 1 1+ 2 1
bn 2 2 2 3 3 4 5 6+2 9

n 18 19 20 21 22 23 24 25 26
an 4 3 12 12 28 49 156 + 24 368 ?
bn 13 16 28 40 68 117 273 + 24 665 ?

Kneser [Kne4] (see also [HerO], [Kol], [Mor3], [Smi6D enumerated

the odd unimodular lattices in dimensions n ~ 16, the authors ([Con27],
[Con34], Chap. 16) extended this to n ~ 23, and Borcherds ([Borl1,
[Bor3] = Chap. 17) recently enumerated the 24- and 25-dimensional odd
lattices. If A is a unimodular lattice so is Zi EB A for all i, so it is simplest
to record how many lattices are not of the form Z EB A' for some A', i.e. do
not contain a vector of norm 1. The numbers (an) of lattices of this type
are given in the first row of Table 2.2, while the second row (b n ) gives the
total number. If n is not a multiple of 8 the table simply gives the number
of odd unimodular lattices; if n == 0 (mod 8) the numbers are written to
show the division into odd and even lattices. For 1 ~ n ~ 8 there is only
one odd unimodular lattice, Zn, while for n = 9, 10, 11, ... there is also
Es EB Zn-S. In dimension 12 a third odd lattice (D(2) appears, and so
there are three odd unimodular 12-dimensional lattices, Z12, E s EB Z4 and
D (2. This explains the first twelve columns of Table 2.2. The enumeration
is extended to dimension 24 in Chaps. 16 and 17. See also the remarks on
extremal self-dual lattices in Chap. 7.
The Minkowski-Siegel mass formulae provide a powerful check that
these enumerations are correct. Let IAut (A) I be the order of the
automorphism group of a lattice A (defined in §4.1 of Chap. 3). The mass
formula gives an explicit constant ao such that

(47)

where the sum is over all inequivalent lattices from a given genus (for
example all n -dimensional odd unimodular lattices). A slightly more
general formula gives an explicit formula for

tI
0 A (z)
Aut(A) I .
so Chapter 2

We illustrate by using the mass formula to check that the first nine
columns of Table 2.2 are correct. First we consider even unimodular
lattices. In dimension 8 the constant ao = 1/696729600 is given in
Chap. 16, Table 16.4, and indeed
1
IAut(E 8) I 696729600

(see Chap. 4, §8.I). The values of ao for odd unimodular lattices are given
in Chap. 16, Table 16.2. For 1 :s:; n :s:; 8, ao = 1/(2n nO, confirming that
zn is the unique odd unimodular lattice (since IAut(zn)I = 2n n!, Chap. 4,
§5). However, for n = 9, ao = 17/2786918400, and indeed there are two
9-dimensionallattices, E8 ED Z and Z9, and
1 + 1
I Aut(E 8 ED Z) I IAut(Z9) I

__-"-__ + _1_ = 17 (48)

2·696729600 29. 9! 2786918400 .

We shall say more about these formulae in Chap. 16. The most dramatic
application of the mass formulae is the verification of Niemeier's list of
24-dimensional even unimodular lattices given in Chap. 16, §2.

The mass formulae also provide a lower bound on the number of

inequivalent lattices of each dimension (since IAut(A) I ~ 2 in (47». It
follows from Chap. 16, Table 16.4, for example that there are at least
80000000 distinct even unimodular 32-dimensional lattices, so it is
extremely unlikely that Niemeier's work will ever be extended to 32 or
higher dimensions.

This book also has results on the classification of (positive definite)

quadratic forms of other small determinants (2,3, .. J in Chap. 15, and
contains some information about indefinite forms (Chaps. 15,26,27).

2.S Modular forms. Further connections between lattices and number

theory arise because the theta series of an integral lattice is a modular
form. We do not state the general theorem here; some important special
cases will be foand in Chap. 7, Theorems 7 and 17. There is a brief
account of the general theory of the connections between quadratic forms
and modular forms in Cassels [Cas3, p. 382], and numerous other
references: [GunI], [Har41, [Hec2], [Hec3], [Kit3], [Knol], [Lan91,
[OggI], [Pet4], [Ran6], [Sch6], [Sch7], [VigIl

The main consequence of this theory is that there are only a limited
number of possibilities for the theta series when the determinant is small.
From this it is sometimes possible to derive explicit expressions for the
coefficients N m in the theta series, or to obtain accurate numerical
estimates for the coefficients. We shall give five samples of such results;
many others will be found in the above references.
Coverings, Lattices and Quantizers 51

(i) In dimensions n ~ 7, as we have mentioned, the lattice Zn with

quadratic form (26) is the unique unimodular lattice. From this it follows
that when n is even there are simple formulae for the coefficients of the
theta series, or in other words for the number of ways of writing a number
as a sum of n squares. For example the final answer to our original
questions is that the number of ways of writing m as a sum of four squares
is equal to
(49)

(cf. Chap. 4, Eq. (49».

(iD There are many other beautiful identities that arise in connection
with the particular modular forms known as Eisenstein series, whose
coefficients involve the arithmetical function (J k (m), the sum of the k th
powers of the divisors of m. For example the theta series of E g, Eq. (40),
is also equal to

0 E8 (Z) = I + 240 ~ (J3(m)q2m

m-l

= I + 240 q2 + 2160q4 + 6720q6 + (50)

- see Chap. 4, §8.1 and Chap. 7.

(iii) Let us define

m-l m-O

= q2 _ 24q4 + 252 q 6 - 1472q 8 + ... , (51)

in which the function T is called Ramanujan's function. It now follows

from modular form theory that there is a simple expression for the
coefficients of the theta series of the Leech lattice A 24 • Let
00

0 A1A(Z) = ~ N m qm
m-O

= I + 196560q4 + 16773120q6 + (52)

Then, for m > 0,

(53)

- see Chap. 4, §Il.

(iv) The second term on the right-hand side of (53) is much smaller than
the first. In fact it is known from Deligne's work [Degll, [Deg21, [Kat3]
that
(54)
52 Chapter 2

where d(m) is the number of divisors of m, and so

rem) = O(m 11 /2+,) (55)

for every positive L Since 0" 11 (m) > m 11,

Nm :::::
65520 0"11 [m2]
"69l (56)

is a very good approximation. This is typical of the estimates that can be

obtained for the coefficients of theta series (see for example [Ran6, §4.51,
[Ran7], and §1.4 of Chap. 3), at least for dimensions n ~ 4.
(v) The theta series of any even unimodular lattice can be written as a
polynomial in A24(Z) and the theta series of E& (Th. 17 of Chap. 7).
Consider for example the 48-dimensional lattices P 4&p and P 4&q mentioned
in Table 1.3. Knowing only that they are even unimodular lattices of
minimal norm 6 we can immediately deduce from this theorem that both
have theta series equal to
0 E8 (Z)6 - 1440 0E8(Z)3A24(Z) + 125280 A 24 (z)2
= 1 + Oq2 + Oq4 + 52416000q6 + 39007332000q& + (57)

The kissing number 52416000 is obtained with almost no extra work!

Further information about this and similar examples will be found in
Chaps. 7 and 19.
2.6 Complex and quaternionic lattices. The lattices defined so far in this chapter
are subgroups of real n-dimensional space Rn that are generated by n linearly
independent vectors. They are closed under the operations of addition, subtraction,
and multiplication by integers m E Z, or in other words are Z-modules. There are
several possible generalizations, for example by using complex or quaternionic
vectors instead of real vectors. These more general lattices are of interest in their
own right, and sometimes lead to simpler constructions of real lattices. They are
introduced in this section; further examples will be found for example in Chaps. 7,
8 and in [Cas3, Chap. 71, [ChaJ], [Con36], [Hsi6al, [Karl], [Lew!], [O'Mel],
[Ongl], [Ong21, [Otrl], [Que41, [Que6], [Rog91, [Rogll], [Ser21, [Smilil.
Let C denote the field of complex numbers {z = X + iy : x,y E R}, where
i 2 = -I, and let H denote the (skew) field of quaternions {z =u +iv + jw +kx :
u,v,w,x E R}, where

i 2 = / = k 2 = -I ,
ij=-ji=k, jk=-kj=i, ki=-ik=j

z
([Birl], [Cox8], [Cox!71, [Herl]). The conjugate of z is defined to be x - iy
(in C) or u - iv - jw - kx (in H), and the norm of z is N (z) = zz, which is
x 2 + y2 or u 2 + v 2 + w 2 + x 2 . Similarly the norm of a vector z = (z \' ... ,zn) is
N(z) = z\z\ + ... + znzn.

Besides changing the field we also change what we mean by an integer. There
are many possibilities, but here we shall just consider three rings of integers to be
used instead of Z. These are the Gaussian and Eisenstein integers
Coverings, Lattices and Quantizers 53

~ = { a + ib : a, b E Z} c C, (58)
8-{a+wb : a,bEZ} CC, (59)

where w - (-I +i.J])/2 is a complex cube root of unity [CoxIO, p. 421), [Cox21,
p. 145), [Fei2), [Har5, p. 179), [LanO, III, pp. 5, 16), and the Hurwitz
quaternionic integers [Hurl), [CoxI8)
:Yf- {a +ib + jc +kd : a,b,c,d E Z or a,b,c,d E Z + 'h} C H. (60)

Let J be one of the rings of integers Z, ~, It, .,r, and let K be the
corresponding field (R, C, C, H respectively). A J-lattice is defined as follows.
Take n vectors VI • • • • • V, E K' that are linearly independent over K. Then the
J -lattice A generated by these vectors consists of all linear combinations
(61)

where EI ....• E, E J. Since J is a ring, A is closed under addition, subtraction

and multiplication (on the left, in the quaternionic case) by elements of J, i.e. is a
J-module.
Aside. Other choices for J that lead to interesting examples include the rings of
algebraic integers { a +b8 : a,b E Z }, where 8 is one of
~
v-2,
r--7
v-5,
-1+H -1+.J=iT 1+.J5 (62)
222

or the ring of cyclotomic integers Zm where t - e 2,,;/m (see Chap. 8, [Bayl),

[Con36), [Cosll, [Cra2)-[Cra5), [Que6l). However, one must be careful. If J is
not a principal ideal domain (e.g. when 8 - H) A may be n-dimensional but not
have n generators, i.e. may not be a free J -module; these are called non-principal
lattices. If J contains an irrational real number (as in the last example in (62»
then A is dense in K', i.e. is not a (discrete) lattice. Nevertheless these examples
can still be used to produce lattice packings by redefining the norm or the
associated quadratic form (see e.g. §4 of Chap. 8). By restricting ourselves in this
section to the special rings mentioned above we avoid these complications.
Let A be a J -lattice. Just as in the case of real lattices we may now define a
generator matrix M (Eq. (1) of Chap. 1) and Gram matrix A - MM' (cf.
Chap. I, Eq. (3». The main difference from the real case is that the inner
product of vectors x = (Xl •... • x,), y = (yl •... • y,) is now defined to be the
hermitian inner product
(63)

The determinant of A is given by Eq. (4) of Chap. 1, the minimal norm IJ. by
Eq. (21) above, the packing radius p by Eq. (22), the covering radius by a formula
analogous to Eq. (3), the theta series by
f)A(Z) - ~ qNGc), q = e";z, (64)
xEA

(cf. Eq. (31)), and the dual lattice by

A' - { X E K' : X· ii E J for all u E A } (65)

(cf. Chap. 1, Eq. (24». A is integral if A ~ A', and unimodular or self-dual if

A = A'.
54 Chapter 2

If A is a complex J -lattice in en there is a corresponding real lattice Areal in

R2n consisting of the vectors
(66)

for (ZI • . . . • zn) E A. Also

(67)

where a,
the absolute value of the discriminant of J, is equal to 3 if J = tf or 4 if
J = '!i. The theta series of the dual lattice is

(68)

If A is a quaternionic Jff..lattice in H n the corresponding real lattice Areal in R 4n is

obtained by replacing every component u + iv + jw + kx of every vector in A by
(u, v, w,x). In this case

(69)

In all cases Areal has the same packing and covering radii as A, and we define the
density, kissing number and thickness of A to be equal to the corresponding values
for Areal.
In the case where A is an tf-lattice (where tf is the ring of Eisenstein integers),
multiplication of vectors in A by w is a fixed-paint-free automorphism of order 3
of A and Areal; only the zero vector is fixed. Conversely, a real lattice in R2n with a
fixed-point-free automorphism of order 3 can be regarded as an tf-lattice in en. If
A has generator matrix M = X + iY (X, Y rea!), then Areal has generator matrix

[V2(X:.J3Y) V2(Y~.J3X)]· (70)

Similarly if A is a '!i-lattice (where '!i is the ring of Gaussian integers),

mUltiplication by i is an automorphism" of A and Area" " has order 4, and the
powers ", ,,2, ,,3 have no fixed points except O. If A has generator matrix
M = X + iY (X, Y rea!), then Areal has generator matrix

[-~ ;].
The situation is a little more complicated if A is an Jff..lattice (where .Yf is the
ring of Hurwitz integers). yt' contains 24 integers of norm I, or units, namely
±I, ±i, ±j, ±k, ±w, ±w',

where

w= Ih(-I+i+j+k),

w= 'h(-I-i-j-k), (73)
Coverings, Lattices and Quantizers 55

and
Wi =i-lwi =jw=wk -w+i = lh(-l+i-j-k),
wi = rlwj - kw = wi = w + j - lh(-l-i + j-k),
wk = k-lwk - iw = wj - W + k = ~'2(-I-i-j+k),
wi = i-lwi = -kw = -wj - w - i = lh(-l-i+j+k),
wj = rlwj = -iw = -wk - w - j = lh(-1+i- j+k),
;;;k =k-lwk =-jw=-wi -w-k =¥2(-1+i+j-k). (74)

As a multiplicative group this set of units is isomorphic to 2A 4, where A4 is the

alternating group of degree 4. Let L" denote left multiplication of vectors in A by
'If, where 'If is one of the 24 units in :Yf. If 'If ¢ 1, L" is a fixed-point-free
automorphism of A and Areal' Conversely, if L is a real lattice in R4n which is a
left (2A 4)-module, and every nontrivial element of 2A4 acts in a fixed-point-free
manner, then L can be regarded as an Jf'.lattice in Hn. If A has generator matrix
M = U + iV + jW + kX (U, V, w, X real), then Areal has generator matrix

U V W X
-V U -X W
(75)
-W X U -V
-X -W V U

Examples. If A = f'fJ in C, Areal is the square lattice Z2; if A - 8, Areal is the

hexagonal lattice (Fig. l.3a); and if A - J'f, Areal is D4 (Chap. 4, §7.2).
The densest six-dimensional lattice E6 has a simple description as an Eisenstein
lattice. If A is the 8-lattice in C1 with generator matrix

[~ ~ :1 (76)

where 8 = w - W = ~,then Areal is E 6 •

Similarly the densest eight-dimensional lattice E 8 has a simple description as a
Hurwitzian lattice. If A is the Jf'.lattice in H2 with generator matrix

[17 ~]
i
(77)

then Areal is E 8•
The Leech lattice can be constructed in all three ways, as a f'fJ-, 8- or Jf'.
lattice - see §1 of Chap. 6 and Example 12 of Chap. 7.
Finally, we briefly discuss the existence of even complex lattices. The norm of
A, denoted by .A1A), is the subgroup of the real numbers generated by
( N(x) : x E A}. We have already seen that if A is a real self-dual lattice then
.AI(A) is either Z or 2Z [O'Mel, pp.227,324). The same proof shows that this
result also holds for complex lattices. Lattices for which .A1A) - 2Z are called
even. Even self-dual f'fJ-Iattices exist in Cn if and only if n is a multiple of 4; there
are no even self-dual 8-lattices [Ger5, Th. 3.6). There are however even self-dual
lattices over other rings of integers - see for example §4 of Chap. 8.
56 Chapter 2

3. Quantizers
3.1 Quantization, analog-to-digital conversion, and data compression. As
mentioned at the end of §1.4 of Chap. I. although this book primarily
discusses geometrical problems. many of them have direct application to
other branches of science. In this section and in §! of Chap. 3 we describe
two of the principal applications. to vector quantizers and to the design of
signals for use over a noisy channel.
The real world is full of nasty numbers like 0.79134989 ...• while the
world of computers and digital communication deals with nice numbers
like 0 and I. A quantizer. or analog-to-digital converter. is a device for
converting nasty numbers into nice numbers. Quantizers are found in
digital measuring instruments or recording devices. and in digital
communication systems (including most medium- or long-distance
telephone networks, those that use pulse code modulation for example -
see §I.l of Chap. 3). We must add. however. that most existing quantizers
are one-dimensional; the n-dimensional quantizers described here are not
yet in general use.
Figure 2.7 shows a hypothetical two-dimensional quantizer. Eleven
representative points PI ..... P" have been chosen in advance. The input

VORONO! CELL
AROUND P4

Figure 2.7. An example of a (hypothetical) two-dimensional quantizer.

Eleven representative points P \ ..... P \\ have been chosen in advance.
Any point x in the plane is then "rounded off'·. or quantized, to the
closest point Pj. The illustration also shows the Voronoi cells around each
Pj • For example. any point x falling in the Voronoi cell of P 4 (the
shaded region) is rounded off to P 4.
Coverings, Lattices and Quantizers 57

INPUT OUTPUT IS
QUANTIZER

x PI' P2 ,·· CLOSEST Pi TO x

Figure 2.8. An n-dimensional quantizer.

to the quantizer is a pair of real numbers (X"X2), which we regard as a

point x in the plane, and the output is the closest one of PI, ... ,P II to x.
Any point x in the plane is thus replaced by (or rounded-off to) the closest
point Pi. This process is also called data-compression; the output could
equally well be the index i of the closest Pi.
The formal definition of an n-dimensional quantizer is as follows
(Fig. 2.8). M points PI, ... ,PM are chosen in Rn. The input x is an
arbitrary point of Rn; the output is the closest Pi to x. If the closest Pi is
not unique, one of the closest Pi is chosen at random to be the output.
The action of this quantizer may also be described by saying that the
space R n is partitioned into the Voronoi cells (§1.2) V(P I ), V(p 2), .••
around the Pi; if the input x belongs to V(P), the output is Pi (see
Fig. 2.7).
Quantizing introduces errors, the magnitude of the error being
measured by the Euclidean distance from x to the closest Pi. The Pi
should therefore be chosen according to the probability distribution of x so
as to be representative or "typical" points. More precisely we shall
attempt to choose the Pi so as to minimize the mean squared error (or
m.s.e.), the average value of N (x - Pi (x», where Pi (x) denotes the
closest point to x .
If x has probability density function p (x), the average mean squared
error per dimension is

(78)

which, since the Voronoi cells partition R n , can be written as

(79)

Remarks. (j) (78) and (79) assume the probability that the input x lies
exactly on the boundary of a Voronoi cell to be negligibly small. (ii) The
factor lin is to ensure a fair comparison between quantizers of different
dimensions. (iii) The mean squared error criterion is only one of many
possible distortion measures; it has the advantage of being widely used and
is mathematically the simplest. For applications to speech or picture
processing the correct choice of distortion measure is a difficult problem.
See for example [Atal1-[Ata3], [Flal], [SchlO1.
58 Chapter 2

The main theorem in this subject is due to P. L. Zador, who showed in

1963 ([Zadl], [Zad2]) that it is possible to reduce the average mean
squared error per dimension by using higher-dimensional quantizers. In
other words it is more efficient to wait until several numbers have arrived
at the quantizer, and then to quantize them all at once by regarding them
as specifying a point in some high-dimensional space. It pays to
procrastinate.
Unfortunately Zador's result, like others we have mentioned, is
nonconstructive, and the problem of explicitly finding good multi-
dimensional quantizers is still unsolved. But the lattices shown in
Table 1.1 perform very well if the input x has a uniform distribution, as
we shall see.
We now give a more precise account of Zador's result. Given the
dimension n, the number of points M, and the probability density function
p (x) of the input, we wish to find the infimum
E(n,M,p) = inf E, (SO)
{Pj}

i.e. the smallest error attainable by any set of M points PI • ...• PM E Rn.
Zador ([Zadll, [Zad2]; see also [Buc2], [Buc3], [Ger21, [Yaml]) showed
under quite general assumptions about p (x) that
n+2
Ji~oo M2/n E(n,M,p) = Gn [£p(x)n/(n+2)dX]-n-, (S1)

where Gn only depends on n. Zador also showed that

2/n

(n
1
+2h
r [ !!.+I )
2 G ~ n~
n1-rr r [ ; + 1 rn r [1+ ~ ). (S2)

For large n the upper and lower bounds in (S2) agree, giving
1
Gn -- -2- = 0.058550... as n -- 00 (83)
-rre

Since the probability density function p (x) only appears in the last
term of (S 1), we may choose any convenient p (x) when attempting to find
Gn • The simplest choice, and one which is of considerable interest in its
own right, is to assume that x is uniformly distributed over a large region
of R n (a large ball, for example). With a uniformly distributed input the
mean squared error is minimized if each point Pi lies at the centroid of its
Voronoi region V(p) [Ger2]. We are most interested in the case when M,
the number of points, is very large. This enables us to ignore boundary
effects, and leads to simpler answers. If the input is uniformly distributed,
(79) and (S 1) become
1.. ~
n
J
i-I V(P j )
N(x - Pi)dx
E = ------M------------- (S4)
~ J
i-I V(Pj)
dx
Coverings, Lattices and Quantizers 59

E(n,M,p)
(85)
[ ~ i~ vt;) dx rn ,
where E(n,M,p) = inf E. Equations (84) and (85) can be interpreted as
saying that, for optimal quantization of a uniformly distributed input using
a large number of points, Gn is the average mean squared error per
dimension. scaled so as to be a dimensionless quantity (independent of the
choice of scale of the Pi)'
3.2 The quantizer problem. From now on we assume that the input x has
a uniform distribution over a large ball in Rn, and that M, the number of
points Pi, is large. The quantizer problem is to choose points PI • ...• PM
in Rn so as to minimize

-1 M
~ f N(x-P)dx
M i-I V(P;)
(86)
n M }1+1-'
{ _1 ~ Vo\(V(p) n
M i-I

where V(Pi ) is the Voronoi cell of Pi' Then (from (84), (85» Gn is equal
to the limit as M - 00 of the infimum of (86) over all choices of the Pi'
By the previous section the solution to the quantizer problem gives both the
optimal quantizer for uniformly distributed inputs and, via (81), the error
probability for an optimal quantizer when the input has an arbitrary
probability density function p (x).
If all the Voronoi cells V(p) are congruent (as in the case when the Pi
form a lattice), (86) can be simplified. Suppose the Voronoi cells are
congruent to a polytope IT. If we place the origin of coordinates at the
centroid of IT, (86) becomes
1- f x . x dx
n II
-=---~ = G (IT), (87)
1+1-
Vo\(IT) n

the normalized second moment of IT defined in §1.2. If the points Pi form

(a large segment of) a lattice A, with Voronoi cells congruent to a polytope
IT, we write G (A) = G (IT). The lattice quantizer problem is to find an n-
dimensional lattice A for which G (A) is a minimum.
3.3 Quantizer problem-summary of results. It is not difficult to show that
the solution to the one-dimensional quantizer problem is to place the points
Pi uniformly along the real line, so as to form (a large segment of) the
integer lattice Z. The Voronoi cells are intervals of length 1, and we take
IT to be the interval [_1/2, 1/21. From (87) we have
1/2
f x 2dx
1
GI =
_-..:.;,1/.::::2__
1/2 = 12 = 0.083333 .... (88)
f
-1/2
dx
60 Chapte r 2

This is a familiar result in quantization theory: if uniformly distributed

numbers are quantized one at a time, the average error is 1/12. Zador's
thesis shows that this error can be reduced by using higher-dimensional
quantizers.
For the two-dimensional problem Fejes Toth ([Fej81; see also [FejlOl,
[Ger21, [NewIJ) showed that the points should form the hexagonal lattice;
correspondingly

G 2 = G(regular hexagon) = 5 r;:; = 0.0801875... (89)

36,,3
(see Chap. 21, §20. As usual the problem is unsolved in higher
dimensions.
The solution to the lattice quantizer problem is known in three
dimensions [Bar161: the best lattice quantizer is the bcc lattice, and
19
G (bcc) = G (truncated octahedron) = 1/3 = 0.078543 ... . (90)
192·2

It is shown in [Bar161 that the bcc lattice is the only lattice at which G (IT)
is even a local minimum . For comparison
G (fcc) = G (rhombic dodecahedron) = 2- 11 / 3 = 0.078745... (9 J)

(Chap. 21 , Eq. (26», but this can be reduced by perturbing the lattice
slightly. In higher dimensions even the best lattice quantizers are not
known .

0 .082

<:)c 0.080

>-
z 0.078
I.LJ
:::;
~ 0 .076

0*n
~ 0 .074
o
u
~ 0072
o
~ 0.070
..J
~ 0.068
0:::
CONJECTUREO
~ 0.066 LOWER 80UND (92)

0 .064 ~-!:--'-~~--'--=--'-:'::-'--:'::--'-:":--'---:'::-'-~'--::'::-'--;';:-.L.:f~
o 2 4 6 8 10 12 14 16 18 20
DIMENSION n

Figure 2.9. The best quantizers known in dimensions n ~ 24.

Coverings, Lattices and Quantizers 61

Table 2.3. Bounds for G., the mean squared error of the optimal n-
dimensional quantizer. (92) is our conjectured lower bound.

Name of Mean squared Lower

n quantizer error Gn bound (92)

0 Ao 0 0
1 Z 0.083333 0.083333
2 A; == A2 0.080188 0.080188
3 Ai == Di 0.078543 0.077875
A3 == D3 0.078745
4 D: ==D4 0.076603 0.07609
A: 0.077559
A4 0.078020
5 D; 0.075625 0.07465
Ds 0.075786
A; 0.076922
As 0.077647
6 E~ 0.074244 0.07347
E6 0.074347
7 E; 0.073116 0.07248
E7 0.073231
8 E; == Eg 0.071682 0.07163
D; 0.074735
Dg 0.075914
A; 0.075972
As 0.077391
12 K;2 == KI2 0.070100 ± 0.000024 0.06918
16 A;6 == AI6 0.068299 ± 0.000027 0.06759
24 A;4 == A24 0.065771 ± 0.000074 0.06561

Tables 1.1, 2.3 and Fig. 2.9 show the best quantizers known in
dimensions n ~ 24. For some of these lattices G (A) can be determined
exactly, as shown in Chap. 21. However, for E~, E;, K 12 , AI6 and A24 ,
G (A) was evaluated in [Con38] by Monte Carlo integration, making use of
algorithms (see §1.4 and Chap. 20) that find the closest lattice point to an
arbitrary point of Rn • (2)
Table 2.3 and Fig. 2.9 also show Zador's bounds (82). The lower
bound in (82) on Gn is the normalized second moment of an n-dimensional
sphere - see Chap. 21, Eq. (8).

(2) Worley [Worl1, [Wor2] has since found exact values for E: and E;
see §3.D of Chap. 21.
62 Chapter 2

In [Con391 we conjecture a new lower bound on Gn which is

significantly stronger than Zador's lower bound. The conjecture is that Gn
is at least

n+3-2Hn +2
4n(n+1)
{ 42[12" cos-II]};
(n+I)(n!) Fn -;; ,

where Hm = I-I + rI+ ... + m- I (m = 1,2, .. .) is a harmonic sum, and

Fn (a) is defined in Eq. (64) of Chap. 1. This bound is also given in
Table 2.3 and Fig. 2.9. Although no formal proof has been found, there
are plausible geometrical reasons for believing that the bound is correct.
The conjectured new bound is analogous to the Rogers bound for the
sphere packing problem (Eq. (39) of Chap. 1), the Coxeter-Few-Rogers
bound (16) for the covering problem, and the Coxeter-Boroczky bound for
spherical codes (Eq. (63) of Chap. 1).
It is worth remarking that the best n-dimensional quantizers presently
known are always the duals of the best packings known. However, we do
not expect that this will always be the case. It would be nice to have more
data. What is G (A) for Dickson's four-dimensional lattices mentioned in
§1.3, or for the Barnes-Trenerry lattices also mentioned there?
For further information about quantizers see [Adol], [Berl1-[Ber31,
[Bucl1-[Buc31, [Dav41, [DunlO], [Elil1, [Gal2], [Gerll-[Ger31, [Gisl1,
[Ora7], [Jayl], [Lin21, [Maxi], [ZivIl, [Ziv21.
 3
Codes, Designs and Groups
J. H. Conway and N. J. A. Sloane

This is the last of the three introductory chapters. We begin by describing

the second main application of sphere packings, the design of signals for
data transmission or storage systems. The remaining sections are devoted
to topics that, although not our primary concern in this book, are always in
our minds: error-correcting codes, Steiner systems, t -designs and finite
groups.

1. The channel coding problem

1.1 The sampling theorem. The second main application is to the design of
signals for use in data transmission or storage systems. A typical system is
shown in Fig. 3.1. Messages are produced by an information source, such
as a human speaker, an orchestra or a computer. The source encoder
converts the messages to digital form. This process often involves
quantization or analog-to-digital conversion, as described in the previous
chapter. The messages are to be conveyed to the destination via a channel,
which may be a data transmission channel, such as a copper telephone
wire, microwave radio link, or optical fiber, or a data storage device such
as a magnetic or optical disk. There is noise on the channel, so what is
received (or retrieved) may differ slightly from what was transmitted (or
stored).
The source produces messages at a certain rate, and we wish to
communicate them to the destination reliably, efficiently and cheaply. The
communication system in Fig. 3.1 accomplishes this by choosing a set of
special signals called the code, the members of which are designed to be
easily distinguishable from each other even in the presence of noise. Only
signals from the code will be used on the channel.
The channel encoder takes the output from the source encoder and
replaces it by one of the code signals, which is then transmitted over the
channel (or stored in the recording device). The channel decoder reverses
64 Chapter 3

Figure 3.1. Block diagram of data transmission or storage system.

this process by taking the received (or retrieved) signal and making a
(hopefully correct) estimate of the code signal; the source decoder then
converts this back to the original message.
From now on we shall just describe the data transmission problem,
always keeping in mind that there are identical results for data storage
systems.
The sampling theorem plays a crucial role in many communication
systems. The sampling theorem (Fig. 3.2) states that if f (t) is a signal
(j.e. a function of time) containing no components of frequency greater
than W cycles per second, then f (t) is completely specified by its samples

... f [ - - I ] f(O) f[-I ] f[_2 ] ... (!)

, 2W' , 2W' 2W'

taken every I!(2W) seconds (see [DymIl, [HigIl, UerIl, [LanIl, [LeeO],
[Pet21, [Sha21, [Ste 11, [Woz 1]). In fact there is an explicit formula, the
cardinal series, expressing f (t) in terms of its sample values:

f(t) =k_~_oof[2kw] sin27rW(t-k/2W)

27rW(t-k/2W)

Since

J sin 27rW(t-k/2W)
27r W(t - k/2W)
sin 27r W(t -1/2 W) dt
27r W(t -I /2W)

is equal to 0 if k ;z!: I, or to 1/(2W) if k = I, the energy in f(t) is given

by

f00

-00
f(t)2dt = -
look
~ f -
2W k--oo 2W
[ ] 2

The sampling theorem is of both practical and theoretical importance in

digital communication systems. The theorem is easily applied in practice,
since from Eq. (2) the signal f(t) can be recovered from its samples by
Codes, Designs and Groups 65

(0 ) f (')

o~
-AT11lI\ ~ T
2W

(b )

o
IIIII
II T

(e)

( f (0). f (;W ) .f (22W ) .... . f (~~) )

(d)

•
Figure 3.2. The sampling theorem is the basis for many digital
communication systems. (a) A function 1 (t) which does not change too
quickly, in particular one which does not contain any frequency
components greater than W cycles per second, is sampled every 1/(2W)
seconds. (b) The sample values 1(0), 1(J/2W), 1(2/2W), ... are
transmitted. The theorem states that 1 (t) can be reconstructed exactly
from these samples. This is the principle underlying pulse code
modulation (PCM) . (c) In T seconds there will be n samples, where
n = 2TW. These samples (f(0),f(J/2W) . ... '/«n-J)/2W» are the
coordinates of a point in n-dimensional space. (d) Thus the complicated
function 1 (t) has been represented by a single point in n-dimensional
space.

feeding them into a simple electrical circuit (a low pass filter). Moreover
the recovery is stable, in the sense that small errors in the sample values
produce only correspondingly small errors in the recovered signal.
For example the sampling theorem is the basis for the digital
transmission system known as pulse code modulation, or PCM ([Bell],
[Ben4], [OliI)). Amplitude modulation (AM) and frequency modulation
(FM) are no doubt familiar to the reader. PCM is less well-known but
just as important, being widely used in medium-distance telephone calls.
The speaker's voice is sampled at the appropriate rate (usually 8000 times
a second), and each sample is rounded off or quantized to one of 256
levels, and then expressed as an 8-bit binary number. The binary numbers
are transmitted and at the receiving end the speaker's voice is
reconstructed from the samples. Since the sampling theorem is a true
theorem, the listener is unaware that this has happened.
66 Chapter 3

The sampling theorem also has theoretical implications. If the signal

f(t) lasts for T seconds(]), as shown in Fig. 3.2, then there are n = 2TW
samples, for instance

But these are the coordinates of a point in n -dimensional space. So the

sampling theorem makes it possible to represent this complicated function
f (t) by a single point (-) in n -dimensional space. This shows the power of
mathematical notation!
Furthermore, if FERn denotes the point with coordinates (4), then
from (3) the norm of F is proportional to the energy in f (t):
T
N(F) = F . F = 2W J f(t) 2dt = 2WTP = nP , (5)
o

where

(6)

is the average power in the signal. If G represents another signal g (t),

then

J (j(t}-g(t)2dt,
T
N(F-G) =!!:.... (7)
T 0

T
F· G = ; Jo f(t) g{t)dt. (8)

1.2 Shannon's theorem. The observation that signals of finite bandwidth

can be represented by points in Euclidean space is a crucial ingredient in
the research that Claude E. Shannon carried out between 1940 and 1960
into the mathematical foundations of communication theory [Sha 1]-[Sha8].
As David Slepian [Sle5] has said, "probably no single work in this century
has more profoundly altered man's understanding of communication than
Shannon's 1948 paper A Mathematical Theory of Communication."
Before describing some of Shannon's results let us mention the two
idealized models for the channel in Fig. 3.1 that are the most important.
One is the binary symmetric channel, in which only sequences of O's and
l's are transmitted and received. In this case the code is called a binary

(1) Strictly speaking it is impossible for a function to have simultaneously

a finite bandwidth and to be nonzero for only a finite interval 0 ~ t ~ T.
To make this precise we should consider bandlimited signals which have
almost all of their energy in the range 0 ~ t ~ T. See [Dymll, [Lan31,
[Lan41, [Sle61, [Sle7].
Codes, Designs and Groups 67

error-correcting code; we discuss such codes in §2. The other is the

Gaussian white noise channel; the problem of designing codes for this
channel is related to the problem of packing spheres in Euclidean space.
The Gaussian white noise channel transmits continuous signals. All
frequencies above a cutoff frequency of W cycles per second (called the
bandwidth of the channel) are attenuated completely; frequencies below W
are passed without attenuation. In the course of transmitting the signal,
the channel adds white Gaussian noise to it.
This channel has a very simple description based on the sampling
theorem. A transmitted signal f (t ) is represented by a point
F = if \' ... In) in n-dimensional Euclidean space. During transmission
this point is perturbed by the addition of a noise vector Y = (y \' ... ,Yn)
whose components are independent Gaussian random variables with mean
o and variance (f2. (Thus (f2 is the average power of the noise,) The
received signal is represented by the vector F + Y.
For this channel the code is a set of points in Rn. If there are M code
points, each representing a signal of bandwidth Wand duration T seconds,
the rate of the code is defined to be

R = T1 log2 M bits/sec. (9)

The decoder finds the closest code point to the received vector and from it
reconstructs the signal. If the noise is small, the closest code point will be
the transmitted point and the signal f (t) will be recovered correctly. But
if the noise is large the received vector may be closer to some other code
point and will be decoded incorrectly; this is a decoding error.
Thus we may reduce the effect of the noise by placing the code points
further apart. On the other hand by (5) this requires signals of greater
energy and increases the cost. One of Shannon's basic theorems is the
surprising result that it is possible to attain essentially error-free
transmission over this channel while using only signals of finite power,
provided the rate of the code does not exceed a critical threshold called the
capacity of the channel. There are similar results for many other channels,
including the binary symmetric channel.
The precise statement of Shannon's theorem for the Gaussian channel
is as follows. For any rate R less than the capacity

C = W log2 [1 + ~ ) , (10)

by making T and hence n = 2 WT sufficiently large we can find a code of

rate R (Eq. (9» and average power at most P (Eq. (6» for which the
probability of a decoding error is arbitrarily small. Conversely such codes
do not exist for rates R ~ C.
We shall sketch a proof of a weaker result, to show the connection with
sphere packing; for a rigorous treatment of the full theorem see [Gall],
68 Chapter 3

[McEll, [Sha21, [Sha61, [Sle51, [Wol3l, [Wyn3]-[Wyn51.

Since the signals F have average power ~ P, by (5) they lie in a
sphere of radius Jnji around the origin. By making n sufficiently large,
we can assume that the noise vector Y, which has n components of
variance u 2 , has norm N (y) ~ n (u 2 + f), where f is arbitrarily small. In
other words, with high probability the received vector F + Y lies in a small
sphere of radius ~ {n(u 2 +f)}1/2 centered at F. The average power of the
received vector is ~ P + u 2 + f, and in particular F + Y lies in a sphere
around the origin of radius {n(P+u 2 +d Jl/2.
Therefore, if we take the code to consist of the centers of spheres in a
dense packing of spheres of radius {n (u 2 + f) Jl/2 in a large sphere of radius
{ n (p +u 2 +d }l/2, we obtain a code in which the transmitted vector will
be decoded correctly with probability close to 1. The number of code
points is, from Eq. (10) of Chap. 1,
] nl2
Ll [ P+u 2
M < 2 '
U

where Ll is the density of the sphere packing. By making n large we can

make M arbitrarily close to (11). The rate of this code is, from (9), (11),

[ p]
R ::::: W log2 1 + -;?
2W
+ -n- log2 Ll.

Therefore, by making T and n sufficiently large, and using a sphere

packing of density satisfying Oog2Ll)/n ~ -I (which we know exists by
Chap. I, Eq. (46», we can achieve essentially error-free transmission at
any rate

R < W log2 [ I + ~]- 2W . (12)

More sophisticated arguments are needed to establish that rates up to

the capacity (I 0), and no higher, can be achieved without error - see the
references cited above. Shannon's argument, which uses an averaging
method, is one of the best known non-explicit proofs in coding theory. (See
also the remarks in §1.5 of Chap. 1.)
The proof does show that, in order to achieve rates close to (10), nand
T must be large. More precise versions of the theorem investigate how the
probability P e of a decoding error drops as the dimension n increases, for a
particular rate R. Slepian [Slel] has examined the tradeoffs between all
seven parameters W, T, n, R, P, u 2 and Pe.
The rigorous proof of (10) also shows that, for small values of u,
finding an optimal code of maximal power P is closely related to finding
the densest packing of spheres in Rn, as we saw in our heuristic derivation
of (12). De Buda and Kassem [Bud21, [Bud3], [Kas2] have shown that
codes obtained from lattice sphere packings are essentially as good as any
codes, for all values of u.
Codes, Designs and Groups 69

A second, very similar problem asks for the best code in which all
signals have the same energy (a constant energy code). For small values of
II the solution to this problem is closely related to the problem of packing

spherical caps on the n -dimensional sphere n n' i.e. to the problem of

constructing spherical codes (see §2.3 of Chap. 1).
1.3 Error probability. We can state the channel coding problem more
precisely by giving an expression for the error probability Pe . Suppose the
code consists of M code points C 1, . . . ,CM in Rn, and let V(Ck ) be the
Voronoi cell for Ck . Given that Ck is transmitted, the decoder makes the
correct decision if and only if the noise vector Y is in V(Ck ), an event of
probability

Assuming that all code points are equally likely to be used, the error
probability for this code is
I M
Pe = I - - ~
M k-l

If all the Voronoi cells are congruent, to some polytope II say, this
simplifies to

Pe = 1-
(~)
1
IIv27r. n
f
II
e -x'x/2u d x.
2
(Is)

Then one version of the Gaussian channel coding problem is the following.
Given the dimension n, the number of code points M, and a power
constraint
k = I, ... ,M,

find a code C 1, . . . , CM E Rn satisfying (I 6) for which the error

probability P e given by (I4) is minimized. The constant energy problem
replaces (I 6) by the requirement that N(Ck ) = nP for all k, or
equivalently that all Ck E n n'
The lattice version of the Gaussian channel coding problem is to find,
for a given value of II, that n-dimensional lattice of determinant I for
which (IS) is minimized, where II (of volume 1) is the Voronoi cell of the
lattice.
The difference between the four main lattice problems is now clear.
Let A be a lattice with Voronoi cell II, of volume 1. For the packing
problem (§I of Chap. 1) we maximize the in-radius of II, for the covering
problem (§I of Chap. 2) we minimize the circumradius, for the quantizing
problem (§3 of Chap. 2) we minimize the second moment G (II), and for
the channel coding problem we minimize Pe (Eq. (IS».
In general the answers to the channel coding problems will depend on
II. Unfortunately, unlike the expression for quantizer error given in
70 Chapter 3

Eq. (87) of Chap. 2, the integrals in (14), (15) cannot be evaluated

exactly even in simple cases (except in one or two dimensions, or if II is a
cube, when Pe can be expressed in terms of the error function (18».
Because of the difficulty in evaluating the integral, most investigators have
replaced (14) by a simpler expression.
Probably the most useful estimate is the following union bound (cf.
[WeI7], [Wozl]). Consider the Voronoi cell V(C k ) of the kth code point
Ck • Suppose the walls of V(Ck ) are determined by r code points
Ck 1 ••••• Ckr (i.e. these are the relevant points for Ck in the notation of
§1.2 of Chap. 2). The complement of V(Ck ) is contained in the union of
the corresponding half-spaces. Therefore the error probability given that
Ck is transmitted, Pe(k), is bounded by

p}k) ~ :i; _1_ j e-x2/2q2 dx ,

j-l u.,J2; Pj

where Pj = lh liCk -Ckjll. We write this as

Pe(k) ~ ~
r lh p. )
erfc [ _1_ , (17)
j-l u.J2
where

f e-
00

erfc (x) = - 2 t2 dt (18)

.J;x
is the error function [Abrl, p. 297]. An upper bound on (18) is obtained
by including the contribution from the half-spaces corresponding to all
C ¢ Ck • Then averaging over k we get
~ _1 ~ ~ 11.
Pe...."M~~72erc f IIC-C'II (19)
In'
C c' ... c uv8

As u -+ 0 the right-hand side of (19) approaches Pe . This is the case when

P » u, the case of high signal-to-noise ratio. Let P = min Pj be half the
minimal distance between code points, and let T be the average number of
code points at distance 2p from a code point. (This is not the same as the
average kissing number.} Then for small u the solution to the coding
problem is found by minimizing

(20)

So even when u is small it is not quite correct to say that the coding
problem coincides with the sphere packing problem, since the latter
maximizes P but ignores T. At the other extreme, when u is large, we can
approximate exp{-x'x /(2u 2 )} by 1 - X 'x/2u 2 + ... in (14), and the
coding problem coincides with the quantizing problem.
Codes, Designs and Groups 71

1.4 Lattice codes for the Gaussian channel. We now consider the lattice
channel coding problem in more detail. In one and two dimensions the
optimal lattice is independent of u. On one dimension there is only one
lattice, and in two dimensions the optimality of the hexagonal lattice for all
u follows by taking j(x) = (ufu)-l exp{-x 2 /2u 2} in the theorem on
page 81 of [FejlOJ.)
We now assume u is small. From (20) we may approximate the error
probability Pe by

Let the code consist of all lattice points C with N (C) ~ nP. The number
of points, M, is given by
M· Vnpn =.:l' Vn (np)nI2 (1+0(1), (22)

(where the term 0(1) - 0 as M - 00), so

lin
[
p=.Jnji! ) (1+0(1). (23)

The average norm of the code points is

_n_. nP(1+o(1) (24)
n+2

(see [Cai7, Th. 4)). For this analysis it is better to define the rate of the
code to be
1
R = - log2 M bits/dimension (25)
n

(rather than (9». Then

(26)

We also define the normalized signal-to-noise ratio of the code to be

By combining (21), (26), (27) we obtain our final estimate for the error
probability of a Gaussian channel code with normalized signal-to-noise
ratio S, obtained from an n-dimensional lattice of density .:l and kissing
number r:

Pe "=; erfc[.j1!f.:ll/n). (28)

72 Cha pter 3

This estimate becomes more accurate as S increases. For large values of x

and therefore

log Pe" - constant - [~ nA2/n 10ge]s. (0)

We have calculated the error probability of various lattices using (28).

Figure 3.3 shows those lattices in dimensions 2, 4, 8, 24 and 48 that have
the smallest known values of Pe", for Pe" < 10- 3 . In this range the best
lattice codes presently known coincide with the densest known lattice
sphere packings. Furthermore in this range 10glO Pe " is a linear function of
S, with slope given by (0).
From (28)-00) we see that, as long as the densest lattice sphere
pac kings in Rn all have the same kissing number, then as (J - 0 (or
S - 00) the lattice coding problem coincides with the lattice sphere
packing problem. But if there are two or more equally dense lattices with
different kissing numbers, then the lattice having the smallest kissing
number is to be preferred.

-4

-6

-6
= •
n.
0

8"
oJ
-10

-12

-14

P48q

10 15 20 25
SIGNAL- TO- NOISE RATIO S

Figure 3.3 . The smallest known error probability P/ (Eq. (28» for any
n -dimensional lattice, as a function of the normalized signal-to-noise ratio
S (Eq. (27», for various dimensions n ~ 48.
Codes. Designs and Groups 73

We have also calculated Pe "exactly" from (15) for various low

dimensional lattices using Monte Carlo integration. The results (as yet
unpublished) agree with those obtained from (28).
A cruder measure of the efficiency of a code is provided by the ratio
IL IE, where IL is the minimal squared distance between code points and E
is their average norm (or energy). The (nomina/) coding gain of a code
with parameters IL" E lover another with parameters 1L2' E 2 is defined to
be

1010g lO [ -ILl I -1L2) db.

EI E2

For example, if we take our code to consist of all points C of norm at most
nP in a lattice A, then

from (23), (24). For comparison we use a roughly equal number of points
arranged in an n-dimensional cubical array centered at the origin (an
essentially one-dimensional code). For this code
.1!:... = _12_ (1+0(1). (33)
E nM2/n

The nominal coding gain obtained from the lattice A is therefore

10 10glO { ~ (n +2)11 ~}. (34)

The coding gains of various lattices in up to 128 dimensions are shown in

Table 3.1. The table also gives the value of 'Y = 402/n (where o=MVn )
for each lattice. 'Y is a lower bound to Hermite's constant (Eq. (47) of
Chap. 1), and in view of (28)-(34) is another measure of the coding gain
of a lattice.
As pointed out by Forney [Fori], the estimate (19) for Pe can be
related to the theta series of the lattice. Since [Mit6, p. 2911
erfc (a) ~ e- a2 (a ~ 0),
we have

(35)

and the right-hand side approaches Pe as u - O.

Welti [Wel6] and Welti and Lee [We17] have used (20) to estimate Pe
for other four-dimensional codes obtained from lattices. Other
approximate formulae for the error probability of lattice codes have been
used in [Bla3] (cf. [Sloll]) and [Bud21.
74 Chapter 3

Table 3.1. Nominal coding gain (Eq. (34» for various lattices in
Tables 1.2, 1.3, and the values of the center density 0 and 'Y = 402/n, a
lower bound on Hermite's constant.

n Lattice 0 Gain (db) 'Y

2 A2 1/(2J3) 0.825 1.155

4 D4 1/8 l.961 1.414
6 E6 1/(8J3) 2.832 1.665
8 E8 1/16 3.739 2
12 K12 1/27 4.514 2.309
16 A16 1/16 5.491 2.828
24 A24 7.116 4
48 P48q (3/2)24 9.037 6
64 P 64c 222 9.397 6.442
80 ." (E 8) 236 10.070 7.464
128 ." (E 8) 288 11.557 10.375

The channel coding problems described in this section are idealized

mathematical problems. In practice the situation is more complicated and
many other factors must be considered. Shannon's theorem implies that T
(the duration of the signals) and n (the dimension) must be increased in
order to reduce the probability of error, but this both delays the encoding
and decoding processes and makes them more complicated. For a code to
be used in practice, efficient encoding and decoding algorithms must exist.
Furthermore other types of distortion besides Gaussian noise must be
considered Gntersymbol interference, for example, or noise which occurs in
bursts) .
A great many constructions of practical channel codes have been
proposed: for example permutation codes [Bi121, [Sle21, "group codes"
(see §4.2), and the codes described by [Cam4], [Einll, [Ein2], [Fosll,
[Fos21, [Gaol], [Ger4], [Gill], [GinI], [Kanll, [Ker21, [Krill, [Lazll,
[Sayl], [Zetl1, [Zet2). See also the references on the construction of
spherical codes given in §2.3 of Chap. 1.
In recent years, geometrical channel codes of the type described in this
section have been combined with error correcting codes (§2), especially
with convolutional codes, to produce what are called trellis code
modulation (TCM) schemes. These are in effect very high dimensional
codes, but have practical encoding and decoding algorithms. See [AndI],
[CaI2]-[CaI8], [Falll, [For2]-[For3], [UngI], [Weill, [WiI20).
Codes, Designs and Groups 75

2. Error-correcting codes
2.1 The error-correcting code problem. The other idealized model for the
channel in Fig. 3.1 is the binary symmetric channel. For this channel the
input and output symbols are O's and 1's, and there is some fixed
probability p < V2 that when a 0 or 1 is transmitted, the other symbol is
received; with probability 1 - P > V2, the correct symbol is received. A
binary cod/2) C of length n is a set of binary vectors (called codewords)
with n coordinates, or in other words is a subset of Fq, where F2 = { 0,1 }
is the Galois field of order 2.
Similarly a q-ary code is a subset of F;, where Fq is the Galois field of
order q and q is a prime or prime-power. Besides F 2, the ternary field
F3 = { 0,1, -I} and the field of order 4, F4 - { 0, l,w,w 2 } (where
w
w 2 ... = w + 1, w 3 = 1) are especially interesting. A q-ary code is used
on the q-ary symmetric channel, where the input and output symbols are
labeled with the elements of F q •
As in §1.2, we wish to choose the codewords so they are easy to
distinguish from each other even when errors have occurred. To make this
precise we define the Hamming distance between two vectors
U = ( U h " ' , U n ), V = ( V h " ' , V n ),

Ui, Vi E Fq , to be the number of coordinates where they differ:

d (u, v) = 1{ i : Ui ;z! Vi } I.
The (Hamming) weight wt (u) of a vector U is the number of nonzero
coordinates Ui; therefore
d (u , v) = wt(u - v). (36)

The minimal distance d of a code is

d -= min { d (u , v) : u, v E C, U ;z! v }. (37)

If a code has minimal distance d, the "Hamming spheres" of radius

p = [V2 (d-l)] (38)

around the codewords are disjoint (so p is the packing radius of the code,
cf. Eq. (22) of Chap. 2), and therefore the code can correct p errors. A
code of length n, containing M codewords and with minimal distance d is
said to be an (n,M,d) code.

(2) Strictly speaking this is a block code. Convolutional codes (see for
example [LinO], [VitI]) are an important rival family of codes. Although
essential for many practical applications, convolutional codes are not
directly related to the geometrical problems discussed in this book.
76 Chapter 3

A linear (or group) code C is a linear subspace of F;: the set of

codewords is closed under vector addition and coordinatewise multiplication
by elements of Fq' The dimension k (also written dim C) is the dimension
of this subspace, and there are qk codewords. The rate of the code is
1 k
R = - log2 M = - log2q bits/symbol. (39)
n n

A linear code of length n, dimension k and minimal distance d is said to

be an [n, k , d) code (or sometimes an [n, k] code). The minimal distance
of a linear code is the minimal nonzero weight of any codeword:
d = min { wt (u) : u E C, u ;c 0 } . (40)

In a good code n is small (to reduce delays), M is large (to make efficient
use of the channel), and d is large (to correct many errors). Naturally
these goals are incompatible. The error-correcting code problem is the
following: given nand d, find A (n, d), the maximal number of codewords
in any (n,M,d) code. There is a similar problem for linear codes. In
general these problems are unsolved. However many upper and lower
bounds on A (n,d) have been found, and a large number of constructions
of codes are known. A table of A (n,d) for n ~ 24 is given in Table 9.1
of Chap. 9. Other tables may be found in [Bes3], [Chel], [Gra2], [HeI7],
[Mac6], [Pet3], [Pr01], [Slo13], [Verl1, [Zin2], [Zin41.
We will see in Chaps. 5 and 7 that there are strong connections
between codes and sphere packings, and between linear codes and lattice
packings. This is not surprising, since the error-correcting problem is also
a sphere packing problem, although in F~ or F; rather than Rn.
A constant weight code is one in which all codewords have equal
weight. Binary constant weight codes are a discrete analog of the spherical
codes described in §2.3 of Chap. 1.
Since we have written at length about error-correcting codes in [Mac6],
and numerous other references are available ([Ass4], [Ber4], [Ber6],
[Bla2], [BiaS], [Bla9], [LinO], [Linll], [Pet3], [VitI]), the discussion in
this chapter will be brief. In the next section we establish some further
notation, and then give a number of examples.
The covering problem of Chap. 2 also has a binary analog: one wishes
to find the smallest number of overlapping Hamming spheres that will
cover Fr Equivalently, let the covering radius of a code C be
maxmind(x,c) (xEF~,cEC). (41)
x

Then the coding version of the covering problem is, for a given length and
covering radius, to find the smallest possible number of codewords. We
shall not discuss such covering codes in this book, but refer the reader to
[Coh3], [Coh4], [Dow4], [Gra4], [Kill], [KiI2], [SI015], [SI0161. Much
less is known about the binary analog of the quantizing problem.
Codes, Designs and Groups 77

2.2 Further definitions from coding theory. Many of the following

definitions are analogs of the corresponding definitions for sphere packings.
Let B be an n x n monomial matrix, containing exactly one nonzero
element from Fq in each row and column. The set of all such matrices
forms the full monomial group over Fq , of order (q -on n!. The matrix B
sends a code Cover F q into the equivalent code
C' = CB = { uB : u E C } . (42)

If C' is the same code as C then B belongs to the automorphism group

Aut(C) of C (the corresponding notion for sphere packings is defined in
§3). Thus the number of codes equivalent to C is
(q -on n! (43)
IAut(C) I .
An [n, k 1 linear code C may be specified by a generator matrix. This
is a k x n matrix M such that C consists of all linear combinations (with
coefficients from Fq) of the rows of M. It is always possible to find a code
equivalent to C which has a generator matrix of the form
G = Uk IAl,

where h is a k x k identity matrix and A is a k x (n - k) matrix.

Conjugation in the field Fq , where q = pa, p prime, is defined by
x
x -+ = x P , for x E Fq . A similar notation is used for vectors:
U = (Ulo ... ,un), matrices, etc. The dual (or orthogonal) code is defined
to be
C· = { x E F; : x . U = 0 for all u E C } (45)

(cf. Eq. (24) of Chap. I, Eq. (65) of Chap. 2). Then

dim C· = n - dim C. A code is self-dual if C = C'; this implies that n
is even and dim C = n12. If C has generator matrix (44) then C· has
generator matrix
(46)

and C is self-dual if and only if

AA 1r = - [nI2' (47)

There are four especially interesting families of self-dual codes, those in

which the weights of the codewords are multiples of a constant c > 1. In
a binary self-dual code every codeword has even weight. In some binary
self-dual codes the weight of every codeword is a multiple of 4: these are
called Type II, or doubly-even codes; the other binary self-dual codes are
of Type I. In a ternary self-dual code the weight of every codeword is a
multiple of 3. These are called Type III codes. Finally in a self-dual code
over F 4 every codeword has even weight. These are called Type IV codes.
78 Chapter 3

It was shown by Gleason and Pierce that these are essentially the only
possibilities: two proofs of this theorem are given in [SlolO, §6.11. Codes
of type I, II, III or IV exist if and only if n is a multiple of 2, 8, 4 or 2
respectively (see Chap. 7). Similar results for lattices were given in §§2.4,
2.6 of Chap. 2. We shall describe what is known about the enumeration of
these codes in Chap. 7.
Let C be an (n, M, d) code over Fq' We let Aj (c) denote the number
of codewords at Hamming distance i from a codeword c E C. The
numbers {A j (c) } are called the weight distribution of C with respect to
c. Of course Ao(c) = 1, Aj(c) ~ 0, and ~j Aj(c) = M. The Aj(c) also
satisfy certain less obvious inequalities found by De1sarte, which play a
crucial role in the theory. These are described in Chap. 9.
For linear codes (and some nonlinear codes) Aj (c) is independent of c
and will be denoted by A j • The (Hamming) weight enumerator of a linear
code C is
WC(x,y) = ~ xn-wt(u) ywt(u) (48)
uEC
n
= ~ Aj x n - j / . (49)
j-O

This is a homogeneous polynomial of degree equal to the length of the

code, and strongly resembles the theta series of a lattice (Eqs. (31), (32) of
Chap. 2).
The MacWilliams identity [Mac2], [Mac6, Chap. 5] gives the weight
enumerator of the dual code C':
1
WC'(x,y) = M W c (x+(q-1)y, x-y) (50)

compare Eq. (43) of Chap. 2. There is a generalization to nonlinear

codes [Mac6, Chap. 5, §5], [Mac7].
The Hamming weight enumerator classifies codewords according to the
number of nonzero coordinates. More detailed information is supplied by
the complete weight enumerator (c.w.eJ, which gives the number of
codewords of each composition. For example the c.w.e. of a ternary code
Cis
_ ~ no(u) nl(u) n_l(u)
c.w.e.c (x,y,z ) - ~ x y z , (51)
uEC

where nj (u) is the number of times i E F 3 occurs in u. There is also a

version of the MacWilliams identity for c.w.e.'s.
There are mass formulae for codes, just as for lattices (cf. Eq. (47) of
Chap. 2) giving explicit constants ao such that

f IAut(C) I = ao,
(52)
Codes, Designs and Groups 79

where the sum is over all inequivalent codes of a certain type (for example,
all Type II codes of length n) - see [Mac61, [SlolO1. A slightly more
general version gives an explicit formula for
Wc(x,y)
t IAut(C) I
(53)

An (n,M,d) code over Fq can be extended to an (n + I,M,d') code by

appending a zero-sum check digit
n-I
(54)

to each codeword Co CI ... Cn-I. If q = 2 and d is odd the extended code

has d' = d + 1.

2.3 Repetition, even weight and other simple codes. We begin with some
easy examples of codes.
(2.3.1) The [n, 0, n] zero code of length n contains just the codeword
00 ... 0.
(2.3.2) The [n,n,l1 universe code F; is the dual of (2.3.0.
(2.3.3) The [n, 1, n] repetition code contains all codewords a a ... a, a E Fq'
(2.3.4) The [n, n -1,2] zero-sum code contains all vectors such that
~ Cj = 0; it is the dual of (2.3.3). When q = 2 this is called the even
weight code, sometimes denoted by tEn, since it consists of all binary
vectors containing an even number of I's.

2.4 Cyclic codes. A code is cyclic if whenever COCI ... Cn-I is a codeword so
is Cn-I COCI'" Cn -2' Unless stated otherwise a cyclic code is assumed to be
linear.
Let q = po where p is prime, and let C be an [n, k, d] cyclic code over
Fq • It is convenient to represent a codeword C = COCI"'Cn-1 E C by the
polynomial c(x) = Co + CIX + ... + Cn-I x n- I in the ring Rn(x) of
polynomials modulo xn - 1 with coefficients from F q • Then xc(x)
represents a cyclic shift of c, and so a linear cyclic code is represented by
an ideal in Rn (x) [Mac6, Chap. 7]. This can be generated by a single
polynomial g (x), a generator polynomial for the code. It is easily seen
that g (x) divides xn - lover F q and that the dimension of C is
k = n - deg g (x ).
Examples. For the universe code g(x) = 1, for the zero sum code
g(x) = x - I, and for the repetition code g(x) = I + x + ... + xn-I.

(2.4.0 The [7, 4, 3] Hamming code :Yt, is the binary cyclic code of
length 7 with generator polynomial g (x) = 1 + x + x 3 ; it has generator
matrix
80 Chapter 3

1 0 100 0
o1 1 0 1 0 0
(55)
o0 1 0 1 o '
000 0

where the rows correspond to g (x), xg (x), etc. ;K., has weight distribution
Ao" 1, A3 = A4 = 7, A7 = 1, which we abbreviate to
0' 37 47 7' .

The automorphism group Aut(;K.,) is Klein's group of order 168. The

dual code .Yt; is a [7, 3, 4] code with weight distribution 0' 47, the
codewords forming the vertices of a regular simplex.
(2.4.2) The [8, 4, 4] extended Hamming code ~ is formed by appending
a zero-sum check digit to;K.,. If the coordinates in (55) are labeled 0, 1,
2, ... , 6, and the zero-sum digit is labeled 00, then ~ contains 14
codewords of weight 4, those with l's in coordinates
00 124,00235,00346,00450,00561,00602,00013 (56)

and their complements. An alternative generator matrix is

000011
o0 1 0 0
(57)
o o o o
1 1 1 1

~ has weight distribution 0' 4'48'; equivalently its weight enumerator is

1/18 == x 8 + 14x4y4 + y8,

and Aut(~) is the affine group of order 8'7·6'4 = 1344. This important
code shares many properties with the E 8 lattice. For example ~ is the
smallest nontrivial example of a self-dual Type II code.
Returning to the general theory, let n be relatively prime to q, and let
a be a primitive n-th root of unity, so that

xn - 1 =
n-'
II (x-a i ).
i-O

If m is the multiplicative order of q modulo n, then a E F q , where

q' = qm. Also
g(x) - II (x -a i ) , (58)
iEK

for some set K !:;;; {O, 1, ... , n -1 }, where since g (x) divides xn - lover
F q , i E K implies q i E K [Mac6, Chap. 71. The a i for i E K are called
the zeros of the code. There are good lower bounds on the minimal
Codes, Designs and Groups 81

distance of a cyclic code in terms of its zeros - see van Lint and Wilson
[Linl3l.
(2.4.3) For example, the binary Hamming code .1f" with parameters
n = 2m - I, k = n - m, d = 3 is obtained by taking K =
{ 1,2,4, ... ,2m - 1 }, so that g(x) is the minimal polynomial of o'.
(2.4.4) By appending a zero-sum check digit we obtain the extended
Hamming code .Yt;m with parameters [2 m, 2m - m -I, 41.
2.5 BC" and Reed-Solomon codes. The Bose-Chaudhuri-Hocquenghem
(BCH) code of length n and designed distance dover Fq is the cyclic code
whose generator polynomial g(x) has roots exactly a, 0'2, 0'3, . . . ,O'd - 1
and their conjugates [Ber4], [Linll], [Mac6, Chaps. 7-9], [Pet3]. The
actual minimal distance between codewords is at least the designed
distance and may exceed it.
BCH codes of length n = qm - I are called primitive. For example
primitive binary BCH codes of length n = 2m - I and designed distance 3
are Hamming codes.
BCH codes over Fq of length n = q - I are usually called Reed-
Solomon codes; they were the first to be discovered and are of great
practical and theoretical importance [Bla2], [Mac6, Chap. 10], [Reell
The minimal distance of a Reed-Solomon code is equal to its designed
distance. Thus a Reed-Solomon code over Fq with parameters
[n = q - I, k, d = n - k + 11 has generator polynomial of the form
g(x) = (x -O' b ) (x _O'b+l) ... (x _O'b +d - 2) , (59)

where a is a primitive element of Fq • Usually b is taken to be 0 or 1.

These codes exist for all q and I ~ k ~ n.
It is possible to extend a Reed-Solomon code by adding I, 2 or 3
additional coordinates [Mac6, Chaps. 10, 111. For any q and k there exist
[q, k, q - k + 11 and [q + I, k, q - k + 2] extended Reed-Solomon codes over
Fq; and there also exist [2 m +2,3,2m ] and [2 m +2,2m -I,4] triply-
extended Reed-Solomon codes over Fq when q = 2m .
For any linear code d ~ n - k + I (the Singleton bound [Mac6,
Chap. I]). Codes for which d = n - k + I are called maximal distance
separable (MDS) codes. The preceding Reed-Solomon and extended
Reed-Solomon codes are all MDS codes, as are the codes (2.3.2)-(2.3.4).
The following are two especially interesting examples of MDS codes;
they are also extended Reed-Solomon codes and are self-dual.
(2.5.1) The [4,2,3] ternary tetracode CC4 has generator matrix

[~ ~ _ ~ ~], (60)

weight enumerator
82 Chapter 3

complete weight enumerator

(62)

and automorphism group 2.S 4 , where Sn is the symmetric group of order

n!.
(2.5.2) The [6, 3, 4] hexacode C(J6 is a code over F4 with generator matrix

0 0 1
-
0 1 0 w w (63)
0 0 -
w w

The following matrix defines an equivalent code:

0 0 w w
1 0 w w (64)
[i 0 w w

Using the definition (63), the 64 hexacodewords are obtained from the five
words
-
01 01 ww
-
ww ww ww
- -
00 11 11 (65)
11 ww ww
00 00 00

by freely permuting the three pairs, reversing any even number of pairs,
and multiplying by any power of w. The five words have 36, 12, 9, 6, 1
images respectively. The Hamming weight enumerator of C(J6 is
(66)

and Aut(C(J6) = 3.A6, where An is the alternating group of order Vzn!. (By
making use of the field automorphism that interchanges w with wwe obtain
the slightly larger group 3.S 6J This code is the basis for the MOG
construction of the Leech lattice - see §11 of Chap. 4 and Chap. 11.
2.6 Justesen codes. For large n, BCH codes are weak: for any fixed rate
R, the ratio din -- 0 as n -- 00. In contrast, Justesen codes are an
explicit family of binary linear codes for which both R and din are
bounded away from zero as n increases. We sketch their construction;
further details can be found in [Jusl], [Mac6, Chap. 10]. The starting
point is a Reed-Solomon code Cover Fq' q = 2m , with length
N = 2m - 1, dimension K and minimal distance D = N - K + 1. Let a
be a primitive element of Fq • If c = (cOCI ... CN-I), Ci E Fq , is an arbitrary
codeword of C, let C be the vector
I

(67)
Codes, Designs and Groups 83

of length 2N over Fq'

Now Fq (for q = 2m ) is a vector space of dimension mover F 2, so we
may set up a one-to-one correspondence between Fq and Fr.
Let c" be
obtained from c' by replacing each component by the corresponding binary
m -tuple. Then c" is a binary vector of length n = 2mN; the set of all c"
for c E C is a Justesen code. The binary dimension is k = mK and the
rate is R = kin = KI2N < '12.
Since C is a Reed-Solomon code, a large number of the Ci are nonzero.
Because each ciis multiplied by a different power of a, the binary 2m-
tuples corresponding to nonzero pairs (ci,a i c) are distinct. This is enough
to guarantee that a significant fraction of the components of c" are
nonzero; one can show (see the above references) that the minimal distance
d of the Justesen code satisfies

!!:... > 0.110(1 -2R) (68)

n -

for 0 < R < V2. Justesen codes of higher rate may be obtained by
deleting the appropriate number of coordinates from c".
Justesen's construction has been generalized by Weldon [Weill and
Sugiyama et at. [Sugll-[Sug31 to yield codes with higher rates and
minimal distances.
2.7 Reed-Muller codes. Let the binary weight W(j) of a non-negative
integer i be the number of ones in the binary expansion of i. For
1 ~ r ~ m - 2, the rth order binary punctured Reed-Muller code of
length n = 2m - 1 is the cyclic code whose generator polynomial has as
roots those a i such that 1 ~ i ~ 2m - 2 and 1 ~ W(j) ~ m - r - 1.
The rth order Reed-Muller (RM) code of length 2m is formed by
appending a zero-sum check digit to the rth order punctured Reed-Muller
code, for 1 ~ r ~ m - 2. An rth order RM code has parameters
r
[n = 2m , k = ~ (7) , d = 2 m -r I. (69)
i-O

The Oth, (m -l)th and mth order RM codes are the repetition, even-
weight and universe codes respectively. The (m - 2hh order RM codes are
extended Hamming codes. The rth and (m -r -l)th order RM codes are
duals, for all r. Thus ~ is a first-order Reed-Muller code.
Reed-Muller codes may also be defined in terms of Boolean functions
- see for example [Mac6, Chap. 131. Quite a lot is known about the
weight distributions of RM codes; for example the number of codewords of
minimal weight d = 2 m - r is
m-r-l 2m - i -1
Ad = 2" II 2m - r- i _1
(70)
i-O

[Mac6, Chaps. 13-151.

84 Chapter 3

It is clear from the definitions that BCH and RM codes are "nested".
More precisely, the BCH code of designed distance d is contained in that
of designed distance d - 1, and the rth order RM code is contained in the
(r + I)th order RM code. The rth order RM code is a subcode of the code
obtained by appending a zero-sum check digit to the BCH code of designed
distance 2m - r - 1.
2.8 Quadratic residue codes. Let p, n be primes such that p is a square
modulo n. For p = 2, the most important case, this means that n is a
prime of the form 8m ± 1. The quadratic residue code of length n over
Fp is the cyclic code whose generator polynomial has roots { c/ : i ¢ 0 is a
square modulo n } ([LinlI], [Mac6, Chap. 16]; see also [Le05], [Le06]).
In other words the roots are (Xi, where i is a quadratic residue modulo n.
The dimension of this code is (n + 1) /2. The extended quadratic residue
code is obtained by appending a zero-sum check digit. If n is of the form
4a - 1 then the extended code is self-dual. In particular, extended binary
quadratic residue codes of length 8m are self-dual Type II codes.
For example when q = 2, n = 7 we obtain the Hamming codes :Yt; and
.tfs again.
(2.8.I) The [23, 12, 7] binary Golay code rtl23 is the quadratic residue
code of length 23.
(2.8.2) The [24, 12, 8] (extended) Golay code rtl24 is obtained by
appending a zero-sum check digit to rtl23 . One of many possible generator
matrices is shown in Fig. 3.4. rtl24 has weight distribution
01875912257616759 24 1, (7 I)

or equivalently its weight enumerator is

X24 + 759x 16 y 8 + 2576x 12 y12 + 759x 8y 16 + y24. (72)

o 1 2 ~ 4 5 6 7 8 9 10
1 1 0 1 0 0 0 1 1 1 0 1 1
1 1 1 0 1 0 0 0 1 1 1 0 1
1 0 1 1 0 1 0 o 0 1 1 1 1
1 1 0 1 1 0 1 0 0 0 1 1 1
1 1 1 0 1 1 0 1 o 0 o 1 1
1 1 1 1 0 1 1 0 1 0 0 0 1
1 0 1 1 1 0 1 1 0 1 o 0 1
1 o 0 1 1 1 0 1 1 0 1 0 1
1 0 o 0 1 1 1 0 1 1 0 1 1
1 1 0 0 0 1 1 1 0 1 1 0 1
1 0 1 0 0 0 1 1 1 0 1 1 1
1 1 1 1 1 1 1 1 1 1 1 1 0

Figure 3.4. A generator matrix for the binary Golay code rtl24 . Blank
entries are zero. The pattern of O's and 1's in the right-hand half is
defined by the quadratic residues modulo II.
Codes, Designs and Groups 85

The automorphism group of C€24 is the Mathieu group M 24 (see

Chap. 10). As we shall see, the code C€24 shares many properties with the
Leech lattice.
(2.8.3) When q = 2, n = 47 we obtain [47, 24,11] and [48, 24, 12] codes.
(2.8.4) The [11, 6, 5] ternary Golay code C€ll is the quadratic residue
code of length 11 over F 3 •
(2.8.5) The [12, 6, 6] (extended) ternary Golay code C€J2 is obtained by
appending a zero-sum check digit to C€ II. One of many possible generator
matrices is
1 o 0 000 0 1
0 1 0 000 -1 0 1 -1 -1
o0 1 0 0 0 -1 1 0 -1 -1
(73)
0 0 0 1 0 0 -1 -1 1 0 1 -1
0 0 0 0 1 0 -1 -1 -1 0 1
0 0 o 0 0 -1 -1 -1 0

C€ 12 has weight enumerator

XI2 + 264x 6y 6 + 440X3 y 9 + 24 y 12, (74)

and complete weight enumerator (assuming the all-ones word is present)

f12 = XI2 + yl2 + zl2 + 22(x6y6+y6z6+z6x6)
(75)

The automorphism group of rt'12 is 2. M 12. The Golay codes are studied in
more detail in Chaps. 10 and 11. See also [Goll], [GoI2], [Mac6,
Chap. 201.
(2.8.6) When q = 3 and n = 23 or 47 we obtain [24, 12, 9] and [48, 24,
15] codes. The c.w.e. of the latter is
~ A ijk Xi yj zk ,
i+j+k-48

where the Aijk for i ~ j ~ k are given in Table 3.2. For the [24, 12, 9]
code see [Mal21. For later use we remark that the codewords of maximal
weight in these [12, 6, 6], [24, 12,9] and [48, 24, 15] ternary codes consist
exactly of the rows of a Hadamard matrix and its negative (cf. §2.13).
2.9 Perfect codes. A perfect code is defined by the condition that its
packing radius is equal to its covering radius. The packing radius of an
(n, M, d) code Cover Fq is p = [(d - 1) /21, and C is perfect if and only if
the Hamming spheres of radius p around the codewords are simultaneously
a packing and a covering of F;. (There is no connection with the concept
of a perfect quadratic form [Cas3, p. 281],)
Perfect codes were essentially classified by Tietavainen and van Lint
(see [Linll], [Mac6, Chap. 6]). The list is as follows.
86 Chapter 3

(i) Certain trivial codes (a code with one codeword, the universe code, or a
binary repetition code of odd length).
(ij) Hamming codes, i.e. linear codes with p = lover any field Fq , with
parameters n = (qm - I)/(q - I), k = n - m and d = 3, for m ~ 2.
(iii) Nonlinear codes with the same parameters as Hamming codes (these
codes have not been completely enumerated).
(iv) The binary [23, 12,7] Golay code rtf23 and the ternary [11, 6, 5]
Golay code rtf ll .

Table 3.2. Complete weight enumerator for both the quadratic residue
and Pless [48, 24, 151 ternary codes.

Number
A ijk i j k of terms

1 48 0 0 3
17296 33 12 3 6
190256 33 9 6 6
190256 30 15 3 6
4280760 30 12 6 6
11225104 30 9 9 3
951280 27 18 3 6
38621968 27 15 6 6
209281600 27 12 9 6
94 24 24 0 3
2092816 24 21 3 6
138220984 24 18 6 6
1343397616 24 15 9 6
2777932180 24 12 12 3
210423136 21 21 6 3
3333094864 21 18 9 6
12362073856 21 IS 12 6
20145351688 18 18 12 3
36133419520 18 15 15 3

2.10 The Pless double circulant codes. Pless ([Ple91, [Plell], [Bla7],
[Bla8], [HoI], [Mac6, Chap. 16] [MaI2]) has constructed a [2p + 2, P + 1]
ternary code S2p+2 whenever p is a prime == - 1 (modulo 6). Let
B = (bij) be a (p + 1) x (p + 1) matrix with rows and columns labeled 00,
0, 1, ... , P -1, where
b"""" = 0, b""i = I, hi; = 0,
hi"" = 1 if p == 1 (mod 4), hie<> = -1 if p == -1 (mod 4),
Codes, Designs and Groups 87

bij = 1 if j - i is a square mod p (j ~ j) ,

bij = -1 if j - i is a nonsquare mod p (i ~ j).

Then S 2p+2 has generator matrix [J B). For example the generator matrix
for p = 5 is shown in (73), and S \2 coincides with the Golay code C6'12'
The next five examples are [24, 12, 9], [36, 18, 12], [48, 24, 15], [60, 30,
18] and [84, 42, ~ 211 codes. The c.w.e. of S48 is given in Table 3.2. The
c. w.e.'s of S 24, S 36 and S 60 are given in [Mal2). S 24 and S 48 have the
same c.w.e.'s as the corresponding quadratic residue codes but different
automorphism groups.
The following properties of these codes will be used later. (j) S 2p+2 is a
self-dual (Type III) code containing the all-ones vector. (ii) The numbers
of -1 's, O's or +1's in any codeword are all multiples of 3. (iii) The
codewords of maximal weight contain among them the rows of a
Hadamard matrix and its negative (cf. §2.13), and for S 12, S 24 and S 48
these are the only codewords of maximal weight. (iv) For S\2, S24 and S48
the codewords containing only O's and +1's are of the types on, 1nand
on/21 n/2.

2.11 Goppa codes and codes from algebraic curves. In recent years it has
been shown that very good codes may be obtained from algebraic cun"~s,
although we shall not make use of these codes in this book. See for
example [GopI]-[GopS], [Hir2], UhaI], [Kat2], [Lac1], [Len2], [Lit4],
[ManI], [Mor71, [Ser3], [Tsfl], [Tsf2], [Vlal1-[Vla3], [Zin3].
2.12 Nonlinear codes. Many nonlinear codes are known which contain
more codewords than the best linear codes of the same length and minimal
distance. Some examples are:
(j) Codes obtained froIp Hadamard (§2.13) or conference matrices [Ma~6,
pp. 44, 55], [SI018].
(ii) Nonlinear single-error-correcting codes, such as Golay's (9, 20, 4)
code, Best's (10, 40, 4) code and Julin's (12, 144, 4) code. For these and
other single-error-correcting codes see Chap. 5, [BesI], [Go12], Dull],
[KatI], [LitI], [Mac6, Chap. 2], [RomI], [SI020).
(iii) The (16, 256, 6) Nordstrom-Robinson code [Mac6, p.73], [Norl],
[Seml1, and the Kerdock and Preparata codes and their generalizations
[Kan4], [KerI], [Mac6, Chap. 15], [Prel1.
2.13 Hadamard matrices. A Hadamard matrix H of order n is an n x n
matrix of +1's and -1 's such that
H H lr = nI. (76)

Multiplying any row or column by -1 changes H into another Hadamard

matrix. By this means we can change the first row and column of H into
+1'So Such a Hadamard matrix is called normalized. If a Hadamard
matrix of order n exists then n is 1, 2 or a multiple of 4. Conversely it is
believed that Hadamard matrices exist whenever the order is a multiple of
88 Chapter 3

4, although this has not been proved. A large number of constructions are
known, and the smallest order for which a Hadamard matrix has not yet
been constructed is (in 1987) 428. Connections with coding theory are
described in [Mac6, Chap. 2, §3]. For this purpose one often changes H to
a binary matrix by replacing + l's by O's and -I's by l's. For further
information see [Hall ]-[Ha13], [Har8], [HedlJ, [lt02], [LeolJ, [LevI],
[Linl0], [LonlJ, [Lon2], [Mac6], [SawlJ, [Toni], [TurlJ, [Wa12].

3. t-Designs, Steiner systems and spherical t-designs

3.1 t-Designs and Steiner systems. t-designs are a class of constant weight
codes that satisfy conditions similar to (but weaker than) those satisfied by
perfect codes. Let X be a v-set (i.e. a set with v elements) whose elements
are called points. A t-design is a collection of distinct k-subsets (called
blocks) of X with the property that any t -subset of X is contained in
exactly A blocks. This is also called a t-(V,k,A) design. If A = 1 a t-
design is called an S (t , k, v) Steiner system. The terminology of the
subject comes from the original application of such designs in agricultural
experiments, although they are now used in all branches of science [Dia3].
If X = { 1,2, ... ,v }, each block B !; X may be represented by its
indicator or characteristic vector (c I, . . . , c.), where Cj = 1 if i E B,
Cj = 0 if i ~ B. In this way a t -design becomes a binary code of length v
in which every codeword has weight k. Usually there is no need to
distinguish between a block and its indicator vector.
The parameters of a t -design must satisfy certain arithmetical
conditions. Let PI, ... ,PI E X be distinct points. For 1 ~ i ~ t the
number of blocks containing PI, ... , P j is

Aj=A [ t - i
V - i) / [kt -- ii)'

and the total number of blocks is

(78)

Each point belongs to r(=AI) blocks, where

bk = vr, (79)
A(v - 1) = r(k - 1) if t ~ 2. (80)

Given t, k and v one wishes to find a design with the smallest A, i.e.
with the smallest number of blocks. Of course a necessary condition for a
design to exist is that the right-hand sides of (77) and (78) are integers.
In some cases these conditions are also sufficient (e.g. for Steiner systems
S (2, 3, v), S (2, 4, v), S (2, 5, v) and S (3,4, v», but not always. In general
the precise conditions under which t -designs exist are not known. For
further information about t-designs see [Ass2]-[Ass4], [Cam2], [Ha13],
Codes, Designs and Groups 89

[Han I], [Han2], [Hug2], [Hug3], Dacl1, [LanS], [Lin3], [Mac6, Chap. 2],
[Ragl1, [Ton4], [WiIl8], [Wi1l9].
Many t -designs may be constructed by taking the blocks to be the
codewords of a particular weight, often the minimal weight, in a good
binary error-correcting code (see especially the Assmus-Mattson theorem,
Theorem 22 of Chap. 7). Conversely, good codes can sometimes be
constructed as the linear span of the blocks of a t -design. We may also
obtain t-designs from non binary codes by taking the blocks to be the
locations of the nonzero components of the codewords of a particular
Hamming weight; i.e. the blocks are the distinct supports of the codewords
of a particular weight. In the nonbinary case it is much harder to recover
the code from the design.
Examples. (i) The seven codewords of weight 3 in the Hamming code :Yt,
form a Steiner system 8 (2, 3, 7), and the 14 codewords of weight 4 in .Yes
form an 8(3,4,8).
(iO More generally the codewords of weight 3 in any Hamming code .1f",
n = 2m - I, form an 8 (2, 3, 2m - 1), and the codewords of weight 4 in any
extended Hamming code of length 2m form an 8(3,4,2 m ).
(iii) The supports of the codewords of weight 5 in the ternary Golay code
rc 11 form an 8 (4, 5,1 1), and the supports of the words of weight 6 in rc 12
form an 8 (5, 6,12). The 132 blocks of this design are called hexads (see
Chap. 11).
(iv) The codewords of weight 7 in the binary Golay code rc 23 form an
8 (4,7,23), and the codewords of weight 8 in rc 24 form an 8 (5, 8, 24).
These 759 blocks (or the corresponding codewords) are called octads (see
Chap. 11).
Many other examples will be found in the references cited above. Until
recently t = 5 was the largest t for which a t-design was known ([Ass3],
[Denl1, [Mi14], [Wit2], [Wit3]), but in 1983 Magliveras and Leavitt
constructed 6-designs [Magl1, [Mag2], and Teirlinck [Teil1 has now
shown that t-designs exist for all t.

3.2 Spberical t-designs. Just as t-designs are a special class of constant

weight codes, so spherical t -designs are a special class of spherical codes
(cf. §2.3 of Chap. 1). The original motivation for studying these objects
came from the numerical evaluation of multi-dimensional integrals. The
integral of a polynomial function over the sphere n n may be approximated
by its average value at the code points; if the code is a spherical t-design,
the approximation is exact for polynomials of degree ~ t. Formally, a
spherical t-design X is a finite subset of n n with the property that

f f(~)dw(~) =
On
I~I ~
X EX
f(x) (81)

for all polynomials f of degree ~ t, where w(~) is Lebesgue measure on

the sphere, scaled so that the total measure of n n is 1. Ot would require
90 Chapter 3

too much of a digression to explain why these are called t -designs - see
[Delli], [DeI16J.)
Many examples of spherical t -designs may be found by using the
vectors of a particular norm, often the minimal norm, in a good lattice
packing. For example the 240 minimal vectors in the E8 lattice (when re-
scaled so as to lie on n 8) form a spherical 7-design, and the 196560
minimal vectors in the Leech lattice form a spherical II-design. The
appropriate value of t for such designs may sometimes be determined by
using theorems of Sobolev (see §4.2) or Venkov (see Chap. 7, §7).
Given nand t one wishes to find a spherical t -design with the smallest
number of points. Delsarte, Goethals and Seidel [Del161 show that the
number of points satisfies

Ix I ~ [n: ~ ~ I ) + [n : ~ ~ 2 ), for t = 2s , (S2)

Ix I ~ 2[
n
n+ _s -I I) ' for t = 2s + I. (S3)

If equality holds in either (S2) or (S3) the spherical t -design is said to be

tighl. Very few tight I-designs exist, and in particular Bannai and
Damerell [Ban7], [BanS1 have shown that, for n ~ 3, there are no tight
(2s) -designs with 2s ~ 6 and no tight (2s + I) -designs with 2s + I ~ 9
except for the just-mentioned II-design obtained from the Leech lattice.
(The latter is also known to be unique - see Chap. 14.) For
generalizations of these results to other spaces see Chap. 9 and [Ban9],
[BanIO], [HogI1-[Hog31, [Neu21.
On the other hand Seymour and Zaslavsky [Seyl1 have shown that
(non-tight) spherical I -designs exist in n n for all values of nand t. For
further information about spherical I-designs see [Banll-[BanI3), [Dell II,
[DeI15], [DeI16], [Dun5], [Goe5], [Goe61, [Honl), [Sob21, [Sob31.

4. The connections with group theory

4.1 The automorphism group of a lattice. First of all, lattices give rise to
groups, their symmetry groups. The automorphism group (or symmetry
group) Aut(A) of a lattice A is the set of distance-preserving
transformations (or isometries) of the space that fix the origin and take the
lattice to itself. For a lattice in ordinary Euclidean space, Aut(A) is finite
and the transformations in Aut(A) may be represented by orthogonal
matrices. Let A have generator matrix M. Then an orthogonal matrix B
is in Aut(A) if and only if there is an integral matrix U with determinant
± I such that
UM=MB (S4)

(cf. Eq. (22) of Chap. I). This implies U MBM tr A-I, where A is the
=
Gram matrix. On the other hand the set of integral matrices U for which
Codes, Designs and Groups 91

there exists an orthogonal matrix B satisfying (84) forms an integral

representation of Aut(A).
For example, the automorphism group of the hexagonal lattice in
Fig. 1.3a is a dihedral group of order 12, generated by a rotation through
60 and a reflection in a line joining the center of two spheres. The
0

automorphism group of the simple three-dimensional cubic lattice contains

all permutations and sign changes of the three coordinates, and has order
23 3! = 48. A lattice and its dual have the same automorphism group.
On some occasions we consider the infinite group of all distance-
preserving transformations of the underlying space (the affine
automorphisms) that take the lattice to itself. This is obtained by
adjoining the translations in lattice vectors to Aut(A).
The automorphism groups Aut(A) of the lattices in Table 1.1 in up to 8
dimensions are especially interesting: they have subgroups of low index
that are reflection groups (or Coxeter groups) - see §2 of Chap. 4. In
higher dimensions other remarkable groups appear. For example the
automorphism group of the 24-dimensional Leech lattice is the group Coo
(or . 0) of order 8315553613086720000, which will be studied in
considerable detail in Chaps. 10 and 11. The corresponding infinite group
(obtained by adjoining translations by lattice vectors) is called Co"", (or
. 00). Actually Coo is interesting for other reasons than its size - the 24-
dimensional lattice D24 has a much larger group, of order

224 '24! = 10409396852733332453861621760000.

The interest of Coo lies in its connection with the classification of finite
groups. Any finite group can be built up out of certain special groups
called simple groups, and the classification of all finite simple groups is a
project that was completed in 1982 after occupying the attention of a great
many mathematicians for over fifty years (see for example [Gor2]-[Gor5]).
When Coo was discovered in 1968 it produced three new simple groups
(Co), CO 2 and C0 3 - see [Con2], [Con3] and Chap. 10). Furthermore
the largest of the "sporadic" simple groups (and one of the last to be
found), the Friendly Giant or Fischer-Griess "Monster" simple group, of
order

808017424794512875886459904961710757005754368000000000

was constructed by R. L. Griess in 1981 using the Leech lattice ([Gri41,

[Gri5]). A simplified construction is given in Chap. 29.
The automorphism group of a lattice in a Lorentzian space is usually
infinite, the even unimodular lattice 11 25 ,1 being a notorious case - see
Chaps. 26-28. One of the surprises is that even those groups have been
illuminated by studying the Leech lattice.
Error-correcting codes (§2) and t -designs (§3) also give rise to groups.
The automorphism group of a code was defined in §2.2, and similarly the
92 Chapter 3

automorphism group of a t -design is the set of all permutations of the

points that takes the set of blocks into itself.
We follow the ATLAS conventions [Con 16, p. xxl, and write A x B
for a direct product; A.B or AB for a group with a normal subgroup
isomorphic to A, for which the corresponding quotient group is isomorphic
to B; A:B for the case of A.B which is a split extension or semi-direct
product; and A· B for the case when A.B is not a split extension.
As general references for groups we mention [Cohll, [Coh21, [Con161,
[Cox201, [Cox211, [Cox281, [Gor2l-[Gor41, [Hupll, [She2l.
4.2 Constructing lattices and codes from groups. We may also take the
opposite point of view, beginning with a finite group G, and attempting to
construct objects of various kinds of which G is the automorphism group.
Three of the most important constructions are the following.
(j) The construction of lattices from integral representations. Given a
representation of a group G by integral matrices U, it is possible to
construct a lattice invariant under G in the following way. We let the
group act on quadratic forms f W = ~ A ~Ir by sending A to UAU 1r , f (~)
to f (~U) (so that U multiplies the generator matrix of the lattice on the
left, as in Eq. (22) of Chap. 1 and Eq. (84». Then if f (~) is an arbitrary
positive definite quadratic form, the quadratic form
~ f(~ U)
UEG

is invariant under G, as is the corresponding lattice. For instance, the

group of order 3 generated by [_ ~ _ ~ 1 permutes the three quadratic

forms

and so preserves their sum 4(~ f + ~ 1 ~2 + ~D, which is a quadratic form for
the hexagonal lattice (Eq. (25) of Chap. 2). In this example the resulting
lattice is actually invariant under a larger group than we started with (the
full group has order 12, as we saw in the previous section). The study of
the integral representations of a given group G is to a large extent the
study of lattices invariant under that group. (See [BrolOl-[Bro 131, [Biilll,
[Con42l, [Crall, [Cra21, [DadIl, [FeiIl, [GudIl, [GusIl, [HemIl, [Plell-
[PleSl, [Rei2l-[Rei41, [RoglOl, [Shull, [Shu2l, [Th03!.)
(ij) The construction of codes from permutation representations.
Similarly, given a representation of group elements by permutations, we
can work modulo 2 and obtain a representation of G on a vector space V
over the field of order 2. The invariant subspaces (the subs paces of V
taken into themselves by every group element) are then all the binary
codes C for which G is a subgroup of Aut(C). Similar methods produce
codes over arbitrary fields. This technique has been used in [BroOl-[Br021,
[CaI9l, [KnaIl, [ParIl. Important information about these codes can be
obtained from the theory of modular representations of groups.
Codes, Designs and Groups 93

Calderbank and Wales [Ca19] used this idea to construct a binary code of
length 176 and dimension 22 whose automorphism group is the Higman-
Sims group. Brooke [BroO]-[Br02] has found all binary codes obtainable in
this way from the primitive permutation representations of many
interesting groups. See also [Kna 11. [Phe 11.
Gii) The construction of spherical codes and designs from orthogonal
representations. Given a representation of group elements by n x n
orthogonal matrices B, we may construct a spherical code X c n n by
choosing an initial point Xo E n n and setting

X = {xoB : BEG}, (85)

the orbit of Xo. These codes are sometimes called group codes, and have
been investigated in [Bill], [Bil3], [Bla4], [Bla6], [Bla9], [Dow11-[Dow3],
[Ing11, [Kar3], [Sle3], [Sle4].
There is a beautiful connection between these codes and invariant
theory, first discovered by Sobolev. Let G be a finite group of n x n real
or complex matrices B. The elements of G act on polynomials
f(x) = f(Xh ... ,xn ) by B 0 f(x) = f(Bx l ' ) . Polynomials for which
f (x) = f (Ex I') for all BEG are said to be invariant under G, and the
set of all invariant polynomials forms a ring RG.
Now suppose G is a group of real orthogonal matrices. Of course the
squared length cp (x) = x f + ... + x; is invariant under G. Sobolev
showed that if the degree of the first invariant that is not a polynomial in
cp(x) is t + 1, then every orbit under G is a spherical t -design (see [Sob2]
and the other references on spherical t-designs given at the end of §3.2).
Examples. The automorphism group of the E8 lattice, namely the Weyl
group of type E 8, is a reflection group (Chap. 4, §2). The ring of
invariants RG has a basis consisting of homogeneous polynomials of
degrees
2,8, 12, 14, 18,20,24,30, (86)

where the invariant of degree 2 is cp (x) = xf + ... + xi [Cox28,

Table 10, p. 1411. It follows from Sobolev's theorem that every orbit
under this group is a spherical 7-design. In particular the set of 240
minimal vectors in E8 is a spherical 7-design, as we remarked in §3.2.
The ring of invariants of the automorphism group of the Leech lattice
has been investigated in [Huf5], where it shown that the first invariant not
a polynomial in cp has degree 12. Therefore every orbit under this group
(for example the set of 196560 minimal vectors) is a spherical II-design.
These examples are studied further in Chaps. 13, 14. Venkov's theorem
(Chap. 7, §7) generalizes these examples. We shall see other applications
of invariant theory in Chap. 7.
 4
Certain Important Lattices
and Their Properties
J. H. Conway and N. J. A. Sloane

This chapter describes the properties of a number of important lattices,

including the cubic lattice zn, the root lattices An, D n , E 6 , E 7 , E 8 , the
Coxeter-Todd lattice K 12, the Barnes-Wall lattice A 16, the Leech lattice
A 24 , and their duals. Among other things we give their minimal vectors,
densities, covering radii, glue vectors, automorphism groups, expressions for
their theta series, and tables of the numbers of points in the first fifty
shells. We also include a brief discussion of reflection groups and of the
technique of gluing lattices together.

1. Introduction
In this chapter we introduce a number of important lattices that will be
used throughout the book, namely zn (n ~ 1), An (n ~ 1), Dn (n ~ 3), E 6,
E 7 , E 8, K 12 , A 16, A24 and their duals, and give a summary of the basic
properties of each lattice. Proofs of most of these results will be given in
later chapters. Some of the reasons for the importance of these lattices can
be found in Table 1.1 of Chap. 1: they include many of the best packings,
coverings and quantizers, as well as the lattices with the highest kissing
numbers, in dimensions up to 24. It is likely that they will also be good
candidates for the solution of other problems the reader may have in mind
in these dimensions.
The "root lattices" zn, An, D n , En are treated first, in §§5-8. The
terminology arises because these lattices are generated by the root systems
of certain Lie algebras, although we shall say very little about this
connection (cf. [BouI], [HumI]). These lattices are also closely related to
reflection groups, as we discuss in §2. Another reason the lattices Zn, An,
Dn , En are the first to arise in many investigations is that any integral
lattice generated by vectors of norms 1 and 2 is a direct sum of these
lattices (Witt's theorem - see §3). There is no corresponding theory for
Certain Important Lattices and Their Properties 95

larger norms. "Gluing theory" (§3) describes how component lattices may
be combined to produce lattices in higher dimensions.
The remaining lattices treated in this chapter (in §§9-11) are K I2 , AI6
and A 24, all of which occur inside the Leech lattice A 24 • In fact Z i;; AI>
A 2, A3 i;; D 3 , D 4 , D s, E 6 , E 7 , E s, AI6 and A24 are all laminated lattices
(see Eq. (25) of Chap. 1 and Chap. 6), and thus are found inside A 24 .
Two important lattices not included in this chapter are the 48-
dimensional self-dual lattices P4Sp and P 4Sq • A summary of their
properties will be found in Example 9 of Chap. 7.
The theta series of a lattice (§2.3 of Chap. 2) tells how many points
there are at each distance from the origin, i.e. gives the number of points
in each spherical layer or shell. The theta series of the lattices in this book
can be expressed in terms of simpler functions such as the Jacobi theta
series 8 2, 8 3 and 84 • These are defined in §4 together with various other
terms.
We have included tables of the first 50 or so theta coefficients of most
of these lattices. Such tables are useful in studying the lattices,
investigating the connections with number theory (§2.3 of Chap. 2), and
because configurations of lattice points often form excellent codes (§2.3 of
Chap. I, §1 of Chap. 3). Individual shells form spherical or constant
energy codes, while the set of lattice points lying inside or on a shell forms
a bounded energy code or spherical cluster.
We also give brief descriptions of the Voronoi cells (or nearest neighbor
regions) around the lattice points (§1.2 of Chap. 2); further information
will be given in Chap. 21. Although much of the material in this chapter
may be found elsewhere (for example [Boul1, [Con271, [Con28], [Con37],
[Cox20], [Intl1, [Nie2], [SI012], [SI019], [Te02], [Te03]), some of the
theta series and tables are published here for the first time. (Some related
tables are given in [Fral1, [Mit5], [New21, [Pry1].) Further information
about the packings in two and three dimensions (including more extensive
tables, both for the packings given here and other translates, plus tables of
the vectors in the first few layers) will be found in [SI017], [SI019], [Te02],
[Te03].

2. Reflection groups and root lattices

Figure 4.1 shows a kaleidoscope whose three mirrors (or walls) cut the
sphere in a spherical triangle having angles 7r /2, 7r /3 and 7r /5. The
reflections in these walls generate a group of order 120, called the
reflection group [3,51. The whole surface of the sphere is divided into 120
triangles, one for each group element.
This is an instance of a finite, or spherical, reflection group.
(Reflection groups are often also called Coxeter groups.) In general such
a group (if irreducible) is generated by reflections in the walls of a
spherical simplex, all of whose dihedral angles are submultiples of 7r. (The
reducible reflection groups are direct products of irreducible ones.) The
96 Chapter 4

Figure 4.1. Kaleidoscope intersecting sphere in a spherical triangle. The

whole surface of sphere is divided into 120 copies of this triangle.

infinite cone bounded by the reflecting walls or hyperplanes (i .e. the

kaleidoscope) is a fundamental region for the reflection group. The group
may be succinctly described by a Coxeter-Dynkin diagram, having one
node for each wall, two nodes being joined by a line marked p when the
corresponding walls are at an angle of 7f /p. Certain abbreviations are
customarily adopted for lines labeled with small values of p, as shown in
Fig. 4.2. For example the diagram for the preceding example is shown in
Fig. 4.3 .
If Ri is the reflection in the ith wall of the fundamental region then
from the diagram we can read off a set of defining relations (or a
presentation) for the group, namely
R? = (R i Rj)Pij = 1 (;,j = 1, . . . , n),

where 7f /Pij is the angle between the ith and jth walls. A celebrated
theorem of Coxeter [Cox4], [Cox28, §9.3] asserts that every finite group
with a presentation of this form is a reflection group.
The classification of the finite reflection groups has a complicated
history [Kan31. The elegant complete enumeration of the finite reflection
groups in terms of diagrams is due to Coxeter [Cox2J, [Cox3), [Cox20J .
However Mitchell [Mit! ]-[Mit4J had already solved an essentially harder

T HE LIN ES

ARE ABBR EVI ATED TO

o o I i i )

Figure 4.2. Conventions for Coxeter-Dynkin diagrams. (See also

Chap. 27.)
Certain Important Lattices and Their Properties 97

= 0 - - - - <0)-----"5_0
ABC

Figure 4.3. Diagram for reflection group [3,51.

problem. We shall concentrate on the crystallographic reflection groups,

for which p takes only the values 2, 3, 4 and 6, since these are associated
with lattices. They are given in the fourth column of Table 4.1. The only
finite indecomposable non-crystallographic reflection groups are shown in
Fig. 4.4.
We can specify each reflecting hyperplane by a vector perpendicular to
it, called a root vector (or simply a root). The root vectors perpendicular
to the walls of the fundamental region are the fundamental roots for the
group, while the entire set of root vectors is called a root system. The
entire root system is found by taking all images of the fundamental roots
under the group. The lattice generated by the root system is called a root
lattice, and the fundamental roots form an integral basis for the root
lattice.
Conversely, given a lattice A, we define a root (vector) for A to be a
vector rEA for which the associated reflection
x 'r
x--+x-2--r
r'r

is a symmetry of A. The group H generated by the reflections in the roots

is the reflection subgroup of Aut(A). If A is integral and unimodular the
roots are just the vectors of norm 1 or 2 in A, which are called short and
long roots respectively. The reflecting hyperplanes corresponding to all the
roots of A partition the space into fundamental regions for H. The roots
corresponding to the walls of anyone fundamental region are a set of
fundamental roots for the lattice (and for H). If A is not a root lattice
then in general the fundamental roots are not a basis for A.
All the indecomposable finite root systems, and so all the
indecomposable crystallographic finite reflection groups, are shown in
Table 4. I. The first column of the table is the usual name for the root
system, the second column is our name for the root lattice generated by the
roots, and the third column gives the determinant d of this lattice. The
fourth column describes the fundamental roots, the reflections in which

~
(p .. 2,3,4,6) ORDER 120 ORDER 14400
ORDER 2p

Figure 4.4. Indecomposable non-crystallographic reflection groups.

98 Cha pter 4

Table 4.1. Indecomposable finite root systems or equivalently,

indecomposable crystallographic finite reflection groups.

ROOT FI NITE INF I N ITE

SYSTEM LATT ICE DET GROUP GROu P

~
Zo
~-~
~ -~
Co Do ~--o--a::(::@

Do Do 4 ~-<>-c
~-~
G2 A2 CE)3>

F4 04 ~ ~

E6 E6 ~ ~
E7 E7 ~ ~
E8 E8 I~ ~

generate the corresponding finite reflection group. The right-most roots in

the fourth column have norm 2. The lengths of the other roots are
determined by the convention that a k -fold edge whose arrowhead (if any)
points from r to s indicates that N (r) = k N (s).
Note that the conventions of Fig. 4.2 imply that if roots rand s are
joined by a k - fold edge then the angle () between them is given by
4cos 2 () = k and cos () ~ 0, i.e. by
k: 0 2 3
(): 90 " 135 " 150"

The final column contains the corresponding (infinite) Euclidean

reflection group, sometimes called the affine Weyl group, obtained by
adjoining the translations by root vectors. This is again a reflection group,
whose diagram is obtained by adjoining one more node (called the
extending node) to the diagram for the finite group. If we had started
from the diagram for the infinite group, there would be exactly d nodes,
called the tips or special nodes, whose removal would leave a diagram for
the corresponding finite group. The tips are indicated in the diagrams by
double circles.
The product of all the generating reflections is called a Coxeter
element, and all the Coxeter elements have the same order h, the Coxeter
Certain Important Lattices and Their Properties 99

number. The Coxeter number has many other interpretations. For

example the total number of roots is nh. Thus for the lattices An, Dn and
En (when all roots have the same length), nh is the kissing number.
Note on Bn. en. G 2 and F 4• We explain why only An' Dn and En (but not
B n, en, G 2 or F 4) appear in the rest of the book. The fundamental root
systems for the Lie algebras Bn and en may be taken as
en
o o 0 2 0 o 0
-1 1 o 0 -1 1 o 0
o -1 1 0 o -1 1 0

Since these differ only by scalar factors, the corresponding finite reflection
groups are identical. However the root lattices generated by these roots
(which are respectively zn and Dn) are different, as are the infinite groups
obtained by adjoining the translations in these lattices to the finite groups.
Similar phenomena occur in the cases G 2 and F 4 • It may be observed that
every root lattice is generated by its roots of shortest length. Thus our list
of lattices includes An, Dn and En, which are root lattices for which all
roots have the same length, but not B n, en, G 2 and F 4 • In fact the root
system for G 2 consists of the vectors of the first two norms (2 and 6) in the
root lattice A 2 • Similarly the root system for F4 consists of the first two
layers (of norms 2 and 4) in the root lattice D 4 .
For further information about reflection groups, crystallographic groups
(including the Bieberbach theorems) and related topics see [Aus11, [Bie11,
[BouI], [Bus11, [Cox20), [Cox28), [Del7], [Fri2), [Gr03], [Haz11, [Hi13),
[Hum11, [Oli2), [Vin11-[VinI5), [Zas11.

3. Gluing theory
Gluing theory is a way to describe the general n -dimensional integral
lattice L that has a sublattice which is a direct sum
L\ ED L2 ED ..• ED Lk

of given integral lattices L \ • . . . • Lk of total dimension n. The typical

vector of L can be written
Y = Y\ + Y2 + ... + Yk,
where each component Yi is in the subspace spanned by Lj, although not
necessarily in Li itself. What are the possibilities for Y\ •...• Yk?
The inner product of Yi with any vector of L j is an integer, since it is
the same as the inner product of Y with that vector. This shows that Yj
must be a member of the dual lattice Lt.
Plainly any Yi can be altered by adding a vector of L i , so we may
suppose that Yi is one of a standard system of representatives for the cosets
of Li in Lt. These representatives are called the glue vectors for L j • It is
100 Chapter 4

usual to choose the glue vectors to be of minimal length in their cosets.

The quotient group LtlL j is called the dual quotient, or glue group for L j •
As remarked in §2.4 of Chap. 2, its order is equal to det L j •
So each possible lattice L is generated by L, EB ... EB Lk together
with certain vectors (3), where each Yj is a glue vector for L j , and we need
only check that the various vectors (3) have integral inner products with
each other and are closed under addition modulo L, EB ... EB L k . We
describe such a lattice L informally by saying that the components
L, , ... , Lk have been glued together by the glue vectors (3). It may
happen that there is a glue vector (3) in which only one Yj is nonzero,
when we say that the component L j has self-glue, and that Y is a self-glue
vector. For example Dn is self-glued to form Dn+.
There is an important special case.
Theorem 1. If a unimodular lattice L is formed by gluing together two
lattices L, and L2 in such a way that there is no self-glue, i.e. if
L, = (L,®R) n L, L2 = (L 2®R) n L,
then the dual quotients L; I L, and L; I L2 are isomorphic groups.
The isomorphism is given by y, + L, - Y2 + L2 whenever there is a
glue vector y = y, + Y2'
There is a standard choice of glue vectors [0], [1], ... , [d -1] for
each root lattice of determinant d, and we use [a, ... ak] as an
abbreviation for the glue vector (3) in which y, = [ad for L" Y2 = [a2]
for L 2, ••.• (There is one glue vector for each tip node.}
In our applications of gluing theory it will usually be obvious that all
automorphisms of L will permute the lattices L" ... ,Lk (often because
L, EB ... EB Lk will be the part of L generated by vectors of norms 1 and
2). In these circumstances there is a simple description of the
automorphism group G (L) of L.
The group of all permutations of the L j that arise from automorphisms
in G (L) we shall call G 2 (L). It is isomorphic to the quotient group
G (L) IG o" where Go, consists of just those automorphisms that give the
trivial permutation.
Let Go(L) be the normal subgroup of Go, consisting of those
automorphisms which, for every i, send each glue vector Yj into a vector in
the same coset Yj + L j , i.e. which fix the glue vectors modulo the
components. Then GOl/Go(L) is isomorphic to a permutation group acting
on the glue vectors of each component: we call this permutation group
G,(L). Thus the full group G(L) is compounded of the groups Go(L),
G , (L), G 2 (L), and has order

where gj (L) is the order of Gj (L). Also Go (L) is the direct product of the
groups Go(L). But in general G,(L) is only a subgroup of the direct
product of the G, (L) and therefore must be computed directly for each L.
Certain Important Lattices and Their Properties 101

For example there is a unique 17-dimensional unimodular lattice with

no vector of norm 1 (see Chap. 16). It is obtained from All $ E 6 by
adjoining the glue vector

2 10 -10 2] [ 22 _1 4 ]]
[2,11 = [ [ 12' U ; 0, 3 ' -3- , 0
which is the concatenation of the glue vectors Yl = [2] for A 11 and
Y2 = [11 for E 6. For this lattice Go is the product of the Weyl groups
G O(A 11 ) and GO(E 6 ); G l has order 2 (we can simultaneously negate Yl and
Y2); while G 2 is plainly trivial (no automorphism can interchange All with
E6)'
Witt's theorem [WitS], [KneS] tells us that, for any integral lattice L,
the sublattice generated by vectors of norms 1 and 2 is a direct sum of root
lattices. Thus the root lattices are particularly appropriate choices for the
L i . For a root lattice X n , Go is the Weyl group (the subgroup generated
by the reflections in the minimal vectors of Xn), and G l is the symmetry
group of the (ordinary) Coxeter-Dynkin diagram (the so-called graph
automorphism group). The symmetry group of the extended Coxeter-
Dynkin diagram has order gl d.
The outstanding application of gluing theory is provided by Niemeier's
list of even unimodular 24-dimensional lattices - see §3 of Chap. 16.
Many other examples are also given in Chaps. 16, 17. Gluing theory has
also proved useful in coding theory [Con21], [Con24], [Leo71, [Le08],
[Ple18}.

4. Notation; theta functions

In the following sections the subscript n on the symbol for a lattice L
usually indicates its dimension, M is a generator matrix for the lattice,
A = MM tr is a Gram matrix, det is the determinant (the square of the
volume of a Voronoi cell), T is the kissing number, h is the Coxeter
number (see §2), p is the packing radius, R is the covering radius,
A = Vn pn 1.J&t is the density, 0 = AIVn is the center density and
e = Vn R n1.J&t is the thickness, where Vn = 7r n12/(n 12)! is the volume of
a sphere of radius 1 (see Chap. 1). Diamond brackets are sometimes used
to indicate the generators of a lattice, as in A 2 = < (1, -1,0), (0,1, -1) >.
The symbol $ indicates a direct sum.
The glue group is L * I L, of order det L, and the glue vectors
(representatives for the cosets of L in L *) are indicated by square brackets
[d. The automorphism group G (L) has order g = gOgb where go, gl are
the orders of the groups Go(L), G 1 (L) (see §2).
We use obvious abbreviations for the components of vectors. For
example the glue vector

1 1 1 1 1 1
[ 4' 3 3]
[1] = 4' 4' 4' 4' 4' -4'-4
102 Chapter 4

for E 7 would be abbreviated to

The norm of a vector x is its squared length x . x. The minimal norm

of L is {min x' x : x E L, x ¢ O}, which is 4p2. The theta series of L is
0 L (z) = ~ qX'x = ~N(m)qm, q = e 7fiz , Im(z) > 0, (5)
xEL

where N(m) (sometimes written N m) is the number of vectors in L of

norm m (see §2.3 of Chap. 2). Thus the coefficients in the expansion of
oL (z) in powers of q give the numbers of vectors in the successive shells of
L around the origin. We also use Eq. (5) to define the theta series of a
non lattice packing L (for example the translate of a lattice).
4.1 Jacobi theta functions. The theta series of the lattices encountered in
this book can be expressed in terms of the Jacobi theta function

m-- oo

(References are given at the end of this section.) For many purposes it is
enough to work with the simpler theta functions /II' (z), /1 2(Z), /I /Z), /1 4(z)
given by

m-- oo

= 2ql/4 - 6 q 9/4 + 10q25/4 _ 14q49/4 + '"

= 2q 1/4 + 2q 9/4 + 2q25/4 + .,.

(8)

m-- oo

= 1 + 2q + 2q4 + 2q 9 + (9)
Certain Important Lattices and Their Properties 103

= I - 2q + 2q4 - 2q9 + (10)

where q = e1fiz . Other useful functions are

l/;k(Z) = e.. iz/k283 [ '1rkZ IZ] = m~oo q(m+l/k)2,

for k = ± I, ± 2 , ... , and

rPo(Z) = 82(2z) 8 2(6z) + 8 3(2z)8 3(6z)

= Y2 {8 3(z /2) 8 3(3z /2) + 8 4 (z /2) 84 (3z /2)}
= I + 6q2 + 6q 6 + 6q 8 + 12q14 + (12)
rPI(Z) = 8 2(2z)8 3(6z) + 8 3(2z) 82(6z)
= 1/2 82(z /2) 82(3z /2)
= 2q 1/2 + 2q 3/2 + 4q 7/2 + 2q 9/2 +

These functions are related by a labyrinth of identities. The deepest of

these is due to Poisson (1827) and Jacobi (1828) - see [Whtl, p.47S]'
[BeI2, p. 4]:

(14)

(where the square root is the principal value>. This implies

8 1' (- I / z) = (z Ii) 3/2 8 1' (z ) , (15)

8 2 (- I / z) = (z / j) 1/2 8 4 (z) , (16)

8 3 ( - I / z) = (z Ii) 1/2 8 3 (z) , (17)

rPo(-l/z) = zr::; rPo(z/3), (18)

iv3

and Jacobi's formula for the theta series of the dual lattice:
eA. (z) = (det A) 1/2 Ci /z)n/2 eA(-1 /Z).

One may regard Eq. (14)-(19), as well as the MacWilliams identity for
weight enumerators of codes (Eq. (50) of Chap. 3), as consequences of the
general version of the Poisson summation formula, which states that the
sum of a function over a linear space is equal to the sum of the Fourier
104 Chapter 4

transform of the function over the dual space [Dyml, Chap. 2, §11.3],
Ugul, p. 44], [Lool, p. 153], [Serl, Chap. VII, §6].
Other useful identities, arranged roughly by degree, and to show the
symmetry<° between 8 2, 8 3 and 84, are:
8 2 (z+1) = ..fi8iz), 8 3 (z+1) = 84 (z), 84 (z+1) = 83 (z), (20)

8 2 (z) + 83 (z) = 8 3 (z /4), -8 2 (z) + 8 3 (z) = 84 (z /4), (22)

83 (z) + 84 (z) = 28 3 (4z), 83 (z) - 84 (z) = 28 2 (4z), (23)

(24)

(25)

(26)

(27)

(28)

(!) Equations (20), (21) show that the substitutions z - z + 1, z - - 1/ z

permute 82, 8 3, 84 up to trivial factors.
Certain Important Lattices and Their Properties 105

These functions may also be expanded as infinite products (Jacobi's triple

product identities):

II O_q2m)(l+q2m-Ie2i~
00

83(~lz) = O+q2m-Ie-2i~), (32)

m-I

II O_q2m)3,
00

8 1'(z) = 2ql/4 (33)

m-I

II
00

82(z) - 2ql/4 O_q2m) O+q2m)2, (34)

m-I

83(z) = II O-q2m) O+q2m-I)2, (35)

m-I

II
00

84(z) = O_q2m) O_q2m-I)2. (36)

m-I

Another important product is

00

~24 - q2 II O_q2m)24 (37)

m-I

}; T (m)q2m ... q2 - 24q4 + 252q 6 - 1472q 8 + (39)

m-O

in which the coefficients T (m) are called Ramanujan numbers. (The first
300 coefficients are given in [Leh21 See also Theorem 19 of Chap. 7.)
Some values of if;k(Z) (Eq. (11» can be expressed in terms of other
functions:

if;1(Z) = 8 3(z), if;2(Z) = 8 2(z), if;3(Z) = ~(83(~)-83(Z»' (40)

if;4(Z) = ~ 82(:), if;6(Z)'" ~ (8 2 ( ~) - 8 2(z», (41)

if;-k(Z) = if;k(Z). (42)

For further information see [BeI2], [Gr02], [Har4], Ugul1, [Kra3],

[Muml1, [Radl1, [Ran6], [Raul1, [Rau2], [Tanl], [Whtll
Rescaling. It is often necessary to rescale a lattice, changing A to
A' - eA - {ex:x E A} for some constant e E R. The parameters of A and
A' are related as follows: M' = eM, A' = e 2 A, det' == e 2n det, (minimal
norm)' = e 2 (minimal norm), p' - ep, R' = eR; the kissing number T,
densities ~ and li, and thickness e are unchanged; (A') * = e- I A *, and
(43)
106 Chapter 4

5. The n-dimensional cubic lattice Zn

The set of integers ... , - 2, - I, 0, I, 2, 3, ... is denoted by Z, and
zn = {(Xl, ... ,Xn ) :x,EZ)

is the n -dimensional cubic or integer lattice. (Z2 is better called the

square lattice, as seen in ordinary graph paper.) As generator matrix M
we may simply take the identity matrix. Then det = I, minimal norm = I,
kissing number T = 2n, and the minimal vectors are (0, ... , ± I , ... , 0).
The packing radius p = 1/2, the covering radius R = -rn
/2 = p -rn,
the
density Ll = VnTn and the center density 0 = Tn. Thus Z has density
Ll = 1, but the densities of Z2, Z3 and Z4 are only 7r/4 = 0.785 ... ,
7r /6 = 0.524 ... and 7r 2/32 = 0.308.... A typical deep hole is (1/2, V2, ... ,1/2),
and the Voronoi cells are cubes. zn is self-dual. Its automorphism group
consists of all permutations and sign changes of the coordinates, and has
order 2nn!. (This is the Weyl group of Bn.)
The theta series of Zn is (h (z)n, and the theta series of the translate
zn + (oa 1/2 n- a ) is
(44)

(§2.3 of Chap. 2). Tables of the numbers of points in the first 50 shells of
Z2, Z3 and their translates are given in Tables 4.2, 4.3. In Table 4.2 the
columns headed N<O), N(I), N(2) give respectively the numbers of points of
norms m, m + 1/4, 2m + 1/2 in Z2 with the origin of coordinates at the

Table 4.2. Numbers of points in the first shells of the square lattice Z2
with respect to various choices for the origin. See text for full
explanation.

m N(O) N(I) N(2) m N(O) N(I) N(2) m N(O) N(I) N(2)

0 I 2 4 17 8 0 0 34 8 4 8
I 4 4 8 18 4 4 8 35 0 0 0
2 4 2 4 19 0 0 0 36 4 8 16
3 0 4 8 20 8 2 4 37 8 4 8
4 4 4 8 21 0 8 16 38 0 4 8
5 8 0 0 22 0 4 8 39 0 4 8
6 0 6 12 23 0 0 0 40 8 0 0
7 0 4 8 24 0 4 8 41 8 0 0
8 4 0 0 25 12 4 8 42 0 6 12
9 4 4 8 26 8 0 0 43 0 4 8
10 8 4 8 27 0 4 8 44 0 0 0
II 0 4 8 28 0 4 8 45 8 4 8
12 0 2 4 29 8 4 8 46 0 8 16
13 8 4 8 30 0 2 4 47 0 0 0
14 0 0 0 31 0 8 16 48 0 4 8
15 0 4 8 32 4 0 0 49 4 4 8
16 4 8 16 33 0 0 0 50 12 0 0
Certain Important Lattices and Their Properties 107

Table 4.3. Numbers of points in the first shells of the cubic lattice Z3
with respect to various choices for the origin.

m N(O) N(J) N(2) N(3) m N(O) N(J) N(2) NO)

0 1 2 4 8 26 72 32 24 72
1 6 8 8 24 27 32 24 48 96
2 12 10 8 24 28 0 32 32 120
3 8 8 16 32 29 72 40 24 48
4 6 16 12 48 30 48 26 40 104
5 24 16 8 24 31 0 48 48 168
6 24 10 24 48 32 12 48 16 96
7 0 24 16 72 33 48 16 56 48
8 12 16 16 24 34 48 32 32 120
9 30 8 24 56 35 48 32 16 72
10 24 32 16 72 36 30 32 64 96
11 24 24 16 48 37 24 56 40 192
12 8 18 28 72 38 72 48 32 72
13 24 24 32 72 39 0 24 32 144
14 48 16 8 48 40 24 64 36 96
15 0 24 32 48 41 96 32 40 72
16 6 32 32 120 42 48 26 48 144
17 48 32 16 72 43 24 56 48 120
18 36 16 40 56 44 24 16 32 96
19 24 32 16 96 45 72 40 48 104
20 24 34 16 24 46 48 64 48 192
21 48 16 40 120 47 0 64 16 72
22 24 48 40 120 48 8 16 80 120
23 0 16 32 48 49 54 40 40 192
24 24 16 36 96 50 84 48 24 48
25 30 56 16 96 51 48 32 80 144

center of a lattice point, an edge, a square; or in other words the

coefficients of qm, qm + 114, q2m + 1/2 in the expansion of
(45)

respectively. In Table 4.3 the columns headed N(O), N(l), N(2), N(3) give
respectively the numbers of points of norms m, m + 1/4, m + 1/2,
2m + 3/4 in Z3 with the origin of coordinates at the center of a lattice
point, an edge, a square face, a cube; or in other words the coefficients of
qm, qm+I/4, qm+I/2, q2m+3/4 in the expansion of

(46)

respectively.
108 Chapter 4

If we expand

8 z .(z) = 83 (z)n = ~ rn(m)qm, (47)

m-O

then the coefficient rn(m) is the number of ways of writing m as a sum of

n squares. In particular for m > 0 we have
(48)

where 0 (m) is the difference between the numbers of divisors of m of the

two forms 4a + 1, 4a + 3,
8 ~ d, if m is odd,

r 4 (m) = 1 dim
24 ~ d ,
dim. d odd
if m is even,
(49)

r8(m) = 16 ~ (_1)m-d d 3. (50)

dim

Many similar, although more complicated, formulas are known for other
values of n. For further information see [Bat!], [Cas3], [Dic2], [Fri11,
[Gla11-[Gla6], [Gro2], [Har3]-[Har5], [MaI6], [Mor2], [Rad11, [Ran6],
[Smi5], [Vall]-[Va13].
If we define r4'(m) = r4(m)/8, then r4'(m) is multiplicative, i.e.
satisfies
r4'(lm) = r4'(l)r4'(m) whenever

land m are relatively prime. (51)

This property simplifies the calculation of r 4' (m), for now it is enough to
know the values when m = pQ is a power of a prime. These are
r4'(2 Q ) =3, (a~1),

r4'(p Q) = 1 + P + P 2 + ... + P Q, (a ~ 0),

where p is an odd prime. A similar multiplicative property holds for the

theta functions of certain other lattices. Bateman [Gro2, p. 1311 has
shown that r n (m)/2n is multiplicative if and only if n is 1, 2, 4 or 8.
(This reflects the fact that there are unique factorization theorems for
suitable rings of integers in the real, complex, quaternionic, and Cayley
numbers.)

6. The n-dimensional lattices An and A;

6.1 The lattice An. For n ~ 1,
An = {(XO,Xl •. ' •• X n) E zn+l :xo+'" +xn =O},
Certain Important Lattices and Their Properties 109

which uses n + 1 coordinates to define an n-dimensional lattice: An lies in

the hyperplane ~ Xi = 0 in Rn + I. The standard integral basis is marked
on the Coxeter-Dynkin diagram in Fig. 4.5. The extending node in this
basis is (1,0,0 ..... 0, -I), and the extended diagram is given in the final
column of Table 4.1. From Fig. 4.5 we can read off the generator matrix
-I 0 0 o0
0 -I 0 o0
M= 0 0 -I o0 (52)

0 0 0 0 -I

Two possible Gram matrices are

2 -I 0 0 0 2
-I 2 -I 0 0 2 1
0 -I 2 0 0 2
(53)
0 0 0 2 -I 1
0 0 0 -I 2 2

(The first of these corresponds to (52).) Also det = n + I, minimal

norm = 2, kissing number T = n (n + I), minimal vectors are all
permutations of (I, -1,0 ..... 0), Coxeter number h = n + I, packing
radius p = 1/J2, center density 0 = r n / 2 (n + I)-1I2, covering radius

1/2
R = 2a(n+l-a)
{ }
(54)
p n+1 '

where a is the integer part of (n + I) /2, a typical deep hole being the glue
vector [a I. The Voronoi cell is given in Chap. 21. Glue vectors:

[il = [_i_ .,. _i_ ~ ...

n+I' 'n+I'n+I'
~)
'n+I'
(55)

with j components equal to i / (n + 1), and i components equal to

- j / (n + 1), where i + j = n + 1 and 0 ~ i ~ n. The norm of [il is
ij / (n + I). Glue group: cyclic en +h with addition [j 1 + [k 1 = [j + k I.
Automorphism group: Go is the Weyl group of An' which is the symmetric
group S n + I of all permutations of the coordinates, and G I is the group of

(0.-1.1.0 •...• 0) (0.0.0 •...• -1.1)

o_----oo----o----~ - -------0
(-1.1.0 •..• 0) (0.0.-1.1.0 •..• 0)

Figure 4.5. Diagram for An showing integral basis for root lattice.
110 Chapter 4

order 2 generated by the negation of all coordinates (which interchanges

[iJ and [n + 1- i)), except that G I = I when n = 1. Theta series:
nj;1 8) [~I z)n
k-O n
An-I: (56)

An-I + [f] : (57)

m-- oo

where r= e 21ri / n .
6.2 The hexagonal lattice. We discuss the first few cases in more detail.
Of course A I ;;;; Z. A 2 is equivalent (or similar, see §1.4 of Chap. 1) to
the familiar hexagonal lattice shown in Fig. 1.3a of Chap. 1, so called
because the Voronoi cells are hexagons. The hexagonal lattice may be
spanned by the vectors (1,0) and (-1/2,.J3/2), and so an alternative
generator matrix to (52) is
o
M= -1 .J3 (58)
2 2

In this form det = 3/4, minimal norm = 1, T = 6, minimal vectors are

(± 1,0) and (± 1/2, ± .J3f2), h = 3, p = 1/2, density
~ = 11'/M = 0.9069 ... (15= I/M), R = 2p/.J3, typical deep holes being
the points (± 1/2, ± 1/2.J3) (see Fig. 1.3 d) , and thickness
o = 211' /3.J3 = 1.2092.... Glue vectors:
[0] = (0,0), [I] = [~, 2~ ), [2]= [-1 -I)
- -
2 ' 2.J3
.

Glue group = C). Automorphism group: go = 3!, gl = 2, g = 12. The

quadratic form associated with (58) is
(59)

and the theta series is therefore

X,Y-- DO x,y--oo

The terms in which y is even contribute (putting r = x - V2y, S = 1/2Y)

',5-- 00
Certain Important Lattices and Their Properties III

and those with y odd contribute (putting r = x - (y - 1) /2, s = (y - 1) /2)

q [r+t]\++tr = 82 (z)fh(3z).
r,s _-00

Therefore

(see (I 2». The series begins

0 hex (z) = 1 + 6q + 6q 3 + 6q4 + 12q7 + (61)

and the first 50 nonzero terms are given in Table 4.4. If the origin is
moved midway between two lattice points (j.e. to the middle of an edge,
when each point is joined by edges to its six neighbors), the theta series is

0 hex (edge) (z) = ~8 ~2[ ] 82 [ 3; ] = ~ ~1[ ]

= 2ql/4 + 2q 3/4 + 4q7/4 + 2q 9/4 + (62)

If the origin is moved to a deep hole the theta series is

0 hex +[!l(z) = 82(z )1f6(3z) + 03(Z )1f3(3Z)
= 3q 1/3 + 3q4/3 + 6q 7/3 + 6q 13/3 + (63)

Table 4.4. Theta series of planar hexagonal lattice.

m N(m) m N(m) m N(m) m N(m)

0 1 28 12 67 12 109 12
1 6 31 12 73 12 III 12
3 6 36 6 75 6 112 12
4 6 37 12 76 12 117 12
7 12 39 12 79 12 121 6
9 6 43 12 81 6 124 12
12 6 48 6 84 12 127 12
13 12 49 18 91 24 129 12
16 6 52 12 93 12 133 24
19 12 57 12 97 12 139 12
21 12 61 12 100 6 144 6
25 6 63 12 103 12 147 18
27 6 64 6 108 6 148 12
112 Chapter 4

(see [SloI91, [Te03]). If we write E>hex{Z) = ~ N{m)qm, then N{m) is

m-O
the number of times the quadratic form (59) represents m, and
N'{m) = N{m)/6 is multiplicative (see (51)). Thus N{m) is specified by
the values
for all a ~ 0,
for p == 1 (mod 3),
for p == 2 (mod 3), a odd,
for p == 2 (mod 3), a even,

where p ¢ 3 is a prime.
We shall use the name A2 only when the minimal norm is 2. From
(60) we have
(64)

6.3 The face-centered cubic lattice. Both A3 and D3 (§7) are equivalent to
the face-centered cubic lattice (or fcc), illustrated in every chemistry
textbook, and found in the pyramids of oranges on any fruit stand
(Fig. l.1). Fruiterers normally use square-based pyramids, but triangular-
based pyramids lead to geometrically equivalent packings. The simplest
definition is via D3: the fcc consists of the points (x,y, z), where x, y and
Z are integers with an even sum. A generator matrix is

-1 -1 0
M = 1 -1 0 (65)
o 1-1

det = 4, minimal norm = 2, T = 12, and the minimal vectors consist of all
permutations of (± 1, ± 1, 0), the twelve vertices of a regular
cuboctahedron. Also h = 4, the packing radius p = I/.Ji, the density
A = 'Tr/m = 0.7405 ... (o=T5/2) and the covering radius R = p.Ji = l.
The Voronoi cell is a rhombic dodecahedron (Fig. 2.2a and Chap. 21).
There are two kinds of holes in the fcc, corresponding to the two types of
vertices of the Voronoi cell: deep or octahedral holes, such as (0,0, 1),
surrounded by six lattice points, and shallow or tetrahedral holes, such as
(V2, V2, V2), surrounded by four lattice points. Glue vectors:
[0] = (0,0,0), [1] = (V2,V2,V2), [2] = (0,0,1), [3] = (V2,i/2,_i/2).

Glue group: C 4. Automorphism group: go = 24, gl = 2, g = 48. The fcc

is the unique lattice with this density, although there are equally dense
nonlattice packings (see §6.5). Theta series:
E> fcc (z ) = V2 (03{z)3 +0 4 (z )3) = 03(4z)3 + 30 3(4z )02(4z)2
= 1 + 12q2 + 6q4 + 24 q 6 + .... (66)
Certain Important Lattices and Their Properties 113

There is a complicated explicit formula for the coefficients [Dic2, Vol. II,
p. 263], [Gro21. The first 50 coefficients are given in Table 4.5. For the
translates we have:
0 fcc +[II(Z) = 1/2 (Jz(z)3 = 4q 3/4 + 12qll/4 + 12q19/4 + (67)
° fcc + [21 (z) = V2 (0 3(z )3 - 0 4(z ) 3)
= 6q + 8q 3 + 24q 5 + 30q9 + (68)

Table 4.5. Theta series of face-centered cubic lattice. The table gives
N(m) for m = lOr + s.
10r\s 0 2 4 6 8 10 12 14 16 18
0+ 1 12 6 24 12 24 8 48 6 36
20+ 24 24 24 72 0 48 12 48 30 72
40+ 24 48 24 48 8 84 24 96 48 24
60+ 0 96 6 96 48 48 36 120 24 48
80+ 24 48 48 120 24 120 0 96 24 108
100+ 30 48 72 72 32 144 0 96 72 72

6.4 The tetrahedral or diamond packing. The tetrahedral packing, as

found in diamond (the hardest substance known), consists of the points
fcc u [11 + fcc. (69)

There is a picture in [Kit4, p. 25], for example. This is the packing Dt

(§7.3) and is not a lattice. All the points are equivalent, and the Voronoi
region around anyone of them has volume 1, the minimal norm = 3/4,
kissing number T = 4, and the minimal vectors = (V2,V2,V2), (V2,-V2,_lh),
(-V2,V2,-V2), (-V2,-V2,VZ). Thus the four neighbors of each point form a
regular tetrahedron. The packing radius p = ..J3/4, density
a = 7r..J3/16 = 0.3401... (Chap. 1, Eq. (14»,0 = r 6 33/2, covering radius
R = 2p = ..J3/2, and the holes are tetrahedral holes such as (Vz, 1/2 , - V2).
Theta series ([SI017]):
0diamond(Z) = V2 (02(Z)3+03(Z)3+0 4(Z)3)
= 1 + 4 q 3/4 + 12q2 + 12qll/4 + 6q4 + (70)

If the origin is moved to a hole the theta series is

odiamond (hole) (z) = V2 (02(z)3 + 0 3(Z)3 -0 4(z )3)
= 4 q 3/4 + 6q + 12qll/4 + 8q 3 +

6.5 The hexagonal close-packing. As remarked in Chap. 1, A], A 2 and A 3

are the densest known packings in dimensions 1, 2 and 3 (and the densest
possible lattice packings in those dimensions). In three dimensions
114 Chapter 4

however there are infinitely many nonlattice packings with the same
density and kissing number as the fcc. The simplest of these is the
hexagonal close-packing or hcp (§1.3 of Chap. 0, which is of considerable
importance in chemistry. The hcp is not itself a lattice, but may be defined
as the union of the lattice L spanned by the vectors (I, 0, 0), (1/2, ../3/2,0)
and (0,0, .JS/3), and the translate L + (I/2, l/m,.J2/3). Alternatively it
is the union of the lattice spanned by the vectors
r l/2(1,1,0), rl/2(1,0,0, r l/2(-4/3,-4/3,-4/3)

and its translate by r l/2(0, 1, O. A third definition is that the hcp consists
of the points
± (6i+l, 6j+l, 6k-2, 2../3(4£-1)), (72)

where i, j, k, £ are integers with i + j + k = 0 [SI019]. With the first

definition the Voronoi cells have volume 1/.J2, minimal norm = 1, T = 12,
minimal vectors = (±1,0,0), (±1/2,±../3/2,0), ±(0,-1/../3,.J2/3) and
± (± 1/2, l/m,.J2/3), p = 1/2, ~ = 7r/3.J2 = 0.7405 ... , and
R = p.J2 = 1/.J2. There are two kinds of holes, deep or octahedral holes
such as (0,1/../3,1/../6) and shallow or tetrahedral holes such as
(1/2, l/m, 1/J24). Theta series:

0 hcp (z) = ¢O(Z){8 3 [8; )- ~ 82 [83Z )} + i¢o [~ ) 82 [ 8; )

= 1 + 12q + 6q2 + 2q 8/3 + ... (73)

(Table 4.6). If the origin is moved to a deep hole the theta series is
0 hcp (hole) (z) = 8 2 (2z /3){8 2 (z) 1/;6 (3z) + 83(z )1/;3 (3z)}
= 6ql/2 + 6q 3/2 + 6qll/6 + 12 q 5/2 + ... . (74)

For further information about three-dimensional lattices see Chap. 1,

[Balll, [Cox18], [Cox20], [Fej8a]-[Fej 10], [IntI], [Kit4], [Sma 11,
[WeI2]-[WeJ41, [Wycl], [Wyc2].

Table 4.6. Theta series of 3-dimensional hexagonal close-packing. The

table gives N(m) for m = (lOr+s)/3.

10r\s 0 1 2 3 4 5 6 7 8 9
0+ 1 0 0 12 0 0 6 0 2 18
10+ 0 12 6 0 0 12 0 12 6 6
20+ 12 24 6 0 0 12 0 12 0 24
30+ 12 12 2 12 6 24 6 12 0 24
40+ 0 12 0 6 24 12 12 24 6 12
50+ 0 24 0 24 18 12 12 24 0 12
Certain Important Lattices and Their Properties 115

6.6 The dual lattice A;. The lattice dual to An is

n
A;= U ([il+An) , (75)
i-O

with generator matrix

-I 0 0 0
0 -1 0 0
M= (76)
0 0 -I 0
-n
n+1 n +1 n +1 n +1 n +1

An equivalent definition uses either the Gram matrix

n -1 -I -I
-I n -1 -1
A=
-I -I -I n

(cf. Eq. (49) of Chap. 8), with determinant (n+l)n-l, or the associated
quadratic form
n n
n ~ xl - ~ Xi Xj' (78)
i-I i""j

often referred to as Voronoi's principal form of the first type [RysI21,

[RysI31. Using the definition (76), det = 1/(n+l), minimal norm
= n / (n + I), T = 2 (n = I) or 2n + 2 (n ~ 2),
1 ~ nnl2
p = T"---;;+1' {) = 2n(n+l)(n-Il/2' (79)

R = p"!FIr -J
~-3- =
n (n +1)
12(n+1) ,
(
80
)

a typical deep hole being

1
- - ( - n , -n +2, -n +4, ... , n -2, n), (81)
2n+2

the thickness is
n/2
e= II: .In+T [ n (n+ 2) ]
(82)
n n 12(n + 1) ,

and the Voronoi cells are permutohedra (see Chap. 21), with vertices all
permutations of (81). A;;;;; AI;;;; Z, A; ;;;; A 2 , and A; is the body-
centered cubic lattice (see §6.7). A;
is the best covering of Rn for n = 2,
116 Chapter 4

the best lattice covering for n ~ 5, and the best covering known for
n ~ 23 [Bamll-[Bam31, [De141, [Fewll, [Gamll, [Gam21, [Rys71,
[RysI21, [RysI31.

The typical lattice between An and An· is An[s 1, where s is any divisor
of n + 1. In Coxeter's notation [Coxl01 this is A~, where rs = n + 1. Its
theta series can be found by summing (57) for
e = 0, s, 2s • ...• n + 1 - s.
6.7 The body-centered cubic lattice. Both Ai and ni (§7) are equivalent
to the body-centered cubic lattice (or bcc), also familiar from chemistry.
The simplest definition is via n;: the bcc consists of the points (x,y, z)
where x, y and z are all even or all odd integers. A generator matrix is
2 0 0
M = 020 (83)
1 1

det = 16, minimal norm = 3, T = 8, minimal vectors are (± 1, ± 1, ± 0,

p = ../3/2, density A = 1r../3/8 = 0.6802... (0=3../3/32), and
R = pM = .J5/2. There is only one type of hole in the bcc, namely
tetrahedral holes such as (0,1/2,0, surrounded by four lattice points. Also
e= 1r5.J5/32 = 1.4635 ... ,

and the Voronoi cells are truncated octahedra (Fig.2.2b, Chap. 20.
Theta series:

(see Table 4.7). If the origin is moved to a hole the theta series is

Table 4.7. Theta series of body·centered cubic lattice.

m N(m) m N(m) m N(m) m N(m)
0 1 36 30 75 56 115 48
3 8 40 24 76 24 116 72
4 6 43 24 80 24 120 48
8 12 44 24 83 72 123 48
11 24 48 8 84 48 128 12
12 8 51 48 88 24 131 120
16 6 52 24 91 48 132 48
19 24 56 48 96 24 136 48
20 24 59 72 99 72 139 72
24 24 64 6 100 30 140 48
27 32 67 24 104 72 144 30
32 12 68 48 107 72 147 56
35 48 72 36 108 32 148 24
Certain Important Lattices and Their Properties 117

0 bcc (holel (z) = V2 82(z )8 2(2z)2

= 4 q 5/4 + 4q13/4 + 8q21/4 + 12q29/4 + (85)

For further information see the references at the end of §6.5.

7. The n-dimensional lattices Dn and D:

7.1 The lattice Dn. For n ~ 3,
Dn = {(x" ... ,xn) E zn : Xl + ... +xn even},

or in other words Dn is obtained by coloring the points of zn alternately

red and white with a checkerboard coloring, and taking the red points. Dn
is sometimes called the checkerboard lattice. The standard integral basis
is marked on the Coxeter-Dynkin diagram in Fig. 4.6. The extending node
in this basis is (0, ... ,0,1,1), and the extended diagram is given in the
final column of Table 4.1. From Fig. 4.6 we obtain the generator matrix
-1 -1 0 o 0
1 -1 0 o 0
M = 0 1-1 o 0 (86)

o 0 0 -1

Then det = 4, minimal norm = 2, kissing number T = 2n (n - 1), minimal

vectors = all permutations of (± 1, ± 1,0, ... ,0), h = 2n + 2, packing
radius p = 1/.J2, center density 0 = 2-(n+2)/2, and covering radius
R = p.J2 (n =3) or p.Jnfi (n ~ 4). There are two kinds of hole, deep
holes such as the glue vectors [2] Of n ~ 4) or [11 (if n ~ 4), and shallow
holes such as the glue vectors [11 Of n ~ 4) or [2] Of n ~ 4). For n = 4
there is only one kind of hole. The Voronoi cell is given in Chap. 21, and
the Delaunay cells in §4.2 of Chap. 5. Glue vectors:
[0] = (0,0, ... ,0), norm 0,
[11= (V2,V2, ... ,V2), normn/4,
[2] = (0,0, ... ,1), norm 1,
[3] = (V2,V2, ... , -V2), norm n/4.

(-1,-1,0 •... ,0)

(0,1,-1,0, .. ,0) (0, ... ,1,-1,0)

p..---o------<:r-- - --o------<J
(0.o,1,-1, ...• 0) (0, ...• 0,1,-11

(1,-1,0, ... ,0)

Figure 4.6. Diagram for D. showing integral basis for root lattice.
118 Chapter 4

Glue group: Klein V 4 ([i] + [;] = 0) if n IS even, cyclic C4

([I] + [2] = [3]) if n is odd. Automorphism group:
n = 4: go = 23 . 4!, gl = 3! (all perms of [1], [2], [3]),
n ;z! 4: go = 2n - 1 . n!, gl = 2 (interchange [I] and [3]).

Go is generated by all permutations together with sign changes of evenly

many coordinates, and G 1 contains the sign change of the last coordinate
and, for n = 4 only, the Hadamard matrix

-1 1-1
1 -1 -1
-1 -1

Theta series (see Chap. 2, Eq. (37»:

Dn : Vd0 3(z)n +04(Z)n) = ~ 'n (2rn)q2m, (87)

m-O

using the notation of (47),

Dn + [I] or Dn + [3] : 'ho 2 (z)n, (88)

(89)

D3 == A3 is the face-centered cubic lattice (§6.3). D 3, D4 and Ds are the

densest possible lattice packings in dimensions 3, 4 and 5, and densest
known packings in these dimensions, although for n = 3 and 5 there are
equally dense nonlattice packings (§6.5 and Chap. 5).
7.2 The four-dimensional lattice D 4 • D 4 , which was also described in §IA
of Chap. 1, is one of the two most useful four-dimensional lattices (the
other being A;). D4 is defined by (86) or equivalently by the generator
matrix

1 0 0 0
0 0 0
M'= (90)
0 0 0
V2 V2 'h 'h

The matrix
0 0
1 1 -1 0 0
T = 0'2 0
(91)

0 0 -1
Certain Important Lattices and Their Properties 119

maps the first version into the second. «90) also defines the dual lattice
D: - see §7.4 - showing that D 4 ;;;; D:.) Using the definition (86),
det = 4, minimal norm = 2, T = 24, minimal vectors = all permutations of
(± 1, ± 1,0,0), h = 10, p = 1/J2, d = 7r2/16 = 0.6169... (0 = 1/8),
R = pJ2 = 1, typical deep holes = (± '12, ± '12, ± '12, ± 'h) and
(0,0,0, ± 1), 0 = 7r2/4 = 2.4674... . The Voronoi cell is the regular 4-
dimensional polytope known as the 24-cell or {3,4,3} (Chap. 21, [Cox20,
§8.2]). The three nonzero cosets D4 + [i], i = 1,2,3, are equivalent.
Theta series

(cf. Eq. (89) of Chap. 7). D4 is the unique lattice with this density.
Using the second definition, (90), D 4 consists of the vectors (a, b, c, d)
with a, b, c, d all in Z or all in Z + V2, and therefore may be regarded as
the lattice of Hurwitz integral quaternions ([Cox8], [CoxI8, p. 25], [Har5,
§20.6], [Hurl]). This gives the lattice points the structure of a skew field.
As an Eisenstein lattice, D4 is generated by (2,0) and (1,0).
The first 50 coefficients of the theta series for D 4 (using the first
definition, Eq. (86» are shown in Table 4.8. These coefficients are given
explicitly by N(2m) = r4(2m), using the second formula in Eq. (49).
Also N(2m) is the number of integral quaternions of norm m, and
(24)-IN(2m) is a multiplicative function of m. In the notation of Schliifii
and Coxeter, D4 is the regular honeycomb {3,3,4,3} (see [Cox20, §7.8]).

Table 4.8. Theta series of 4-dimensional lattice D 4 • The table gives

N(m) for m - lOr + s.
r\s 0 2 4 6 8 10 12 14 16 18
0 1 24 24 96 24 144 96 192 24 312
2 144 288 96 336 192 576 24 432 312 480
4 144 768 288 576 96 744 336 960 192 720
6 576 768 24 1152 432 1152 312 912 480 1344
8 144 1008 768 1056 288 1872 576 1152 96 1368
10 744 1728 336 1296 960 1728 192 1920 720 1440

7.3 The packing D;i. The covering radius of Dn increases with n, and
when n = 8 it is equal to the minimal distance between the lattice points.
So when n ~ 8 we can slide another copy of Dn in between the points of
D n , doubling the number of points (and the density) without reducing the
distance between them! Formally, we define
Dn+ = Dn U ([I]+D n). (93)

D;i is a lattice packing if and only if n is even. Dj is the tetrahedral or

diamond packing (§6.4) and Dt ;;;; Z4. When n = 8 this construction is
especially important, the lattice Dt being known as £8 (§8.1). The
120 Chapter 4

Voronoi cells of D: have volume 1, the minimal norm = n/4 (n ~ 8) or

2 (n ~ 8), center density 0 = (.Jri/4)n (n ~ 8) or r
n/2 (n ~ 8), kissing
number T = 2n- 1 (n ~ 7), 240 (n =8), or 2n (n - 1) (n ~ 9), and the theta
series is
(94)

7.4 The dual lattice D;

D; = Dn U ([1]+Dn ) U ([2]+D n) U ([3]+Dn ).

Generator matrix
1 0 o 0
o o 0
M= (95)

det-l/4, minimal norm =3/4 (n=3) or l(n~4), T=8(n-3),

24 (n =4) or 2n (n ~ 5), p = ../3/4, 0 = 31.5 r5 (n =3) or 2-(n-l) (n ~ 4),
R=pn l /2/.J2 (n even), p.J573 (n=3) or p(2n-I)1I2/2 (nodd~5),
while the Voronoi cells are given in Chap. 21. Theta series:
(96)

D; is the body-centered cubic lattice (§6.7), and D: == D 4 .

8. The lattices E 6, E 7 and E 8

8.1 The 8-dimensional lattice E g. The existence of an even unimodular 8-
dimensional form was first established nonconstructively in [Smi6, p. 5211,
and an explicit construction was given in [Kor11-[Kor3] and [MinO].
Gosset in 1900 [Gosl1 seems to have been the first to study the lattice Eg
itself. Eg = Dt is a special case of the family of packings constructed in
§7.3, and so might be called the 8-dimensional diamond lattice. In the
even coordinate system E g consists of the points
{(XI, . • . , Xg) : all Xi E Z or all Xi EZ + V2,
~ Xi =0 (mod 2)} . (97)

The odd coordinate system is obtained by changing the sign of any

coordinate: the points are
{(XI, . • . ,Xg) : all Xi E Z or all Xi EZ + V2,
2; Xi = 2xg (mod 2)}. (98)

The standard integral basis (in the odd coordinate system) is marked on
the Coxeter-Dynkin diagram in Fig. 4.7. The extending node is (V2,-V2 7),
Certain Important Lattices and Their Properties 121

I 5 I 3
(2 '-2)
ES

Figure 4.7. Diagram for £8 showing integral basis for root lattice in the
odd coordinate system. A basis in the even coordinate system can be
obtained by changing the signs of any odd number of coordinates (for
example the last three coordinates).

and the extended diagram is given in the final column of Table 4.1. A
generator matrix (in the even coordinate system) is
2 0 0 0 0 0 0 0
-1 0 0 0 0 0 0
0 -1 0 0 0 0 0
0 0 -I 0 0 0 0
M= (99)
0 0 0 -1 0 0 o'
0 0 0 0 -1 0 0
0 0 0 0 0 -1 0
V2 'h 'h V2 1/2 1/2 V2 1/2

det = 1, minimal norm = 2, kissing number T = 240, and the minimal

vectors are all vectors (± 12,06), together with those vectors (± 1/2g) that
have an even number of minus signs in the even coordinate system, or an
odd number of minus signs in the odd coordinate system. The Coxeter
number h = 30, packing radius p = 1/J2, density A = 11'4/384 - 0.2537 .. .
(/)= 1/16), covering radius = pJ2 = 1, thickness e = 11'4/24 - 4.0587 ... ,
and the Voronoi cell is the reciprocal of the polytope 421 (see Chap. 21 and
[Cox20, §11.8]). There are two kinds of holes in E g, deep holes such as
(0 7,1) (halves of norm 4 vectors), surrounded by 16 points, and shallow
holes such as ((1/6)7,5/6), surrounded by 8 lattice points (see Fig. 21.8).
E; = E g, so there is no glue (or more precisely the only glue is
[0] = (08».

Eg is the unique lattice with this density and minimal norm [Vet2].
Automorphism group: go = 214 35 52 7 = 696729600, gl = 1. Go is the
Weyl group W(E g), generated On the even coordinate system) by all
permutations of 8 letters, all even sign changes, and the matrix
diag {H 4, H 4} (100)

(cf. §7.t). See also [Cox20], [Cox28], [Edg3].

E g may also be obtained by applying Construction A to the Hamming
code ,tfs (§2.5 of Chap. 5, Example 5 of Chap. 7). The minimal vectors
(after rescaling) now consist of 24 . 14 = 224 vectors «±
1/2 )4, 04), where
the nonzero coordinates support a minimal weight codeword in ,tfs,
122 Chapter 4

together with 16 vectors (± 1,0 7). This definition is the best for showing
the identification of E 8 with the integral Cayley numbers or octaves
([Bli5], [Bli6], [ConI6, p. 85], [Cox91, [CoxI7], [CoxI8, Chap. 2]). Let
the coordinates be labeled 00, 0, I, 2 ..... 6 as in §2.4.2 of Chap. 3. The
real Cayley algebra is spanned by "numbers" i= = I, i o, i, ... " i6 with
the property that for the quadruples {a, b, c, d} = {oo, 1,2, 4} , ... ,
{00,0, I, 3} given in Eq. (56) of Chap. 3 (corresponding to the minimal
vectors of ,tfs that contain 00) the sets lia,ib,icoid} ;;;; {J,i,j,k} span a
quaternion subalgebra. Thus i,i 2 = i 4, i 2i, = -i 4, i,i 3 = i o, etc. The
integral Cayley numbers are the members of the E8 sublattice spanned by
vectors of the form (± V24, 04 ) for which the distinguished sets of four
coordinates are those obtained from Eq. (56) of Chap. 3 by interchanging
00 and 0, together with their complements. The 240 minimal vectors
become the 240 unit (or norm 1) Cayley numbers.
Alternatively, E 8 can also be identified with the ring of icosians, as
described is §2 of Chap. 8. Table 8.1 gives icosian names for the 240
minimal vectors. A quaternionic construction for E 8 is given in Eq. (77) of
Chap. 2. Theta series:
0 E8 (Z) = 1/2 (02(z)8 + 03(z)8 + 04(z)8)
= 02(2z)8 + 14 02(2z)40 3(2z)4 + 03(2z)8
(101)
m-O

(102)

where
O',(m) = ~ d' (103)
dim

(Table 4.9). N m is the number of integral Cayley numbers of norm mi2,

and 240- 1 N m / 2 is multiplicative. If the origin is moved to a deep hole the
theta series is

=
~ N m' qm= 16q + 128q2 + 448 q 3 + 1024 q 4 + (104)
m-I

(105)

with the convention that 0'3 (x) = 0 if x is not an integer.

Table 4.10 gives coordinates for representative vectors in the first eight
layers of E 8. The full list of vectors is obtained by applying arbitrary
permutations and signs to the vectors in the table, except that if the vector
is prefixed by an E (resp. D) then an even (resp. odd) number of minus
Certain Important Lattices and Their Properties 123

Table 4.9. Theta series of Gosset's lattices E 6, E7 and Eg.

m N m (E 6) N m (E 7) Nm(Eg) m N m (E 6) N m (E 7) Nm(Es)
0 1 1 1 52 45900 470232 4747680
2 72 126 240 54 59040 505568 4905600
4 270 756 2160 56 46800 532800 6026880
6 720 2072 6720 58 75600 615384 5853600
8 936 4158 17520 60 51840 640080 7620480
10 2160 7560 30240 62 69264 701568 7150080
12 2214 11592 60480 64 73710 799092 8987760
14 3600 16704 82560 66 88560 809424 8951040
16 4590 24948 140400 68 62208 853776 10614240
18 6552 31878 181680 70 108000 1006992 10402560
20 5184 39816 272160 72 85176 1051974 13262640
22 10800 55944 319680 74 98640 1031688 12156960
24 9360 66584 490560 76 97740 1195992 14817600
26 12240 76104 527520 78 122400 1286208 14770560
28 13500 99792 743040 80 88128 1313928 17690400
30 17712 116928 846720 82 151200 1469664 16541280
32 14760 133182 1123440 84 110700 1474704 20805120
34 25920 160272 1179360 86 133200 1547784 19081920
36 19710 177660 1635120 88 140400 1797768 23336640
38 26064 205128 1646400 90 157680 1776600 22891680
40 28080 249480 2207520 92 114048 1809360 26282880
42 36000 265104 2311680 94 198720 2104704 24917760
44 25920 281736 2877120 96 147600 2130968 31456320
46 47520 350784 2920320 98 176472 2123982 28318320
48 37638 382536 3931200 100 162270 2382156 34022160
50 43272 390726 3780240

signs are required (in the even coordinate system). The group Go acts
transitively on the sets of vectors of norms 2, 4, 6, 10 and 12. (There are
two orbits of norm 8 vectors: such a vector mayor may not be twice a
lattice vector. The vectors of the second orbit, divided by 3, give the
shallow holes near 0.)
For further information about E 6 , E7 and E8 see later chapters and
[Boul], [CoxIO), [CoxI8), [Cox20), [Cox28), [Hazl1.

Table 4.10. The first 8 shells of Es.

Norm Number Vectors
0 1 Os.
2 240 1206, E (I/2)s.
4 2160 207,1 4 04 , D(3/2)(1/2)7.
6 6720 21 20 5,1 602, E(3/2)2(1/2)6.
8 17520 2206, 21 403, IS, D (3/2)3(\/2)5, E (5/2)(1/2)7.
10 30240 3106, 221 204, 21 60, D (5/2)(3/2)(\/2)6, E (3/2)4(\/2)4.
12 60480 31 304, 2305, 221 402, E (5/2)(3/2)2(\/2)5, D (3/2)5(\/2)3.
14 82560 32105,31502,231203, 2216, D(7/2)(1/2)7,
E (5/2)2(1/2)6, D (5/2) (3/2)3(\/2)4, E (3/2)6 (\/2)2
16 140400 40 7 ,321 303, 31 7, 2404, 23140, EO/2) (3/2) (\/2)6,
0(5/2)2(3/2) (\/2)5, E(5/2) (3/2)4(1/2)3, D(3/2)7(1/2).
124 Chapter 4

8.2 The 7 -dimensional lattices E 7 and E; . The vectors 10 Es

perpendicular to any minimal vector vEE s form the lattice E 7:
E 7 = {x E E s : x . v = O}. (106)

There are several possible coordinate systems. Using the even coordinate
system for E s and taking v = (1/2S) we obtain
E7= {(XI ....• Xs) E ES:XI + ... +X8=0}; (107)

using the odd coordinate system and taking v = (l/z 7 , - Yz) we obtain
E7 = {(Xl ..... Xs) E Es: ~Xi = 2xs}; (108)

and Gn either coordinate system) v = (0 6 , 1,-I) leads to

E 7 = {(x I. . . . • X 8) E E 8 : X7 = X 8} . (109)

A standard integral basis (using definition (107» IS marked on the

Coxeter-Dynkin diagram in Fig. 4.8. The extending node is (0 6 , -1, I). A
generator matrix for E 7 as defined by (107) is
-1 0 0 0 0 0 0
0 -1 0 0 0 0 0
0 0 -1 0 0 0 0
M 0 0 0 -1 0 0 0 . (I 10)
0 0 0 0 -1 0 0
0 0 0 0 0 -1 0
I/Z Y2 Yz Yz -Yz -Yz -Yz -Yz

Still another definition (see §2.5 of Chap. 5) leads to the generator matrix

20 00 00 0
02 00 00 0
00 20 00 0
M' = 1.. 00 02 00 0 (III)
2
11 10 10 0
o1 11 o1 0
00 11 10

1-1.1,0 6 )

10,-1,1,0 5 ) 10',_1.1,0') (05,-1,1,0)

Figure 4.8. Diagram for E 7 showing integral basis for root lattice in the
coordinate system (107).
Certain Important Lattices and Their Properties 125

For the definition via (107), (110) we have det = 2, minimal norm = 2,
T = 126, the minimal vectors consist of 56 of the form (1, -1,0 6) and 70 of
the form (V24,_V24), h = 18, p = 1/..fi, ~ = 1r 3/105 = 0.2953 ... (15=1/16),
R = p.J3 = .J31i, a typical deep hole being the glue vector [1], while the
Voronoi cell is given in Chap. 21. Glue vectors:
[0] = (0,0,0,0,0,0,0,0), of norm 0,
of norm 3/2 •

Glue group = C 2 . Automorphism group: Go is the Weyl group W(E 7 ), of

order go = 2 10 . 34 . 5· 7 = 2903040, G 1 = 1. E7 is the unique lattice
with this density [Vet21, although there are equally dense nonlattices (§4.2
of Chap. 5). Theta series (Table 4.9):
0 E /z) = IJ 3(2z)7 + 7IJ 3(2z)3IJ 2(2z)4
= 1 + 126q2 + 756q4 + 2076 q 6 + (112)
0 E7 +[I!(Z) = IJ 2(2z)1 + 7IJ 2(2z)3IJ 3(2z)4
= 56 q 3/2 + 576q7/2 + 1512qll/2 + 4032 q 15/2
+ 5544 q 19/2 + 12096 q 23/2 + 13664 q 27/2 +

Other constructions will be given in Chap. 5.

The dual lattice is
(114)

with generator matrix

-1 0 0 0 0 0 0
0 -1 0 0 0 0 0
0 0 -1 1 0 0 0 0
M 0 0 0 -1 1 0 0 0 , (115)
0 0 0 0 -1 1 0 0
0 0 0 0 0 -1 0
_3/ 4 _3/ 4 1/4 1/4 1/4 1/4 1/4 1/4

det = V2, minimal norm = 3/2, T = 56, minimal vectors = ± (1/46 ,_3/42),
P =.J378, R = pM =.J778, a typical deep hole being (7/8,(-1/8)7),
whose images are the vertices of the Voronoi cell (see [Wor2D. The theta
series is the sum of (112) and (113).

8.3 The 6-dimensional lattices E 6 and E;. The vectors in Eg

perpendicular to any Arsublattice V in Eg form the lattice E6:
E6 = {x EE g : x·v=O for all v E V}. (116)

Again there are several possible coordinate systems. Using the even
coordinate system for Eg and taking V = < (1,0 6,1), (V2 g) > we obtain
E6 = {(x I • . . . • x g) E E g : x I + X g = X 2-1- .•. + X 6 = O}; (117)
126 Chapter 4

(0,-1,1,0 5 ) (03,_1,1,0 3 ) (0 5 ,-1,1,0)

Figure 4.9. Diagram for E6 showing integral basis for root lattice in the
coordinate system (I 17).

while (in either coordinate system) V = < (0 5,1,-1,0), (06 ,1,-I) >
leads to

A standard integral basis (using definition (117)) IS marked on the

Coxeter-Dynkin diagram in Fig. 4.9. The extending node is (- I; 06 ; ]).
From Fig. 4.9 we obtain the generator matrix

0 -I 0 0 0 0 0
0 0 -I 0 0 0 0
M 0 0 0 -I 1 0 0 0 (119)
0 0 0 0 -I 0 0
0 0 0 0 0 -1 0
1/2 1/2 1/2 V2 -V2 -V2 _1/2 _1/2

E 6 also has a simple construction as a three-dimensional complex

lattice, a lattice over the Eisenstein integers (§2.6 of Chap. 2). A
generator matrix for this version of E6 is

~ ~ ~l, (120)
1 I

where (J = w - W = R.
Using the definitions (117) and (119), we have det = 3, minimal norm
= 2, T = 72, the minimal vectors consist of 30 of the form (0; 1, -1,0 4 ; 0),
40 of the form ± (Vd'2 3 - V2 3. - V2) and 2 of the form ± (1. 06 • -1) h = 12
p = 1/.Ji, ~ = 7r 3/48J3 = 0~3729... (0= 1/8·)3), R = p'J8r3, ~ typicai
deep hole being the glue vector [11, and the Voronoi cell is given in
Chap. 21. Glue vectors:
[01 = (0;0,0,0,0,0,0;0), of norm 0,
[11 = (0; - 2IJ, - 2IJ, 1/3, IIJ, 1/3, IIJ; 0) , of norm 4/ 3 ,
[21 = -[I!, of norm 4/ 3 •

Glue group = C 3 . Automorphism group: Go is the Weyl group W(E 6 ), of

order go = 27 . 34 . 5 = 51840, G j = C 2 (generated by negation). (See
[Coxl!, [Cox?], [Coxl3].) E6 is the unique lattice with this density
Certain Important Lattices and Their Properties 127

[Bar?], [Vet2], although there are equally dense nonlattices (§4.2 of

Chap. 5). Theta series (Table 4.9):
0 E6(Z) = c!>O(z)3 + 1/4 {c!>O(Z /3) - c!>o(z))3 ,
= 1 + 72q2 + 270q4 + 720q6 + ... , (121)

0 E6 (Z) = 1/3[c!>O(Z /3)3 + 1/4{3c!>o(z) - c!>o(z /3))3]

= 1 + 54q4/3 + 72q2 + 432 q 10/3 + 270q4

+ 918 q 16/3 + 720q6 + 2160q22/3 + ... , (122)

0 E6 +III(Z) = 0 E6+12J(Z) = lf2 (0 E6 (Z) -0Eiz)}

= 27 q 4/3 + 216 q lo/3 + 459q16/3 + ... . (123)

(Eq. (121) follows from definition (120), as shown in Chap. 7, and (122) is
derived from (121) via (18).)
The dual lattice is
E~ = E6 U ([1I+E 6) U ([2]+E 6) , (124)

with generator matrix

0 -1 0 0 0 0 0
0 0 -1 0 0 0 0
M 0 0 0 -1 0 0 0 . (125)
0 0 0 0 -1 0 0
0 2/3 2/3 -If3 -1/3 -1/3 -1/3 0
V2 lf2 lf2 lf2 -lf2 -V2 -V2 -lf2

As a three-dimensional complex lattice E~ has generator matrix

8 0 0
M' = -1 0 (126)
o -1

Using the definition (125), det - 1/3, minimal norm = 4/3, 'T = 54, the
minimal vectors consist of 30 of the form ± (0; (-2/3)2, (1/3)4;0) and 24
of the form ± (±1/2;(-1/6)5,5/6;+1/2), p = 1/.J3, and the covering
radius R = p..fi = .J273, corresponding to the deep hole
(0; 1, 1,1, - 1, -1, - 1; 0) /3, whose images are the vertices of the Voronoi
cell (see [Worl]). Theta series: Eq. (122).

9. The 12-dimensional Coxeter-Todd lattice K 12

So far all lattices mentioned were known in the nineteenth century, but
now we move into the twentieth century. The automorphism group of KI2
was discovered by Mitchell in 1914 [Mit3], while the lattice itself was first
explicitly described by Coxeter and Todd in 1954 [Cox29]. Like E 6 , KI2
128 Chapter 4

has a simple description as a complex lattice. K 12 is the real form of the

6-dimensional complex lattice over the Eisenstein integers that is generated
by the vectors

(127)

where 0 = w - W = R may be in any position and there are an even

number of minus signs. An alternative definition (Example lOa of
Chap. 7) uses the hexacode (§2.5.2 of Chap. 3, in the version defined by
(64» and corresponds to the generator matrix
2 0 0 0 0 0
0 2 0 0 0 0
0 0 2 0 0 0
M= w w I 0 0
(128)
w w 0 I 0
w w 0 0

(See also Example 11 c of Chap. 7.) In the definition given by (127), K I2

may be regarded as the set of all Leech lattice vectors of the form

(I 29)

(see Chap. 6). (I 29) corresponds to the K 12 vector

(UI + OVI • ...• U6 + OV6)/J8.
For example the Leech vector with VI = u2 = = U6 = 2,
UI = V2 = ... = V6 = 0 corresponds to (0,1 5) /.J2. K]2 may also be
obtained as the set of Leech lattice vectors of the form

o o o
(130)

where Xj + Yj + Zj = 0, i = 1 ..... 6 (Fig. 6.3). Using the definition

(127), det = 729, minimal norm = 4, kissing number T = 756, the minimal
vectors consist of 576 of the form w"( ± 0, 15) /.J2, v = 0, 1, 2, with an even
number of minus signs, and 180 of the form w"(±22,04)/.J2, packing
radius p = 1, density ~ = 11'6/19440 = 0.04945... (0= 1/27), and covering
radius R = p J873, a typical deep hole being (4/0,0 5)/.J2. The Voronoi
cell has 4788 faces, 756 corresponding to the minimal vectors and 4032 to
those of the next layer [Con38, Th. 31. As a complex lattice over the
Eisenstein integers K 12 is an integral unimodular lattice, and so (as a real
lattice) K;2 ;;;; K!2' The automorphism group of the real lattice has order
Certain Important Lattices and Their Properties 129

Table 4.11. Theta series of 12-dimensional Coxeter-Todd lattice K 12.

m N(m) m N(m)
0 1 38 48009024
2 0 40 64049832
4 756 42 70709184
6 4032 44 102958128
8 20412 46 124782336
10 60480 48 142254252
12 139860 50 189423360
14 326592 52 237588120
16 652428 54 248250240
18 1020096 56 344391264
20 2000376 58 397510848
22 3132864 60 433936440
24 4445532 62 554879808
26 7185024 64 671393772
28 10747296 66 677557440
30 13148352 68 908374824
32 21003948 70 1018507392
34 27506304 72 1079894844
36 33724404

2'0. 37 . 5 . 7 = 78382080. The automorphisms preserving the complex

lattice form Mitchell's complex reflection group of half this order (and
isomorphic to 6· POIi(3) . 2 and to 6· psu4 (3) . 2). Other names for
Mitchell's group are 6· u4 (3) ·2,6· HO(4,3 2) ·2 [DiclJ, [2 1 ; 3]2 (see
[ShelJ) , [321]3 (see [BenO]), and W(K 6) [Cohn It is number 34 on
Shephard and Todd's list [She2l See also [EdglJ, [Edg2l Theta series:
9 K12(z) = r/>0(2z)6 + 45 r/>0(2z)2r/> , (2z)4 + 18r/>,(2z)6
= 1 + 756q4 + 4032 q 6 + 20412 q 8 + ... (131)

(Table 4.11). For further information see later chapters and [Con37],
[Con38], [Fei2], [Haml], [Har6], [Har7], [Lin6], [Lin7], [TodlJ, [Tod2].

10. The 16-dimensional Barnes-Wall lattice A 16

This lattice appears to have been first published by Barnes and Wall in
1959 [BarI8], and since then has been rediscovered by many authors.
There are several constructions. Construction B of Chap. 5 applied to the
first-order Reed-Muller code of length 16 leads to the generator matrix
shown in Fig. 4.10, for which det = 256, minimal norm = 4, kissing
number T = 4320, the minimal vectors consist of 480 of the form
r'/2(± 22,0'4) and 3840 of the form r 1/ 2(± 18,08), where the positions of
130 Chapter 4

4
2 2
2 0 2

o
2 0 0 2
2 0 0 0 2
2 0 0 0 0 2
2 0 0 0 0 0 2
_I 2 0 0 0 0 0 0 2
./22 0 0 0 0 0 0 0 2
2 0 0 0 0 0 0 0 o 2
2 0 0 0 0 0 0 0 002
I I I 0 I 0 I I 0 0 I
o I I 0 I 0 I I 0 0 I
001 I 0 I 0 I I 0 0 I
000 I o I 0 I I 0 0 I
I I I I I I I I I

Figure 4.10. Generator matrix for 16-dimensional Barnes-Wall lattice

A 16• The last five rows are a generator matrix for the first-order Reed-
Muller code of length 16.

the ± 1's form one of the 30 codewords of weight 8 in the first-order

Reed-Muller code and there are an even number of minus signs. The
packing radius p = 1, density .:l = 1['8/16'8! = 0.01471... (15=1/16), and
covering radius R = p.J3, a typical deep hole being r 1/ 2 (I6,0IO), where
the six 1's are at a deep hole in the first-order Reed-Muller code
(corresponding to a "bent function" of four variables [Mac6, Chap. 14]) -
see §5 of Chap. 6. Theta series (Table 4.12):

TallIe 4.12. Theta series of 16-dimensional Barnes-Wall lattice A 16 •

m N(m) m N(m)
0 1 32 8593797600
2 0 34 11585617920
4 4320 36 19590534240
6 61440 38 25239859200
8 522720 40 40979580480
10 2211840 42 50877235200
12 8960640 44 79783021440
14 23224320 46 96134307840
16 67154400 48 146902369920
18 135168000 50 172337725440
20 319809600 52 256900127040
22 550195200 54 295487692800
24 1147643520 56 431969276160
26 1771683840 58 487058227200
28 3371915520 60 699846624000
30 4826603520 62 776820326400
Certain Important Lattices and Their Properties 131

e Aiz) = V2 {8 2 (2z) 16+8 3(2z) 16+8 4(2z) 16+ 308 2 (2z )88 3(2z )8}

= 1 + 4320q4 + 61440q6 + ... . (132)

A 16 may be constructed from the Leech lattice A24' There are
involutory symmetries of A24 that fix a 16-space, and the portion of A24
that lies in such a space is a copy of A 16. The automorphism group of A 16
is the centralizer of such an involution in the automorphism group of A24 ,
considered modulo the involution itself. The group has order
g = 221 . 35 . 52. 7 = 89181388800, and structure 2 1+ 8 . Ot(2), where
Ot(2) is the simple group of order 212 . 35 . 52. 7 = 174182400.
A 16 may also be obtained by applying Construction C (or D) to either
of the following sequences of binary codes: [16, 1, 161, [16, 11,4],
[16, 16, 11 or [16, 5, 81, [16, 15, 21 (see §3.4 of Chap. 5, Example 7 of
Chap. 7, §8.2f of Chap. 8). Other constructions will be found in §4 of
Chap. 6 and in §4 of Chap. 8, where A 16 appears as a self-dual lattice over
Z[e 1ri / 4 1. The last-mentioned construction implies that A~6 == A 16 .

11. The 24-dimensional Leech lattice A24

This lattice was discovered by Leech in 1965 [Lee5]. We shall give several
constructions in this book. The following definition is probably the
simplest. We state it in three equivalent forms.
(i) The Leech lattice A24 is generated by all vectors of the form

(133)

where the + 3 may be in any position, and the upper signs are taken on a
"~-set", i.e. the set of coordinates where a codeword of the binary Golay
code ~24 is 1.
(ij) Equivalently, we label the coordinate positions 00, 0, 1, ... ,22 and let
Q denote the set of nonzero quadratic residues modulo 23. Then A24 is
spanned by the vectors
a (212,0 12), 23 vectors, supported on a translate
{ {oJ U Q} + i, 0 ~ i ~ 22,
a (- 3, 123 ), a single vector,
a (± 42,0 22 ), 2· 24· 23 vectors, (134)

where a = 1/.J8 (see §3.2 of Chap. 10).

(iii) Equivalently, A24 consists of the vectors
a(0+2c+4x), (135)
a(1+2c+4y), (136)

where a = 1/.J8, 0 = (0 24 ), 1 = (124), C E ~24 (regarding the components

of c as real O's and l's rather than elements of F 2) and x, y E Z24 satisfy
132 Chapter 4

2: Xi ==0(mod2), 2:Yi == 1 (mod 2). (135) are called the even vectors
and (136) the odd vectors. (See §4.4 of Chap. 5.)
Other constructions for the Leech lattice are given in §5.7 of Chap. 5
(from the [24,12,9] quadratic residue or Pless code over F 3); §6 of Chap. 6
(as a laminated lattice); Example 12 of Chap. 7 and §3.6 of Chap. 10 (as
a complex 12-dimensionallattice, using the [12, 6, 6] Golay code ~12 over
F 3); §2 of Chap. 8 (as a three-dimensional icosian lattice, an analog to the
Turyn construction of ~24); §3 of Chap. 8 (from Construction A f , a
generalized version of Construction A); §5 of Chap. 8 (McKay's
construction from a Hadamard matrix of order 12); §7.3b of Chap. 8
(Craig's construction as an ideal in the cyclotomic field Q(ehi/39»; §7.5 of
Chap. 8 (Thompson's construction from the cyclotomic field Q(ehi/23»;
Chap. 11 (the MOG definition - see below); Eq. (7) of Chap. 17 (from
D 24 ); Chap. 17 and §5 of Chap. 18 (as a neighboring lattice to A [4);
Chap. 24 (twenty-three constructions, one for each Niemeier latticeD;
Theorems 1, 2 and 3 of Chap. 26 (three constructions of the Leech lattice
as a Lorentzian lattice), and Theorem 1 of Chap. 27 (as the Coxeter
diagram of the automorphism group of the Lorentzian lattice 11 25 ,1)'
In a certain sense the construction given in §6 of Chap. 6 is the
simplest of all, since it states that the Leech lattice may be built up by
"induction", starting with the I-dimensional lattice of integers and each
time extending to a densest lattice in the next dimension. The lattices
constructed in this way can be seen in Fig. 6.1 of Chap. 6; in 24
dimensions one obtains the Leech lattice. This is a "no-input"
construction! Another is given in Chap. 27.
But for calculating with the Leech lattice it is best to use MOG
(Miracle Octad Generator) coordinates, which are arranged in a 4 x 6
rectangle. The MOG coordinates for ~-sets are obtained from the
codewords of the hexacode (§2.5.2 of Chap. 3) by replacing each digit by a
column of length 4 as shown in Fig. 4.11. The ~ -sets are obtained either
by giving each digit an odd interpretation in any way such that the top row
contains an odd number of points, or an even interpretation in any way
such that the top row contains an even number of points. The Leech
lattice is spanned by the vectors (± 28 ,0 (6 ), where the support is a ~-set
and there are an even number of minus signs, together with the vector
(-3,1 23 ). MOG arrays are extensively studied in Chap. 11.

o w w o w w
ODD EVEN
Figure 4.11. The two odd interpretations of a digit are the sets on the
left or their complements. Similarly the two even interpretations are the
sets on the right or their complements.
Certain Important Lattices and Their Properties 133

8000 0000 0000 0000 0000 0000

4400 0000 0000 0000 0000 0000
4040 0000 0000 0000 0000 0000
4004 0000 0000 0000 0000 0000
4000 4000 0000 0000 0000 0000
4000 0400 0000 0000 0000 0000
4000 0040 0000 0000 0000 0000
2222 2222 0000 0000 0000 0000
4000 0000 4000 0000 0000 0000
4000 0000 0400 0000 0000 0000
I 4000
2222
0000
0000
0040
2222
0000
0000
0000
0000
0000
0000
./8 4000
2200
0000
2200
0000
2200
4000
2200
0000
0000
0000
0000
2020 2020 2020 2020 0000 0000
2002 2002 2002 2002 0000 0000
4000 0000 0000 0000 4000 0000
2020 2002 2200 0000 2200 0000
2002 2200 2020 0000 2020 0000
2200 2020 2002 0000 2002 0000
0222 2000 2000 2000 2000 2000
0000 0000 2200 2200 2200 2200
0000 0000 2020 2020 2020 2020
-3 I I I I I I I I I I I I I I I I I I I I I I I

Figure 4.12. Generator matrix for the 24-dimensional Leech lattice A24
in standard MOG coordinates (the rows of the matrix are divided into six
blocks of 4, corresponding to the columns of the MOG, read downwards).

The generator matrix for A24 given in Fig. 4.12 has been chosen to be
compatible with the MOG coordinates. For A 24, det = 1, minimal norm
= 2, and kissing number T = 196560, the minimal vectors consist of
27 .759 = 97152 of the form 8- 1/2(± 28,0 16), where the positions of the
± 2's form one of the 759 codewords of weight 8 in f(j24 and there are an
even number of minus signs, 24· 212 = 98304 of the form (133), and
2· 24· 23 = 1104 of the form 8- 1/ 2(± 42,0 22 ). The vectors in the first
three layers are shown in Table 4.13, with the signs suppressed (see also

Table 4.l3. Vectors in the first three shells of the Leech lattice. A(n)i
indicates the Leech vectors of norm 2n (or type n) and shape i; the signs
have been suppressed.

Class Shape Number Class Shape Number

A(O)I (0 24 ) 1 A (4)2+ (2 16 0 8) 2 11 .759
A (2)2 (2 80 16 ) 27 . 759 A (4)2- (2 16 0 8) 2 11 .759.15

A (2)3 (31 23 ) 2 12 . 24 A (4)1 (351 19) 212 [254 ]

A (2)4 (4 20 22 ) 22 [224] A (4)4 (440 20) 24 [244]

A (3)2 (2 12 0 12) 2 11 .2576 A (4)4+ (422 80 14) 29 . 759. [126]

A (3)3 (331 21 ) 212 [234 ] A (4)4- (42 12 0 11 ) 2 12 .2576. 12

A (3)4 (42 80 15 ) 28 . 759 . 16 A (4)5 (53 2 121) 212 [234 ]. 3

A (3)5 (5121) 2 12 . 24 A (4)6 (62 70 16 ) 27 . 759 . 8

A (4)8 (80 21 ) 2 1 .24
134 Chapter 4

§3.2 of Chap. 10). The first 15 layers are described on page 181 of
[ConI6l. The packing radius p = I, density 11 = 1r12/12! = 0.001930 ...
(0 = 1), covering radius R = .J2 p = .J2, there are 23 different types of
deep hole (Chap. 23), one of which is 8- 1/2(4,0 23 ) (this is an "octahedral"
hole, surrounded by 48 lattice points), and 284 types of shallow hole
(Chap. 25). The thickness 0 = (21r)12/12! = 7.9035 .... The Voronoi cell
has 16969680 faces, 196560 corresponding to the minimal vectors and
16773120 to those of the next layer (from [Con38l or Theorem 10 of
Chap. 21). A24 is the unique integral lattice of determinant I and minimal
norm 4 (Chap. 12). Its automorphism group is Coo (or '0), of order
2223 9 54 72 11' 13· 23 = 8315553613086720000 (137)

(see Chaps. 10, 1 J). Theta series:

0A;iz) = 0 E /z)3 - 720 1l 24(Z) (138)

1 45
= g{02(z)8 + 0 3(z)8 +0 4(z )8)3 - 16 {02(Z )0 3(z)O 4(z)}8 (139)

1 69
= T{02(Z )24+0 3 (z )24+04(Z )24) - 16
{02(Z )0 3 (z)O 4(Z)}8 (140)

00

= ~ N(m)qm = 1 + 196560q4 + 16773120q6 + (l4J)

m=O

where (cf. Eq. (53) of Chap. 2)

N(m) = 6~~~0 [~II [; ]-T[;]] (142)

and 1l24(Z), T(n) are defined in (39), ~II (n) in (103). The first 50
coefficients N (m) are given in Table 4.14. The decompositions into prime
factors of the first 20 coefficients N (m) are given in [SI08, Table xl. The
coefficients grow rapidly, and

N (m) ::::: 65520

691 I
~ [l!!:-]
2
(J 43)

is a good approximation (see Eqs. (54), (56) of Chap. 2). If the origin is
shifted to a deep hole of octahedral (or A [4) type the theta series is
1 3
g{02(z)8 +0 3(Z)8 -0 4(Z )8]3 + 16{02(Z )0 3 (Z )0 4(z)}8 (J 44)
00

= ~ N m ' qm = 48q2 + 4096 q 3 + 97152 q 4 + ... , (J45)

m-2
 Certain Important Lattices and Their Properties 135

Table 4.14. Theta series of 24-dimensional Leech lattice A 24 .

m N(m) m N(m)
0 1 52 348188700764268000
2 0 54 527108117540659200
4 196560 56 786767288036446080
6 16773120 58 1156841376897024000
8 398034000 60 1680521645295642240
10 4629381120 62 2409208986562560000
12 34417656000 64 3417887115322439760
14 187489935360 66 4792384230947389440
16 814879774800 68 6658492791534948000
18 2975551488000 70 9154704673132462080
20 9486551299680 72 12486419179545402000
22 27052945920000 74 16869989277755228160
24 70486236999360 76 22632233269428619200
26 169931095326720 78 30102943984468992000
28 384163586352000 80 39789443408903042400
30 820166620815360 82 52181704975196160000
32 1668890090322000 84 68053817735463006720
34 3249631112232960 86 88114798569141227520
36 6096882661243920 88 113523923563982568000
38 11045500816896000 90 145290076878792867840
40 19428439855275360 92 185116016465911440000
42 33213186220032000 94 234407918373703925760
44 55431591273414720 96 295640558277712240320
46 90344564568760320 98 370725724086681600000
48 144355739339448000 100 463209975930723119280
50 226066364540190720

with the convention that 0' II (x) and T(X) are 0 if x is not an integer.
Other references dealing with the Leech lattice are [Bayl], [Bro4), [Lep2),
[Lin6], [Lin8], [Mill], [Tit6], [Tit7], [VosIl. The "odd Leech lattice" 0 24
(Chap. 6, Appendix) was discovered by O'Connor and Pall in 1944
[O'Col).
 5
Sphere Packing and
Error-Correcting Codes
John Leech and N. J. A. Sloane

Error-correcting codes are used to construct dense sphere packings in n-

dimensional Euclidean space Rn , and other packings are obtained by taking
cross-sections or building up by layers. In this way we construct the
densest packings known in all dimensions n ~ 29 and in some higher
dimensions, as well as a number of other interesting packings.

1. Introduction
In this chapter we make systematic use of error-correcting codes to
construct dense sphere packings. By use of cross-sections we then obtain
packings in spaces of lower dimension, and by building up packings by
layers we obtain packings in spaces of higher dimension. The packings
constructed in this chapter include the densest packings presently known in
all dimensions up to 29, as well as a number of other interesting packings.
They are summarized in Tables 1.2, 1.3 and Fig. 1.5 of Chap. 1.
We begin by defining the coordinate array of a point in §1.1, and then
§2 describes Construction A, which is most effective in up to
15 dimensions. §3 describes Construction B, effective in dimensions 8 to
24. §4 digresses to deal with packings built up from layers, while §5 gives
some special constructions for dimensions 36, 40, 48 and 60. §6 deals with
Construction C, which generalizes A and B and is especially effective in
dimensions that are powers of 2. The constructions make use of the
material on error-correcting codes given in §2 of Chap. 3.
Packings built up by layers are considered further in Chap. 6, while
Chaps. 7 and 8 give additional properties and generalizations of
Constructions A, Band C. This chapter is a revised and updated version
of [Leel0).
Sphere Packing and Error-Correcting Codes 137

1.1 The coordinate array of a point. The coordinate array of a point

x = (x 1, . . . ,xn ) with integer coordinates is obtained by writing the
binary expansions of the coordinates Xi in columns, beginning with the
least significant digit [Lee5, §1.421 Complementary notation is used for
negative numbers, as illustrated in the display below. The first, second,
third, ... rows of the coordinate array are called the l's, 2's, 4's, ... rows
respectively. The l's row of the array comprises the l's digits of the
coordinates, and thus has O's for even coordinates and l's for odd
coordinates. The 2's, 4's, 8's, ... rows similarly comprise the 2's, 4's, 8's, ...
digits of the coordinates. For example, the coordinate array of the point
(4,3,2, 1,0, -1, -2, -3) is
o o 0 o 1 l's row
o 1 1 0 0 o 2's row
o
0 0 0 4's row
00000 8's row

The number of rows is potentially infinite, but all rows after the first few
are identical.

2. Construction A
2.1 The construction. Let C be an (n,M,d) binary code. The following
construction specifies a set of centers for a sphere packing in Rn.
Construction A. x = (x h"" xn) is a center if and only if x is congruent
(modulo 2) to a codeword of C.
Thus a point x with integer coordinates is a center if and only if the l's
row of the coordinate array of x is in C. A lattice packing is obtained if
and only if C is a linear code. Construction A is a generalization of the
construction of A4 as given in [Lee4, §1.11, and is studied further in
Chaps. 7 and 8, where several generalizations are given.
2.2 Center density. On the unit cube at the origin,
{O ~ Xi ~ 1 : i = 1, ... ,n },

the centers are exactly the M codewords. All other centers are obtained
by adding even integers to any of the coordinates of a codeword. This
corresponds to shifting the unit cube by two in any direction. Thus all the
centers may be obtained by repeating a building block consisting of a
2 x 2x x 2 cube with the codewords marked on the vertices of the
1 x 1 x ... x 1 cube in one corner.
Each copy of the 2 x 2 x ... x 2 cube contributes M spheres of
radius p (say), so the center density obtained from Construction A is
(2)
138 Chapter 5

If two distinct centers are congruent to the same codeword their

distance apart is at least 2. If they are congruent to different codewords
then they differ by at least 1 in at least d places and so are at least Vd
apart. Thus we may take the radius of the spheres to be
p = V2 min {2, Vd}.

2.3 Kissing numbers. Let S be a sphere with center x, where x is

congruent to the codeword c. Candidates for centers closest to x are as
follows. (a) There are 2n centers of the type x + «±
2)on-l) at a
distance of 2 from x. (b) Let {Ai (c)} be the weight distribution of C with
respect to c (§2.2 of Chap. 3). Since there are Ad (c) codewords at a
distance of d from c, there are 2d Ad (c) centers of the type
x + «± J)don-d) at a distance of vd
from x. Therefore the number of
spheres touching S, the kissing number of S, is
2d Ad(c) if d < 4,
r<S) = 1
2n + 16A 4 (c) if d = 4,
2n if d > 4.

2.4 Dimensions 3 to 6. Let C be the [n,n-I,2] linear code consisting of

all codewords of even weight (§2.3.4 of Chap. 3). There are V2 n (n - 1)
codewords of weight 2, and applying Construction A we obtain the
checkerboard lattice Dn in Rn (§7.1 of Chap. 4), with p = 1/.J2,
{j = r(n+2)/2 and kissing number T = 2n (n - 1). The lattice points are
alternate vertices of the cubic lattice zn.
As mentioned in §1.5 of Chap. I, for n = 3,4, 5, Dn is the densest
possible lattice packing in Rn, and is denoted by An' since it is also a
laminated lattice (see the next chapter). In R3 and R5 there are equally
dense nonlattice packings (§4.2 below, §6.5 of Chap. 4, [Lee6]).
The densest six-dimensional lattice, E6 == A6 , is not directly given by
this version of Construction A, but will be obtained by stacking layers of
A5 in §4.2, as a section of A7 in §4.5, and from a generalization of
Construction A in §S of Chap. 7.
2.5 Dimensions 7 and 8. From now on we apply Construction A to codes
with minimal distance d = 4. The packing obtained is of spheres of unit
radius, and has center density (j = Mr n and kissing number
T = 2n + 16A4(c).
Let Hn denote the binary matrix obtained from an n x n Hadamard
matrix (§2.13 of Chap. 3) upon replacing the +I's by O's and -I's by l's.
We assume that Hn has been normalized so that the first row and column
are all O's. Up to obvious equivalences there is a unique H s, and if the
first column is deleted the rows form the [7,3,4] code .Yt; (§2.4.1 of
Chap. 3), with A4 = 7. By applying Construction A we obtain E7 == A7 ,
the densest lattice packing in R7, with {j = r 4 and T = 2·7 + 16·7 = 126
(§S.2 of Chap. 4).
Sphere Packing and Error-Correcting Codes 139

The rows of H8 together with their complements form the [8,4,41

Hamming code .Yl8, with A4 = 14 (§2.4.2 of Chap. 3). This is also a
first-order Reed-Muller (or RM) code. From this code we obtain E8 ;;;; A8 ,
the densest lattice packing in R8 , with () = r4 and T = 2·8 + 14· 16 = 240
(§8.1 of Chap. 4).
2.6 Dimensions 9 to 12. H12 is also unique up to equivalence. Let ~ be
the 11 x 11 matrix consisting of the vector 11011100010 (with 1's at
position zero and at the quadratic residues modulo 11) together with its
cyclic shifts. Then H 12 may be taken as

0 otr 1
[ (5)
o ~ .
The sums modulo 2 of pairs of rows of H 12 and the complements of such
sums form the 132 vectors of a Steiner system S(5, 6,12) (see [Lee41 for
example). Since no two of these vectors can overlap in more than four
places, they form a (]2, 132, 4) code, every codeword having weight six.
This code may be increased to a (]2, 144,4) nonlinear code by adding six
codewords of type 010 12 and six of type 0 2 110 , the six 12 and the six 02
being disjoint sets. Several different versions of this code are possible,
depending on the relationship between the positions of the 1's in the 12
"loose" codewords and the vectors of the Steiner system.
By shortening the (]2, 144, 4) code we obtain (] 1, 72, 4), (10, 38, 4)
and (9, 20, 4) codes. Codes equivalent to these four were first given by
Golay [Go121 and Julin Dull], and alternative constructions and
generalizations are given in [Litl], [Mac61, [Roml], [S10181, [S1020l.
Applying Construction A to the various versions of the (]2, 144, 4)
code we obtain non lattice packings in R12 with center density () = 144· r12
= r 8 . 32 = 0.03516 ... , which is less than that of K12 (§9 of Chap. 4), but
in the most favorable versions some spheres touch as many as 840 others,
as we now show. Let c be a codeword of weight 6. Any vector of weight 4
and length 12 is contained in exactly 8 vectors of weight 5, and therefore
in exactly 4 codewords of weight 6 (since any vector of weight 5 is
contained in a unique codeword of weight 6). So the number of codewords
of weight 6 that are at a Hamming distance of 4 from c is 3 [: 1 = 45.

If the 12 loose codewords are chosen so that there are three words
010 12 whose 1's coincide with pairs of 1's from c, and three words 0 2 110
whose O's coincide with pairs of O's from c, then there are an additional 6
codewords at a Hamming distance of 4 from c and A4(c) = 45 + 6 = 51.
Based on these codes we obtain packings in R12 with a maximal kissing
number of 2· 12 + 16·51 = 840 for those spheres whose centers are
congruent to c.
There are still several inequivalent versions of these packings,
depending on the choice of the remaining loose codewords, but they all
have the same maximal kissing number of 840, and are collectively called
P12a in Table 1.2.
140 Chapter 5

Other choices for the 12 codewords give packings with maximal kissing
number of 824 or 808. It can be shown that, regardless of the choice of
the 12 loose codewords, the average kissing number obtained is equal to
770 2/3. (This is true even for those packings where the maximal kissing
number is less than 840.)
For RlI we set XI2 = 1 in P 12a , and find the maximal kissing number to
be 566 in the most favorable cases, collectively called P lIa , but it can be
only 550 or 534 in other cases (if 010 I is altered to 011). Except in this
last case the average is 5197/9. (P lIa is studied further in [Vetil.)
For RIO, if we set Xli = XI2 = I in P l2a , where XII, Xl2 are a pair of
both 010 12 and 11°02, we obtain packings P lOb with () = r 8 • 32 and a
maximal kissing number of 500. If instead we set xl2 = I, xlO = 0 where
XIO, Xl2 is not a pair of either form, we obtain packings with () = r 9 ·19.
In the most favorable cases, called P lOa , the maximal kissing number is
372. The average kissing number is 353 91I9 for P lOa and 340 1/3 for P lOb •
For R9 we set XIO = 0, XII = Xl2 = I in P l2a and obtain packings P 9a
with () = 2-7 . 5 and a maximal kissing number of 306; the average kissing
number is 235 3/5.
An alternative derivation of P lOb may be given. The tetrads of a
Steiner system S (3, 4, 10) (see §3.1 of Chap. 3) form 30 codewords of
length 10, weight 4 and Hamming distance at least 4 apart. For example
the tetrads may be taken to be the cyclic permutations of 1110001000, of
1101100000 and of 1010100100. By including the zero codeword and five
codewords of weight 8, several different 00, 36, 4) codes are obtained.
Construction A then gives the packing P lOb .
Similarly, the packings P 9a , P lOa , P lla may be obtained by applying
Construction A to the (9, 20, 4), 00, 38, 4) and (11, 72, 4) codes. All
these are nonlattice packings.
A (10, 40, 4) code was found by Best [Bes1l; it consists of the cyclic
permutations of 1010000001, 1100101100, 0001010111 and 0111111010.
Although nonlinear, the weight distribution with respect to each codeword
is 0 1 422 6 12 85 . By applying Construction A a lO-dimensional nonlattice
packing P lOe is obtained with () = r 7 . 5, in which each sphere touches
2·10 + 16· 22 = 372 others.
Best [Besl], [Bes3] also found eleven different constant weight codes,
each consisting of 35 codewords of length II, weight 4 and distance 4; one
such code is shown in Fig. 5.1. By applying Construction A to anyone of
these codes an II-dimensional arrangement of spheres is obtained in which
some spheres touch 2· 11 + 16·35 = 582 others. These are collectively
denoted by P lIe in Table 1.2. The densities of these packings are quite
low.
2.7 Comparison of lattice and nonlattice packings. It is still an open
question whether there exists any nonlattice packing with density exceeding
that of the densest lattice packing in its dimension. But the non lattice
packings P lOe , P lIa and P l3a of §4.3 below have density greater than that
Sphere Packing and Error-Correcting Codes 141

I I I I I I I I I I I 100 0 000 0 0 0 000 0 0 0 000 0 0 0 0 0

I I 100 0 000 0 001 I I I I I I I 100 0 0 0 0 000 0 0 0 0 0
000 I I I 000 000 I I I 000 000 I I I I I I 0 000 0 0 0 0
000 I 001 I 0 0 0 0 100 I I I 000 I 100 001 I I I 0 0 0 0
o 0 0 0 I 000 I 100 I 0 0 000 I I 100 I 100 I I I 0 I 000
10000 I 001 0 I 000 0 I 0 0 100 101 0 I 0 I 0 0 I 0 I I 0
o I 000 I 100 0 0 I 000 0 I 0 0 100 I 0 101 100 0 I I I 0
001 0 000 I 100 I 0 I 0 0 I 000 I 000 I 100 I 0 I 0 I 0 I
001 000 I 001 100 0 I 001 100 I 0 0 0 0 I 0 I 001 101
100 I 000 001 0 I 0 I 000 I 0 I 000 I 001 001 100 I I
o I 0 0 I 001 0 0 I 000 I 100 0 0 I 0 I 0 0 I 000 I 0 I 0 I I
Figure 5.1. The columns form a constant weight code found by Best,
containing 35 words of length 10, weight 4 and minimal distance 4.

of the densest known lattice packings. It has already been mentioned in

§2.2 of Chap. 1 that it follows from Watson's work [Wat7] that P 9a has a
higher kissing number than is possible in any 9-dimensional lattice.

3. Construction B
3.1 The construction. Let C be an (n, M, d) binary code with the
property that the weight of each codeword is even. A sphere packing in R n
is given by:
Construction B. x = (XI, .•. ,xn) is a center if and only if X is
congruent (modulo 2) to a codeword of C, and ~7-1 Xi is divisible by 4.
Thus a point X with integer coordinates is a center if and only if the l's
row of the coordinate array of x is a codeword c E C and the 2's row has
either even weight if the weight of c is divisible by 4, or odd weight if the
weight of c is divisible by 2 but not by 4. A lattice packing is obtained if
and only if C is a linear code. Construction B is a generalization of the
construction of Ag as given in [Lee4, §1.11, and will be studied further in
Chap. 7.
3.2 Center density and kissing numbers. The number of centers is half
that of Construction A, so 0 = M pnrn-I.
Let S be a sphere with center x, where x is congruent to a codeword c.
Candidates for centers closest to x are (a) the 2n (n - 1) centers of type
x + «± 2)2on-2), and (b) the 2d- 1Ad(c) centers congruent to the
codewords differing minimally from c. Therefore the kissing number of S
is
2d-1 Ad (c) if d < 8,
T(S) = 1 2n (n - 1) + 128Ag(c) if d = 8, (6)
2n (n - 1) if d > 8;

and S has radius p = 1/2 min {.Jd, Jg}.

142 Chapter 5

3.3 Dimensions 8, 9 and 12. In R8 we apply Construction B to the

repetition code (0 8, 18) to obtain the packing E 8 again. In R9 we use the
code (0 9, 180) and obtain the lattice packing A9, with p = .J2, 0 = r 4S
and T = 272.
In R 12 we use the (I 2, 24, 6) code formed from the rows of H 12 and
their complements to obtain the nonlattice packing L 12 (see [Lee4)) with
0= r 16 '3 7 and T = 704. The [12, 2, 81 code (012,0 418,1 40414, 1804) gives
the lattice packing Arr, with 0 = r S and T = 648.
3.4 Dimensions 15 to 24. In Rl6 the [16, 5, 81 first-order RM code gives
the lattice packing Al6 having 0 = r4 and T = 4320 (§10 of Chap. 4).
Shortening this code by equating a coordinate to zero we obtain the [15, 4,
81 simplex code, which gives the lattice packing A 1S in R 1S having 0 = r 4 .S
and T = 2340.
Let Ci , i = 0, 1, ... , 5, denote the shortened code obtained from the
[24, 12, 81 Golay code f(j24 by setting i coordinates equal to zero, and let
ai denote the number of codewords of weight 8 in Ci . Then aD = 759,
al = 506, a2 = 330, a3 = 210, a4 = 130 and as = 78 (see Table 10.1 of
Chap. 10).
The sequence of lattice packings in R24-i, i = 0, 1, ... ,5, obtained
from Ci has Oi = r(i+2)/2 [Lee4, §2.41. In R 19 , R20, R21 these are A 19 ,
A 20, A 2b the densest packings known, but in R22 to R24 they can be
improved, as will be shown in §§4.4, 4.5. We denote the 24-dimensional
lattice by hA 24 ; this consists of half of the points of the lattice A24
constructed in §§4.4, the "even" part of that lattice.
Remark. Constructions A and B cannot be successful for large n.
Consider for example Construction B applied to a code with d = 8. It is
easy to show that log2 0 ~ nl2 - 3 log2 n + 0 OOg2n), as n - 00, and so,
using (I8) of Chap. 1, log2 ~ ~ - V2 n log2 (n hr e). But much denser
pac kings exist - see §1.5 of Chap. 1. Also A 8 ~ (~), and therefore
T = 0 (n 8 ) at most, whereas much higher values of T are known - see §2.2
of Chap. 1 or §6.5 below.

4. Pac kings built up by layers

4.1 Packing by layers. The basic idea is very simple (cf. [Lee61, [Lee7)).
Let A be a lattice sphere packing in R n with finite covering radius R
(Eq. (3) of Chap. 2), and let D (A) be the set of deep holes in A. (The
same construction may also be applied to suitable nonlattice packings.)
A layer of spheres in Rn + l is a set of spheres whose centers lie in a
hyperplane, and whose cross section in the hyperplane is A. In this
hyperplane there are two distinguished sets of points, the set C of centers
of A, and the set D of the deep holes in A.
We shall try to build up a dense sphere packing in Rn + l by stacking
such layers as closely as possible. We therefore place adjacent layers so
that the set C of one layer is opposite to some or all of the set D of the
Sphere Packing and Error-Correcting Codes 143

next. Since the layers are lattice packed, if one point of C is opposite to a
point of D, then so are all the points of C.
It may happen that D is more numerous than C, in which case several
inequivalent packings may be produced in Rn+l. These may be lattice or
nonlattice packings or both. Examples will be found below and in
succeeding sections. (So far we have not found an example where no such
lattice packing can be formed, but this does not seem impossible. Indeed,
there might even be a case where the deep holes fail to give either lattice
or nonlattice packings.)
For example let A be the square lattice Z2 in R2. Then C comprises
the vertices of the squares and D their centers. The layers are placed so
that the vertices of the squares in one layer are opposite to the centers of
the squares in the next. Since C and D are equally numerous, this
arrangement is unique, and the face-centered cubic lattice packing A3 in
R3 is obtained (§6.3 of Chap. 4).

The situation is different if A is chosen to be A 2 , the hexagonal lattice

in R2. Now D consists of the points marked 'b' or 'c' in Fig. 1.3b, and is
twice as numerous as C (the points 'a'). The spheres of a layer can be
placed opposite to the points 'b' or to the points 'c' of the adjacent layer.
As already discussed in §1.3 of Chap. 1, if the layers are stacked so that
the spheres in the two layers adjacent to any layer are opposite points
marked with different letters, then the lattice A3 is obtained. But if the
adjacent spheres on both sides of each layer are opposite points marked
with the same letter, then the (non lattice) hexagonal close packing (§6.5
of Chap. 4) is obtained, having the same density as A3 • If we require
uniform packings there is no further choice, as all layers have to be fitted
alike, and this construction gives precisely these two packings.
In general, the layers are placed just far enough apart that the smallest
distance between centers in adjacent layers equals the smallest distance
between centers in the same layer. There are four cases which can arise
here. The spheres of one layer may (i) not reach, (ij) just touch, or (iii)
penetrate, the central hyperplane of an adjacent layer, or (iv) it may be
possible to fit the spheres of one layer into the spaces between the spheres
of the adjacent layer and so merge the two layers.
Examples of case (i) are n = 2 (as above) and 3-5 (in §4.2). In case
(ii), extra contacts may occur because of spheres on opposite sides of a
layer touching. Examples are the case n = 6 in §4.2, and the packing P l3a
in §4.3. In case (iii), the layers on each side of a given layer must be
staggered to avoid overlapping. Examples are the construction of E 8 from
layers of D7 as given in [Lee7], and the local arrangements P 14b , P I5a in
§4.3. Case (iv) doubles the density of the original packing. Examples of
this in R 8, R12, R24 and R 48 are given in §4.4 and §5.6.

It is sometimes advantageous to stack layers which are not lattices.

Examples are the local arrangements of spheres in R 11, R 14 and R 15 and
the non lattice packing P l3a in RI3 to be described in §4.3.
144 Chapter 5

This laminating construction is investigated further in the next chapter.

4.2 Dimensions 4 to 7. We consider packings in Rn +1 formed by stacking
layers of the lattice Dn for n ~ 3. The Delaunay cells (§1.2 of Chap. 2)
are of two kinds: each omitted vertex of zn (such as the glue vector [2]
given in §7.1 of Chap. 4) is the center of an octahedral cell f3n, while the
center of each cube of zn (such as the glue vectors [1] and [3]) is the
center of a "half-cube" h'Yn [Cox20, p. 155]. The latter cells are twice as
numerous as the former, and we shall regard the cells h'Yn as being colored
alternately black and white.
For n = 3 the octahedral cells f33 are larger than the tetrahedral cells
h'Y3, and so we place the spheres of each layer opposite to octahedra of the
adjacent layer. As the octahedra and spheres are equally numerous we
arrive uniquely at the lattice packing D4 == A 4 .
For n = 4 the cells f34 and h'Y4 are congruent, the lattice D4 being the
regular honeycomb {3, 3, 4, 3} [Cox20, p. 136]. Thus there is a threefold
choice for placing each layer: the spheres of each layer may be placed
opposite to the omitted vertices or the black cells or the white cells. In this
case we obtain either the lattice packing D5 == A5 or three distinct uniform
nonlattice packings, all having the same density and kissing numbers
[Lee6].
For n > 4 the cells h'Yn are larger than the cells f3n' so for maximal
density in Rn +! we stack the layers with the spheres of each layer opposite
to the cubes of the adjacent layers. At each stage the spheres may be
placed opposite to the black or the white cubes. For n = 5 or 6 we obtain
in this way two uniform packings, the lattice packing A n + 1 and a non lattice
packing of equal density. For n = 5 both have the same kissing numbers.
For n = 6 each sphere in the lattice packing E7 == A7 has two more
contacts than in the nonlattice packing, because it touches spheres in layers
two away from it. This is an example of case {in above. For n == 7 only
the lattice packing E s == As is produced. This is an example of case {iii}
above, where the adjacent layers have to be staggered to avoid overlap.
4.3 Dimensions 11 and 13 to 15. In RII we construct a local arrangement
P lIb of 580 spheres touching one sphere (inferior to P lie of §2.6), by
stacking three partial layers. The central layer consists of the 500 centers
of PlOb touching one sphere, with eleventh coordinate zero. The two outer
layers consist of all points of the form
(CI, ... ,CIO, 0) - (Ih, ... ,V2, ± Ih.J6),

where (CI, ... ,CIO) runs through the (10, 40, 4) code, and contain 80
points. It does not seem possible to extend this to a dense space packing.
In RI3 we take anyone of the packings P 12a and its translates by
multiples of the vector «Ih)12, ± 1), and obtain a family of nonlattice
packings collectively called P 13a • Any of these packings has (j = T S·3 2 and
a maximal kissing number of 840 + 2·144 + 2 = 1130, the last 2 being
caused by a sphere touching spheres two layers away. The average kissing
number is 1060 2/3.
Sphere Packing and Error-Correcting Codes 145

The tetrads of a Steiner system S(3, 4, 14) (see [Hanl]) form 91

codewords of length 14, weight 4 and Hamming distance at least 4 apart.
For example, if the fourteen coordinates are labeled 1 2 3 l 5 6 7 and l' 2'
3' 4' 5' 6' 7', the tetrads may be taken to be 1 2 3 6, 2' :,' 6' 7', 1 2 4 2',
64'6'7',271'6',244'6',571'3',153'7',144'7',471'4',
67 l' 2', 562' 3', 453' 4', and their transforms under the permutation
(J 2 3 4 5 6 7) (J' 2' 3' 4' 5' 6' 7'). Construction A then gives a local
arrangement Pl4a of 1484 spheres touching one sphere. This can be
improved, however, by cutting down to thirteen dimensions and rebuilding.
The best section of S (3, 4, 14) in RI3 has 65 codewords, giving a local
arrangement P l3b with 26 + 16·65 = 1066 contacts, inferior to P l3a'
We now form a local arrangement P I4b in RI4 by stacking five partial
layers. The central layer is P l3b . The adjacent layers are (c, 0) - «V2)13,
V2.Ji) and (c', 0) - «V2) 13, - Vl.Ji), where c runs through the [13, 8, 41
shortened Hamming code and c' denotes c + (12,0") reduced modulo 2.
The two outer layers contain just two spheres each: (±1, 0 12 , ±.Ji).
Thus the central sphere touches 1066 + 2·256 + 2· 2 = 1582 others.
Similarly in R I5 we form a local arrangement P I5a from five partial
layers. The central layer is P 14a . The adjacent layers are <C, 0) - «V2)14,
V2.J2) and (c', 0) - «V2)14, -V2.J2), where c runs through the [14, 9, 41
shortened Hamming code and c' = c + (J 2, 0 12) reduced modulo 2. The
outer layers are (XI • ... • XI4, ± .J2), with XI • ... • X14 all zero except for
one pair X2i-J, X2i which are ± 1 in all four combinations. The central
sphere touches 1484 + 2· 512 + 2· 28 = 2564 others.
4.4 Density doubling and the Leech lattice A24 . In R8 two copies of D8
can be fitted together without overlap to form the lattice packing E 8 ;;;; A8,
the second copy being a translation of the first by «Vl)8).
In R24 it was shown in [Lee5, §2.311 that two copies of the 24-
dimensional packing obtained in §3.4 from the Golay code CC24 may be
fitted together without overlap to form a packing A24 (now usually called
the Leech lattice), with 0 = 1 and T = 196560. The second copy is a
translation of the first by (-1 V2, (V2)23), and this produces A24 in the form
given in Eqs. (J 33) - (J 36) of Chap. 4, rescaled.
4.5 Cross sections of A 24 . All of the densest known lattice packings in
fewer than 24 dimensions occur as sections ofA 24 . There are two main
sequences of sections, Ai and K i , i = 0,1 ..... 24, where the subscript
indicates the dimension, and Ai ;;;; Ki for i ~ 6 and i ~ 18. Because of
the symmetry of A24, there are several different ways of describing some of
the sections.
The sequence of lattice packings Ai is defined as follows ([Lee4, §2.4;
Lee5, §2.41]). A 23 , A22 and A21 are obtained from A24 by equating any
two, three or four coordinates to each other. A21 (again), A 20 and AI9 are
obtained by equating any three, four or five coordinates to O. We recall
from §3.1 of Chap. 3 that codewords of weight 8 in CC 24 form the octads of
a Steiner system S (5, 8, 24). Because of the fivefold transitivity of the
Steiner system, the choice of coordinates in forming A 19 , ... ,A 23 is arbitrary.
146 Chapter 5

AI9 (again), A IS , AI7 and AI6 are obtained by equating to 0 the sum of
eight coordinates forming a Steiner octad and also any four, five, six or all
of them.
The [16, 5, 8] RM code occurs as a subcode of C€24, and we associate
the remaining 16 coordinates with coordinate positions of this subcode.
Any two intersecting octads of the [16, 5, 8] code split the coordinates into
four tetrads. A 15 , A 14, A\3 and AI2 are obtained by equating to 0 the sum
of the coordinates forming a tetrad and none, anyone, any two or all of
them. All, AIO and A9 are obtained from A\2 by equating any two, any
three or all four coordinates to each other in one of the remaining tetrads.
A9 (again) and As are obtained by equating to 0 any three or all four
coordinates in a tetrad.
A7, ... ,A4 are obtained as sections of As in the same way that AII, ... ,As
are obtained from A \2, defining any four coordinates of the eight to be a
tetrad. Finally, A 3, ... ,A I are obtained from A4 (calling the four
coordinates a tetrad) in the same way that A I5 , ... ,A\3 were obtained from
A 16·
The sequence A), ... ,A24 includes the densest known lattice packings in
R I, ... ,R 10 and R 14, ... ,R24. But in dimensions 11 to 13 the lattices K II, K \2,
K \3 of the Kn sequence are denser than An [Lee5], [Leel0]. These are
described in the next chapter. K\2 is the Coxeter-Todd lattice (§9 of
Chap. 4), with 0 = r 3 and T = 756. K II , K\2 and K\3 are the densest
known lattice packings in these dimensions, although P lla and P\3a are
denser nonlattice packings. Also the nonlattice packings P lOa, P lOe are
denser than A 10.
We remark that there is a close analogy between the packings D 3, ••• ,Ds
of §2.4 and the packings in RI9, ... ,R24 of §3.4, in that the first three are the
densest known, the last can be doubled in density, and the densest
intermediate packings can be derived either as sections of the doubled
packing or by building up in layers.

5. Other constructions from codes

5.1 A code of length 40. We construct a binary code of length 40 which
will be used in §5.2 to obtain a sphere packing. Let C I be the [16, 11,4]
Hamming code. The 140 codewords of weight 4 form the blocks of a
Steiner system S(3, 4,16). Let C€24 be the [24, 12,8] Golay code, with
coordinates arranged so that I S0 16 is a codeword. There are 759 codewords
of weight 8, forming the octads of a Steiner system S (5, 8, 24).
Let C be the code of length 40 consisting of all codewords of the form
(x,y,z) where y E C), z E C), and (x,y +z) E C€24.
Theorem 1. C is a [40, 23, 8] linear code.
Proof We note that if the eight coordinates belonging to an octad are
deleted from all the codewords of C€24, the truncated codewords form two
copies of C I (see [Ber5] or Chap. 11). In C, y and z can each be chosen
in 211 ways and then, by the previous remark, there are two ways of
Sphere Packing and Error-Correcting Codes 147

choosing x so that (x, y+z) E ~24. Therefore C contains 223 codewords.

Since
(x,y,z) = (x,y +z,O) + (O,z,z), wdx,y,z) ~ wdx,y +z,O) ~ 8

by construction, and so C has minimal weight 8.

Theorem 2. The number of codewords of minimal weight in C is 2077.
Proof Let (x ,y, z) = (x ,y + z, 0) + (0, z, z) be a codeword of weight 8.
There are five cases. (1) If y = z = 0, then x = 18. (2) If y ¢ 0, z = 0,
then there are 758 codewords of the form (x,y) E ~24' y ¢ o. (3) If
y = 0, z ¢ 0, again there are 758 possibilities. (4) If y = z ¢ 0, then
wdy) = 4, x = 0, and there are 140 choices for y. (5) If y ¢ 0, z ¢ 0,
Y + z ¢ 0, then x = 0, wt (y + z) = 8 = wt (y) + wt (z). Therefore y and
z have disjoint sets of 1'so y can be chosen in 140 ways and for each of
these there are three choices for z so that (0, y + z) E ~24, giving 420
codewords. The total of (1)-(5) is 2077.
Remark. If (O,y,z) E C,then (O,y+z) E ~24andsoy+z is in the [16,
5,8] 1st order RM code and therefore (y,z) is in the [32, 16,8] 2nd order
RM code.
5.2 A lattice packing in R40 . A lattice packing in R40 can be obtained
from the code C of §5.1 as follows. The 40 coordinates are divided into
8 + 16 + 16 corresponding to the division of the codewords of C.
Then X=(x,y,z) is a center if and only if the 2's row of X is in C,
the 1's row is all O's or is 0 on one of the sets 8, 16, 16 and 1 on the other
two, and the 4's row has even weight if the set of 8 is even or has odd
weight if the set of 8 is odd.
The centers closest to the origin are shown in Table 5.1. (The centers
described in the first line of the Table have 1's row 08132 , 2's row equal to
any codeword (O,y,z) E C, and 4's row chosen to make all the nonzero
coordinates equal to ±1, and by the remark at the end of § 5.1 the number
of such centers is 216. The other types of centers are easily counted.}
Thus we have a lattice packing ~ with p = 2 .J2, (j = 24 and T = 531120.

Table 5.1. Minimal vectors (x,y,z) in A40 • In the lines marked <*), one
entry ± 1 must be replaced by +3.

x y z Shape Number

zero odd odd 08(± 1)32 2 16

odd zero odd (±1) 8016(±1)16 (:j:) 24.2 12
odd odd zero (± 1)8(± 1) 160 16 (:j:) 24.2 12
even even even 032(±2)8 2 .2077
7

even even even 038(±4)2 40·39·2

531120
148 Chapter 5

5.3 Cross sections of A40 . A40 contains A24 as a section, and for
1 ~ n ~ 16 has a section A40 - n in R40 - n of center density 16 times that
of A24 - n • For instance in R36 there is a section A36 with {) = 2 and
T = 234456. A36 is obtained from A40 by equating to zero four coordinates
from a tetrad contained in one of the sets of 16. A32 is obtained by
equating to zero all coordinates from an octad contained in one of the sets
of 16, and has {) = 1 and T = 208320.
5.4 Packings based on ternary codes. Let C be an (n, M, d) ternary
code. By analogy with the constructions based on binary codes we obtain
sphere packings in Rn from the following constructions.
Construction A 3. x = (XI, . . . ,xn ) is a center if and only if X is
congruent (modulo 3) to a codeword of C.
Construction B3. In addition, ~7-1 Xi is divisible by 2.
From Construction A3 we obtain a packing of spheres for which
p = min {3/2, 1/2Jd}, {) = M pnrn, (7)

and from Construction B 3

min {3/h, 112 Jd}' if d is even,
{ (8)
p = min {3/h, 112 (d + 3) 1/2}, if d is odd,

(9)

Other constructions from ternary codes are described in §8 of Chap. 7.

5.5 Packings obtained from the Pless codes. By applying Construction B 3
to the Pless double circulant codes (§2.10 of Chap. 3) we obtain the lattice
packings shown in Table 5.2.

Table 5.2. Lattice obtained by applying Construction B3 to the Pless

codes.

n Code Radius {) Name

12 [12,6,61 rl/23 1/2 r7 DI2

24 [24, 12, 91 3 1/2 rl hA24
36 [36, 18, 121 31/2 rl P36p
48 [48, 24, 151 r1/2 3 r 2S 324 hP 48p
60 [60, 30, 181 rl/23 r 31 330 P 60p

By applying Construction A3 to the [12, 6, 61 Golay code ~12 we

obtain the lattice Df2 with p = .J30. and {) = r6. It is known that D\2 can
be doubled in density to give Df2, and in fact quadrupled to give A 12
[Lee4, §2.l1.
We show in §5.7 that the packings in R24 and R48 of Table 5.2 can also
be doubled in density, to give A24 and a lattice P 48p ' On the other hand
Sphere Packing and Error-Correcting Codes 149

we have not found doublings of the packings P36p and P 60p , the technique
of §5.7 failing because of the existence of codewords of maximal weight
with odd numbers of each sign [MaI2]. (See Postscript, p. 156.)
In P 36p the centers closest to the origin are those of the type
«± 1) 120 24 ) and correspond to codewords of minimal weight, so
T = AI2 = 42840. In P 60p , T "" AI8 + 2·60·59 = 3908160.
5.6 Packings obtained from quadratic residue codes. As pointed out in
§§2.8, 2.10 of Chap. 3, there are [24, 12, 9] and [48, 24, 15] ternary
quadratic residue codes with the same weight distributions as the
corresponding Pless codes. Applying Construction B3, we obtain packings
h A24 and h P 48q. It will now be shown that these can be doubled in
density to give A24 in R24 and another lattice P 48q in R48.
5.7 Density doubling in R24 and R48. Any of the lattices h A24 , h P 48p and
h P 48q may be doubled in density, by adding a second copy which is a
translation of the first by (-2V2, (Y2)n-I).
To show this, we must verify that no point of the original lattice is
closer to (-2V2, (V2)n-l) than the origin. Every point of the original lattice
differs from this point by at least V2 in every coordinate. If any coordinates
are congruent to -1 (modulo 3) then at least three are, by §2.10 (ij) of
Chap. 3, and so at least three coordinates differ by 1 V2. On the other hand
if all coordinates are congruent to 0 or 1 (modulo 3), then by §2.10 (iv)
they are congruent to one of the types On, 1n or on/21 n/2. Since the sum is
even, at least one coordinate differs by 2 V2.
In R24 the doubled packing has center density 1 and so must be A24
(regardless of whether it came from the Pless code or the quadratic residue
code) since this lattice is unique (Chap. 12).
In R48 we obtain a lattice which we call P 48p by doubling that given by
the [48, 24, 15] Pless code, and one which we call P 48q from the [48, 24,
15] quadratic residue code. Both P 48p and P 48q have 0 ., r 24 . 324 . The
kissing numbers for P 48q and P 48p can be obtained from the weight
distribution given in Table 3.2. The centers closest to the origin are shown
in Table 5.3; their total is T = 52416000 (in agreement with the value
found in Eq. (57) of Chap. 2). The cross-sections of P 48p are investigated
in §2 of Chap. 6. The lattice P 48q was first found by Thompson [Th07]

Table 5.3. Minimal vectors in either P 48q or P 48p.

Coordinates Number
(±3)2046 2·48·47
(+2) (± 1) 14033 15·415104
(± 1) 18 030 20167136
(+2V2) (±Y2)47 48·96
(±IY2)3(±Y2)45 4·6503296
150 Chapter 5

using the construction given in §7.5 of Chap. 8. We return to these lattices

in Example 9 of Chap. 7, where we give their determinants and theta
series, etc.

6. Construction C
6.1 The construction. Let Ci = (n,Mi,d), i = 0, 1. ... ,a, be a family
of codes with d i = 1" 4a - i , where I' = 1 or 2. A sphere packing in Rn is
given by:
Construction C. A point x with integer coordinates is a center if and only
if the i's row of the coordinate array of x is in Ci , for i = 0, 1, ... ,a.
This is a generalization of the constructions for Rn, n = 2m , given in
[Lee4, §1.6]. Construction C follows the trend of A and B in successively
imposing more restrictions on the coordinate array. In general a non lattice
packing is obtained. A modification of this construction (Construction D)
given in §8 of Chap. 8 always produces lattice packings provided only that
the codes Ci are linear and nested, i.e. satisfy Ci ~ CHI'
6.2 Distance between centers. If the first row in which two centers differ

°
is the i's row, then (i) if i > a their distance apart is at least 2a+l, and
(ij) if ~ i ~ a they differ by at least 2i in at least d i coordinates, and
so are at least (di · 4i) 1/2 = .J:Y. 2a apart. Thus we may take the radius of
the spheres to be p = .J:Y. 2a- l .
6.3 Center density. How many integer points x satisfy the conditions of
Construction C? The fraction of x's with l's row in Co is Morn, of these
a fraction Mlr n have 2's row in C b and so on. Thus the fraction of
integer points which are accepted as centers is MoMI ... M a 2-(a+lln, and
so the center density is
0= MoMI ... M a 2-(a+lln pn = MoMI ... M al'n/2r 2n . (10)

If the codes are linear, and dim Ci = k i , then

a
log2 0 = ~ ki - 2n if I' = 1, (Ila)
i-O
a 3n
log2 0 = ~ k i - - if I' = 2. (1 t b)
i-O 2

For large n we have (from (18) of Chap. 1)

a n
log2 d = ~ k i - - log2 (8n/Tre) + o (JOg2n) . (12)
i-I 2

For example, we may obtain a 64-dimensional non lattice packing by using

the following codes from Table 5.4:
C o:[64,t,641, C 2 :[64,57,4], C I :[64, 28,16], C 3 :[64, 64, I1 (I3)
(here 1'=1, a=3). From (Ita) the center density is 0 = 222. In §8.2e of
Chap. 8 we obtain a lattice packing with the same density.
Sphere Packing and Error-Correcting Codes 151

6.4 Kissing numbers. We shall calculate the number of spheres touching

the sphere at the origin. From the discussion in §6.2 this means that we
must find the number of centers of type (±2,)drO n-dr for each
r = 0,1 ..... a. A coordinate equal to +2' contributes to the coordinate
array a column with a single one in the 2"s row; while a coordinate equal
to -2' has ones in the i's row for all i ~ r. The l's in the 2"s row form
a codeword c (say) in C" and the minus signs must be at the locations of
the ones in a codeword in C,+l n C,+2 n ... n Ca. The 2"s row can
be chosen in Adr ways, and for each of these the number of ways of
choosing the signs is equal to the number of codewords in
C,+l n C,+2 n ... n Ca which are contained in the codeword c of C,.
Let the latter number be N, (c).
Then the number of centers of the desired type is ~ N, (c), and

(14)

where ~ denotes the sum over all c E C,. If N, (c) = N, is independent

of c, this becomes T = ~~_oAdrN,.
6.5 Packings obtained from Reed-Muller codes. As an example of
Construction C we take C, to be the (2r)1h order RM code of length
n = 2m , to obtain a packing Pn, in Rn. This and similar pac kings were
given in [Lee4].
If m is even, 'Y = 1 and a = m12, while if m is odd, 'Y = 2 and
a =(m - 1) 12; in both cases the radius is 2(m -2)/2. Since RM codes are
nested, N, (c) is equal to the number of codewords in the (2r + 2)1h order
RM code of length 2m which are contained in a codeword of minimal
weight in the (2r)1h order RM code, and this is the same as the number of
codewords in the 2nd order RM code of length 2m - 2,. Thus

log2 N, (c) = 1 + [m -1 2r] + [m -2 2r] . (15)

Using the properties of RM codes given in §2.7 of Chap. 3, the center

density and maximal kissing number are then found to be
(16)

(17)

The details may be found in [Lee41 (The relation between the k-parity of
a vector as defined in [Lee4] and RM codes is that a binary vector of
length 2m has k -parity if and only if it is in the (m - k)1h order RM
code.}
152 Chapter 5

The packing Pnr in Rn , n = 2m , is a lattice if and only if n ~ 32. The

Barnes-Wall lattice packing BWn (§8.2f of Chap. 8) in dimension n = 2m
coincides with P nr for n ~ 32 and has the same density and maximal
kissing number for all n.
6.6 Packings obtained from BCH and other codes. Both very low and very
high order RM codes contain the maximal possible number of codewords,
but intermediate order RM codes are poor. For n ~ 64 there are BCH
codes with more codewords, and these too are known not to be optimal.
For length 64 extended cyclic codes are known which are better than BCH
codes. Unfortunately for lengths greater than 99 cyclic non-BCH codes
have not been extensively studied. Cyclic codes of length n ~ 63 are listed
in [Chel1, [Pet3, Appendix 0], and Promhouse and Tavares [Prol1 have
extended this table to n = 99. It would be nice to have a list of the best
cyclic codes of length 127. In Table 5.4 we give a selection of the best
codes presently known with length n = 2m (5 ~ m ~ 8) and minimal
distance d = i. The Goethals codes are nonlinear (see [Goell, [Goe2],
[Mac6, Chap. 15, §7]); the others are linear. The trivial [n, n, 11,
[n, n-l, 2] [n, 1, n] codes have been omitted from the table.

Table 5.4. The best codes known of length 2m .

n log2 M d Name

32 26 4 Hamming
16 8 2nd order RM
6 16 1st order RM
64 57 4 Hamming
45 8 BCH
46 8 extended cyclic
47 8 Goethals
24 16 BCH
28 16 extended cyclic
7 32 1st order RM
128 120 4 Hamming
106 8 BCH
78 16 BCH
43 32 BCH
8 64 1st order RM
256 247 4 Hamming
231 8 BCH
233 8 Goethals
199 16 BCH
139 32 BCH
55 64 BCH
9 128 1st order RM
Sphere Packing and Error-Correcting Codes 153

For completeness we mention that the [64, 46, 8] code in Table 5.4 is
obtained by adding a zero-sum check digit to the [63, 46, 7] cyclic code
having generator polynomial
g(x) - M(I) (x) M(5) (x) M(9) (x) M(21) (x), (I 8)

where M(;) (x) is the minimal polynomial of ol, a = primitive element of

F64 (see §2.4 of Chap. 3). Similarly the [64, 28, 16] code comes from the
[63, 28, 15] cyclic code with generator polynomial
5
g(x) - M(2I) (x) II M(2i+I) (x)
i-O

[Pet3, Appendix D].

Since the center density in Construction C is proportional to the
number of codewords, by using the best codes from this table we obtain
considerable improvements over the packings of the previous section. In
R64 for example, as we saw in §6.3, we obtain a nonlattice packing with
{j = 222. For m ~ 7 we use extended BCH codes to obtain an infinite
family of nonlattice packings Pnb in Rn, n -= 2m . For m = 7, 8, 9 these
have {j = 285 , 2250, 2698 respectively, giving improvements over both BWn
and P", by factors of 221 , 258 , 2 186 respectively. Lattices with the same
density as Pnb will be obtained in §8.2g of Chap. 8 using Construction D.
For dimensions n ~ 256 Craig's lattices (§6 of Chap. 8) are denser still -
see the discussion in §I.5 of Chap. 1.
The density of Pnb is estimated for all m in the next section.
By using Goethals' code in place of the BCH code with d = 8 we can
increase the density of P nb by a factor of 4 for n = 2m ~ 256 when m is
even. However, this does not affect the following analysis.
BCH codes are nested and so also are the extended cyclic codes of
length 64 in Table 5.4. Therefore calculation of the kissing numbers in
these packings requires knowledge of the number of codewords of minimal
weight d in each code used and the number of codewords in the code with
minimal distance d /4 that are wholly contained in such minimal weight
codewords. This appears to be a difficult problem.
6.7 Density of BCH packings. This section contains lower and upper
bounds and an asymptotic expansion (Theorem 3) for the center density of
the packings in Rn obtained by using extended BCH codes in Construction
C. Here "code" will mean "extended BCH code of length n = 2m ." Two
packings are considered. Packing (a) uses the codes of (actual) Hamming
distance 1,4, 16, ... , 4[m/21, and packing (b) uses the codes of (actual)
Hamming distance 2,8,32, ... ,2·4[(m-I)/21. Then let Pnb denote the
denser of the two packings.
By the remark at the end of §2.7 of Chap. 3, the code of designed
distance 2h has actual distance 2h. Let this code have dimension k h. But
there may be codes of designed distance less than 2h also having actual
distance 2h. Let Kh be the largest dimension of such a code. Then
154 Chapter 5

k>.. .:;;; K>.. < k>.._I' For example, for codes of length 128 it is known that
ks = 36, Ks = 43, k4 = 78 [KasIl
If 6a and 6b denote the center densities of the two packings, then from
(lla), (llb) we have
[m/21 [(m-ll/21
log26a = ~ K2i - 2n, log2 6b = ~ K2i+1 - 3n/2.
i-O i-O

An algorithm for calculating k>.. is given in [Ber4, §12.31 The results

of this algorithm may be stated as follows. Let numbers ai,j be defined by
ai,j = 2j - 1, for .:;;; j .:;;; i,
ai,j = ai,j-i + ai,j-i+1 + ... + ai,j-b for 1 .:;;; i < j.
Then ko=n,km=1 and k>..=m+am->..,m for O<X<m. Let
Ai(X) = ~j:laiJxj. From the definition of aiJ,
x + 2X2 + ... + iXi
Ai(X) = 2 ..
1 - x - x - ... - x'

Let 1 - x - x2 -' .. - Xi = II~ _I (I - tvx). The partial fraction

expansion of Ai (x) is then
i
Ai(X) = ~ --=~
v-I 1 - tv x '

and so
i
aiJ = ~ tt.
v-I

The polynomial G(y) = yi+1 - 2yi + 1 = (y - I)(yi - yi-I - ... - Y - I)

has roots 1, tl, ... ,ti' A sketch of G(y) shows that one root, tl (say), is
close to 2. In fact for i > 1, G (2 - 2 1- i ) < 0 and G (2 - 2- i ) > 0, so
2-2 1- i < tl < 2 - r i , i> 1.

It has been shown by Mann [Man21, while calculating the number of

codewords in the BCH codes of length 2m - 1 and distances 2>" and 2>" + 1,
that Itvl < 1 for v = 2,3, ... ,i. Then
(2 - 21-i)m - i + 1 < ai,m < (2 - ri)m + i-I, i > 1.

Collecting these results we find that lower and upper bounds to log26a
are respectively

i-I
[m/21
2m ~ (I - 22i - m - 2)m - 2m + O(m 2).
i-I
Sphere Packing and Error-Correcting Codes 155

The sum in the lower bound may be written as

[m/21-clogm [m/21-dlogm [m/21
~ + ~ + ~ = ~l + ~II + ~III '
i-O i -[m/21-clogm i -[m/21-dlogm

where c > 1 and d < 1 are constants, and the logarithms are to base 4.
Then
[m/21-clogm m
~I > ~ (I-m4 i ·rm ) > T- Io g4 m +0(1),
i-O

since c may be made arbitrarily close to 1. Also ~II = 0 OOg4m), since c

and d can be made arbitrarily close together. Finally, since d < 1,
o< ~III < (I-m- d )m(d·log4 m)+1 = 0(1).

Therefore

since n = 2 m . Similar arguments apply to the upper bound and to bounds

for log2ob. This proves:
°
Theorem 3. Let be the center density of either of the packings in Rn ,
n = 2m , obtained by using extended BCH codes in Construction C. Then
log2o - 11m log2n - 11m log210g2n as n -+ 00. (20)

It follows using Eq. (I8) of Chap. 1 that the actual density Ll = Vno of
these packings satisfies
log2 Ll - - V2 n log210g2 n .

6.8 Packings obtained from Justesen codes. A higher density than (21)
can be achieved by replacing some of the BCH codes by Justesen codes
(§2.6 of Chap. 3). Suppose n = m2m = 22a, and take 'Y = 1. For
o ~ i ~ 5 let Ci consist of the 0 codeword alone, for 6 ~ i ~ 5/ slog2 (2a)
let Ci be a Justesen code of minimal distance 4 a - i shortened to have length
n, and for 5/ slog2 (2a) < i ~ a let Ci be an extended BCH code of
minimal distance 4 a - i as in §6.7. Then it can be shown [Slo1] that
Construction C produces a nonlattice packing with
I
- log2 Ll > -6. (22)
n -

Remark. If we used codes meeting the Gilbert bound (Eq. (3) of Chap. 9)
in Construction C we would obtain nonlattice packings with
1
- log2 Ll > -1.2919 .... (23)
n -
156 Chapter 5

(However, although such codes exist, it is not known how to construct

them,) On the other hand, from the JPL bound (Eq. (I) of Chap. 9), the
density of any packing obtained from Construction C is bounded above by
1
- log2 ~ < -0.9042" .. (24)
n

Eqs. (23), (24) are to be compared with Eq. (46) of Chap. 1.

Postscript: a uniform construction for extremal Type II lattices in

dimensions 24, 32, 40, 48, 56, 64. Ozeki [Ternary code construction of
even unimodular lattices, preprintJ asserts that the following method can be
used to double the density of lattices obtained by applying Construction B3
to a ternary self·dual C of length n, for all n = 24, 32, 40 , ... , 64, just
as we did for n = 24 and 48 in §5.7. One adjoins a vector v of the form
«± '/2)n-l(:P/2», «± '/2)n-2, (_3/2)2) or «± '/2)n-l,3), according as
n == 0, 8 or 16 (mod 24), where the signs are chosen so that 2v is in the
undoubled lattice. He shows that if C has minimal distance 9 for
n = 24, 32, 40, or distance 15 for n = 48, 56, 64 then the resulting lattice,
suitably scaled, is an extremal Type II lattice (cf. Chap. 7, §7). The
required codes are briefly discussed in §10 of Chap. 7. We remark that
such codes of lengths 32 and 56 may be obtained by "subtracting" [Con24,
rc
§VII] the tetracode 4 (Chap. 3, (2.5.1) from the double circulant codes
S36 and S60 (Chap. 3, §2.10).
 6
Laminated Lattices
J. H. Conway and N. J. A. Sloane

We study the densest lattice packings that can be built up in layers. Start
with the I-dimensional lattice Al of even integer points; at the nth step
stack layers of a suitable (n - I)-dimensional lattice An - I as densely as
possible, keeping the same minimal norm; the result is a laminated lattice
An. In this chapter the density of An is determined for n ~ 48, all An are
found for n ~ 25, and at least one An is found for 26 ~ n ~ 48. The
unique A24 is the Leech lattice. Denser lattices than An are now known
for n ~ 30.

1. Introduction
A natural and familiar way to construct a lattice packing of spheres in n-
dimensional Euclidean space Rn is the following. Starting with the one-
dimensional lattice consisting of the even integer points, 2Z, we obtain a
two-dimensional packing by drawing a row of circles of radius 1 centered
at the even integer points along one axis, then a similar row next to it as
close as possible, then another row at the same spacing, and so on. This
produces the hexagonal lattice packing A 2• To go to three dimensions we
place a layer of billiard balls in the hexagonal lattice packing, then place a
similar layer next to it and as close as possible, then another layer at the
same spacing, and so on, obtaining the face-centered cubic lattice packing
A3 == D 3 . At each step, in going from i to i + 1 dimensions, we stack
layers (or laminae) consisting of copies of a suitable i -dimensional lattice
Ai as close together as possible. The resulting lattice(s} - for in some
higher dimensional spaces more than one lattice can be obtained in this
way - we call laminated lattices (a more precise definition is given in §2).
A typical n-dimensional laminated lattice will be denoted by An.
In this chapter we determine the density of An for n ~ 48 (see Fig. 1.5
of Chap. I and Table 6.I), find all An for n ~ 25 (see Fig. 6.1), and find
at least one An for 26 ~ n ~ 48. We see that laminated lattices are the
densest packings known in dimensions n ~ 29 except for n = 10, 11, 12
and 13.
158 Chapter 6

Table 6.1. The laminated lattice An has minimal norm 4, determinant An

and center density (, = A;;I12. The distance between the laminae An in
An + I is 7r;; 1/2, and the greatest subcovering radius of any An is h n. Note
that 7r n = 4 - hi and An = 7r n-1 An-I' Also hn ,,;; rn (the greatest
covering radius of any An), with equality if either h n ,,;; J3 or rn ,,;; J3.

n An "lrn hn2 n An "lrn hn2

0 1 4 0 24 I 2 2
1 4 3 1 25 2 3/2 512
2 12 8/3 4/3 26 3 4/3 8/3
3 32 2 2 27 4 I 3
4 64 2 2 28 4 I 3
5 128 312 512 29 4 3/4 13/4
6 192 4/3 8/3 30 3 2/3 10/3
7 256 I 3 31 2 112 712
8 256 2 2 32 1 1 3
9 512 312 512 33 1 3/4 13/4
10 768 4/3 8/3 34 3/4 2/3 10/3
11 1024 1 3 35 112 112 712
12 1024 1 3 36 114 112 712
13 1024 3/4 13/4 37 118 3/8 29/8
14 768 2/3 10/3 38 3·r6 113 1113
15 512 112 7/2 39 r6 114 15/4
16 256 1 3 40 r S 112 7/2
17 256 3/4 13/4 41 r9 3/8 29/8
18 192 2/3 10/3 42 3·r 12 113 1113
19 128 112 7/2 43 r l2 1/4 15/4
20 64 112 7/2 44 rl4 1/4 15/4
21 32 3/8 29/8 45 r l6 3/16 61116
22 12 113 1113 46 3·r 2O 1/6 23/6
23 4 1/4 15/4 47 r21 118 3118
24 1 2 2 48 r24 ? ?

History. A detailed account of this laminating process in up to eight

dimensions (and a description of the nonlattice packings that can be
constructed in the same way) was given in §4 of the previous chapter and
in [Lee7], where it is shown that A I, . . . ,As are unique and coincide with
the densest lattice packings Al ;;;; Z, A 2 , A 3 ;;;; D 3, D 4 , D s, E 6 , E 7 , Es in
these dimensions (see §1.5 of Chap. 1). It appears that in 1946 Chaundy
[Cha41 found the laminated lattices A9 and A lD , but did not show these to
be the densest possible lattices in 9 and 10 dimensions (which have still not
been determined). E. S. Barnes has informed us that in unpublished work
Chaundy also found some of the An in the next few dimensions.
Laminated Lattices 159

Ilo' ONE-POINT LATT ICE

II, ' I TEGER LATTICE
liZ' HEXAGONAL LATT ICE
1l 3 ' FACE -CENTERED CUBIC LATTICE
114' D.
~ ' O5
116'E6
117 ' E7
IIS'E8
IIg
IlW

11'5
11'6
1117
II,s
11'9
1120
liz,
1122

LEECH LATTICE IlZ4 IIZ3

11(' 1 II(~
25 . ..

Figure 6.1. Inclusions among the laminated lattices An. All An for
n ~ 24 are shown, while there are 23 A2s 's and probably a large number
of A 26 's. At least one An is known for 26 ~ n ~ 48.

We shall see that there are unique laminated lattices A IS and A 16,
which are in fact isomorphic to lattices found by Barnes and WaH [Barl8]
in 1959. The unique A24 is the Leech lattice, and the "main sequence" of
cross-sections
Ao, AI •.. .• AJO, Aria" A~a" A~a" A 14, A ls . .. . • A24 (1)

was described in [Lee4]-[Lee7] and the previous chapter, although there

these lattices were not proved to give the densest laminations for n > 8.
The members of this sequence are distinguished by having the highest
kissing numbers among the An (see Table 6.3 below). In Fig. 6.2 we give
particularly simple coordinates for these lattices as cross-sections of the
Leech lattice (cf. [LeeS]) using the MOG notation of Chap. 4, §11. Their
determinants are given in Table 6.1. The lattice A ~a, is Chaundy's J 12
(see [Cox29]).
It was shown in [Lee4], [LeeS], [LeelO] that there is a second
important sequence of sections of A 24 , denoted by Ko, K I • . . .• K 24. These
coincide with An for n ~ 6 and 18 ~ n ~ 24, but for n - 7 ... . . 17 are
not laminated lattices. K 12 was found by Coxeter and Todd ([Cox29]; see
also §9 of Chap. 4), Kll by Barnes [Bar9], and KJ3 by Leech
(unpublished) . Figure 6.3 gives coordinates for the Kn as sections of A 24 ,
160 Chapte r 6

00 00 00 ** 00 00 ** ** 00 00 ** **
00 0 0 00 ** 00 00 ** 00 A ** **
00 0 0 00
00 0 0 00 :1 00 00
0 0 00 ** *
** * 0
0 00
00
18 * * **
** **
00 00 00 ** 00 00 ** * 0 00 * * ** 00
00 00 00 It* 00 00 ** * 0 00 ** ** *0
O~
00
00
00
00 :1 00 0 0
00 00 **
** :0
00
00 ** **
** **
*0
*0

00 00 00 ** 00 00 ** ** *0
no 00 00 ** 00 00 ** ** *0
00 00 ** 00 00 11.20
*0
lJ~ 00 00 ** 00 00 ** **
** ** *0

00 00 00 ** *0 00 ** * 0 00 ** ** ***0
*0 00 00 ** 00 00 ** ** 00 ** **
*0 00 00 ** 00 00 ** ** 00 11.21 * *
** *0
*0 00 00 ** 00 00 ** ** 00 ** ** * 0

*0
*0
00
00
00
00
00
00
**
** ~ I: 00
00
00
** **
** **
00
00
00
**
**
** **
** *.
*0
*0 00 00 ** li~ 00 ** ** 00 ** ** :1
** ** ** ** **

** 00 00 ** *0 00 ** ** 0 0 ** ** **
*0 00 00 ** *0 00 ** ** 0 0 ** ** **
:1
AII\8lC
*0 00 00 11 ** 1~ 00 ** ** O ~ ** **
*0 00 00 ** 00 ** ** ** ** --

00 00 *0 00
o 0 0 0 II\8lC **
** *0 00 **
It It **
**
**
**
00 00 11.12 ** *0 00 ** ** **
00 00 ** * 0 00 ** ** It*
Figure 6.2. MOG coordinates for Ao • . . .• An as sections of the Leech
lattice A 24 . The figure shows Ao, AI;;;; Z;;; Ai> A 2 ;;;; Ab A3 ;;;; A 3 ;;;; D 3,
A4 .. D 4 , A5;;;; D 5, A6 ;;;; £6, A7 ;;; £7, Ag ;;;; E g, A 9, A IO , Al'l", A~" ,
Al'.\", A I4 , .... A small circle represents a zero coordinate, a hollow loop is
a set of coordinates adding to zero, an asterisk is a free coordinate, and a
line of asterisks is a set of equal coordinates.

and Table 6.2 gives their determinants. Some of the An for n ~ 16 and
the Kn for n ~ 13 were rediscovered by Skubenko [Sku I], who appears to
have been unaware of the earlier work on this subject. Particular lattices
A 2s •. . . • A40 are also described in the previous chapter. Leech has shown
that, for n ~ 21, the layers can be stacked to give nonlattice packings that
are as dense as An and Kn, except for AI> A 2, A 4 , As, KII and K 12 .
Sequels to this chapter. We have generalized this work in two ways in
[Con36], by considering (a) laminated complex and quaternionic lattices
(cf. §2.6 of Chap. 2), and (b) integral laminated lattices, where An is
required to be an integral (real, complex or quaternionic) lattice with the
Laminated Lattices 161

00 00 00 0 *o ** *0 00 00
00 00 00 00 **
00 ~ In* ** **
00 00 00 o 00 00 K12~ ** **
00 00 00 o 00 00 I ~: ** **

00 00 = ** ** -11* *0 00 00
00 00 00 11 0 00 00 ** ** **
00 00
00
00
00 00
00 00 It ~ 00 00 ** **
** **
**
**

00 00
000000
K 21-0-0-+-0-0-+-0-0-1
******
00 00 ~
00 00 K14[[]]]]]Q[] K20
**
** ** **
000000 00 00 ** *-11 **
** ** **
00 000* ** ** *0 00
00 00 ** ** **
00 00 00 ** ** **
00 00 00 ** ** **

IHljiftil
00 00 00
00 **
00 **
00 ** ** ** *
00 00 00 ******
00 00 00
K10 ** ** K22 ** * * * *
*****-11

0* ** ** 00 00 -11* *'*
00
00
00
00
00
00
~ *-11 ** **
** ** *-11
00 00 00 K11l!.1l11ltlJ ** **
-11* **
** **
**
** **
** **
*0 00 00 *-11 ** **
n )( ** ** ** ** *-11
:
~ ** **
** **
-11*
**
**
*-11
-11*
**
Figure 6.3. MOG coordinates for the lattices K o• . . .• Kn as sections of
the Leech lattice. The symbols have the same meaning as in Figure 6.2.

prescribed minimal norm. Let us temporarily denote the particular

laminated lattices constructed in §7 below in dimensions 25·48 by A~a) .
Then we show in [Con36] that the lattices
Ao, A4, Ag, ArlaX, A 16, A20, A 24 , Ai~), AW ..... Al~)

can be given the structure of quaternionic lattices (over the Hurwitz

integers), and that these are laminated quaternionic lattices. Furthermore
they also have the structure of Gaussian or Eisenstein lattices (cf. §2.6 of
Chap. 2) , and as such are again laminated lattices. Integral laminated
lattices with minimal norm M over a ring of integers J have a similar
definition to laminated lattices, with the additional requirement that the
layers must be stacked so as to form an integral lattice, i.e. so that all
inner products u . v belong to J. Real integral laminated lattices have been
investigated by computer by Plesken and Pohst [Ple6], [Poh2], and the
162 Chapter 6

Table 6.2. Determinants I( n of lattices Kn.

n Kn n K n n K n n K n n K n n Kn

0 1 8 576 16 576 24 1 32 9/4 40 32·r 1O

1 4 9 864 17 384 25 2 33 33 ·r4 41 3·r 1o

2 12 10 972 18 192 26 3 34 35·r 8 42 3·r 12

3 32 11 972 19 128 27 4 35 35·r 9 43 r l2

4 64 12 729 20 64 28 4 36 36 ·r 12 44 r l4

5 128 13 972 21 32 29 4 37 35 ·r ll 45 2- 16

6 192 14 972 22 12 30 3 38 35·r 12 46 3·r 2O

7 384 15 864 23 4 31 3 39 3J · r 1o 47 r21

8 576 16 576 24 1 32 9/4 40 32· r 1O 48 2- 24

theorems if. [Con36] explain some of their results. For completeness we

include a table from [Con36] as an appendix to this chapter, giving the
densest integral lattices presently known with minimal norms 2, 3 or 4 in
dimensions up to 24.

Certain other lattices also arise in a canonical way as integral

laminated lattices. For example the Coxeter-Todd lattice K 12 is the unique
6-dimensional integral laminated Eisenstein lattice with minimal norm 2
[Con36, §4.2J.

The method used to prove Theorem 10 of [Con36] could easily be

extended to obtain a proof of the theorem that the root lattices A o, AI> A 2,
A3 ;;;: D 3, D 4, Ds, E 6, E7 Eg (see Theorem 2 below) are the unique
laminated lattices Ao, ... ,Ag. The only extra step needed is a proof that
the deep holes in Ao, ... ,E7 belong to the dual lattices A~, ... ,E;. This
can be supplied using Norton's technique (see Chap. 22 and §5 below). In
the present chapter we deduce this theorem from the results of Blichfeldt
and Vetchinkin. The alternative proof would make our determination of
the An for n ~ 48 independent of Blichfeldt and Vetchinkin's work (the
inductive argument would then proceed one dimension at a time instead of
in steps of eight dimensions). Of course Theorem 3 below would still
depend on [Bli4] and [Vet2].

Lexicographic codes, which are a coding-theory analog of laminated

lattices, are studied in [Con41l.
Outline. The rest of this chapter is arranged as follows. The next few
paragraphs establish some notation, and §2 states the main results. In §3
we collect some properties of A), ... ,Ag that will be used throughout the
chapter and then §§4-7 are devoted to the proof of Theorem 1.
Laminated Lattices 163

Notation. As in §I.4 of Chap. 1, if two lattices Land M differ only by a

rotation, a scale factor and possibly a reflection, we say they are equivalent
and write L == M. We shall use three different norms for lattices. The
laminated lattices An will always have minimal norm 4, to correspond to a
packing of unit spheres. The root lattices An' Dn and En may have
minimal norm 2 or 4, depending on which coordinates are being used.
Finally in MOG notation (see §II of Chap. 4) the minimal norm is 32. If
the minimal norm of a lattice is f,L, its center density (in this chapter the
adjective "center" will usually be omitted) is

which becomes 5 = (det 0- 1/2 for a laminated lattice. A hole (see §1.2 of
Chap. 2) in a lattice Ln ~ Rn is a point of Rn whose distance from Ln is a
local maximum. The greatest distance of any hole from Ln is the covering
radius of Ln (Eq. (3) of Chap. 2), and a hole at this distance is called a
deep hole. By a k-dimensional section of a lattice Ln we mean a k-
dimensional lattice Mk ~ Rk ~ Rn such that Mk = Ln n Rk. The dual
lattice L; = {x ERn: x·y E Z for all y E Ln } (Eq. (24) of Chap. 1). If V
is a subspace of Rn the orthogonal subspace V-L = { x ERn: x· y =0 for
all y E V J.

2. The main results

Definition of laminated lattice An. Let Ao be the one-point lattice. For
n ~ 1 we take all n-dimensional lattices with minimal norm 4 that have at
least one sublattice An-I> and select those of minimal determinant. Any
such lattice is a laminated lattice An. When necessary we use superscripts
(e.g. A ~ax, A ~in) to distinguish between different An's.
The following geometrical ideas will be useful. The projection of a
given laminated lattice An onto the I-dimensional subspace (RAn_I)-L,
where An -I is one of its sublattices, is a I-dimensional lattice of minimal
norm 1I"n-1> say. Then An may be regarded as a union
... U A~:H U A~~I U A~~I U ...
of translates of An-I> A~j~1 having the coordinate i.J1I"n-1 in (RAn_I)-L.
The smallest norm, s say, of vectors in A~!) is at least 4 (and it will turn
out that in all the cases we use, s actually is 4). If v is a vector in A~~I
of norm s and the distance of v from A~~I is hn-I> then
1I"n-1 = S - h;_1 ~ 4 - h;_I. (3)

So hn-I> which we shall call the subcovering radius, is a lower bound for
the covering radius rn-I of An-I. It is easy to see that h n- I and rn-I are
equal if either is ~../3. (It seems likely that hn - I is always equal to rn-I>
although we cannot prove this.) If the determinant of An is denoted by Xn,
we have
(4)
164 Chapter 6

The density of An is A;; 1/2. It is clear from the definition that all An have
the same determinant, but only those An with the maximal subcovering
radius (h n ) can be extended to An+I'S.
Our main result is the following.
Theorem 1. The density An of any laminated lattice An for n ~ 48 is as
shown in Table 6.1 (see also Fig. 1.5 of Chap. O. The greatest
subcovering radius hn of any An for n ~ 47 is also shown in Table 6.1.
When hn ~ .J3. hn is also the greatest covering radius of any An. All An
for n ~ 25 are shown in Fig. 6.1. At least one An is known for
26 ~ n ~ 48.
In dimensions 26 and above the number of inequivalent An appears to be
very large (see §7). We see from Fig. 6.1 that for n ~ 24 there is only
one An, A~id, that is not contained in a An + l . However, Plesken (personal
communication) has shown that A~id is a sublattice of A 24 .
On the scale at which the minimal norm is 4, the An are integral
lattices if and only if n ~ 24. (For n ~ 25, An contains A 24 , a self-dual
lattice, but does not contain A24 as a direct summand, and so is not
integral.) The quadratic form associated with any laminated lattice is
perfect. (This follows immediately from [Bar9, Part II, Th. 2.11.)
The proof of Theorem 1 depends heavily on the next two theorems, the
first of which collects known results.
Theorem 2. For n = 0, 1 ..... 8 the densest lattice packing in Rn is
isomorphic to A o• Al ;;; Z. A 2• A3 ;;; D 3. D 4• D 5• E 6• E 7• Es respectively.
The laminated lattices Ao• A I •...• As are unique and are isomorphic to
these lattices. Their determinants and covering radii are shown in
Table 6.1.
Proof For the first assertion see [Bli4], [Vet2]; for the second see [Lee7]
and the previous chapter; and for the covering radii see for instance
Chap. 21, [Con28], [Cox201.
Theorem 2 certainly establishes the results of Theorem 1 for n ~ 8,
although of course it is much stronger than that.
Theorem 3. Let Lsm +, (m ~ 0, 0 ~ r ~ 8) be any lattice in RSm +, of
minimal norm ~ 4 with a section ASm having covering radius rSm. Then

- 42]'
[-4-rs m
det Lsm +, ~ ASm A, - (5)

If equality holds in (5) there is an additive map 8. a "gluing" map. from

A, into RAsm. such that the vectors of LSm+' have the form
u + vO + CV, where u E Asm , u + vO E RA sm , v E A" (6)

and
4-rs2 ] 1/2
c = [ --
4
-
m

'
Laminated Lattices 165

and such that

(8)

for all u E Asm. v EAr. (u, v) ~ (0,0). In other words LSm+r is

obtained by gluing a scaled copy of Ar to Asm. Furthermore LSm+r is a
A sm +,. and

2]r
4-rsm
[--4-
~Sm+r = ~Sm ~r (9)

Proof A typical vector of LSm+r may be written as VI + V2 where

VI E RA sm , V2 E (RAsm)..i.. The vectors V2 span an r-dimensionallattice
Lr , which is isomorphic to Lsm+r/Asm. Thus det LSm+r =
det Asm . det L r •
By adding a vector of ASm we can reduce the norm of VI to at most
r~m. But 4 ~ N(vl +V2) = N(vl) + N(V2), so the minimal norm of Lr is

r~r,
at least 4 - rim. Therefore by Theorem 2

det Lr ~ [4-;im

which proves (5).

If equality holds in (5) then we must have Lr = cAr' with c as in (7),
and the gluing map (J described in the theorem must exist. If 0 ~ s ~ r,
Ar has a section As (by Theorem 2), and restricting v to As in (6) we
obtain a section LSm+s of L sm +r . This shows that LSm+r is a laminated
lattice and completes the proof.
The Kn sequence. We shall not say as much about the rival sequence of
sections Kn of the Leech lattice. Up to n = 6, Kn =- An; K7 - KIO have a
lower density than A7 - A IO ; but then K II , KI2 and K\3 have higher
densities than A II - A \3. For n = 18 to 24 again Kn == An.
There is a similar sequence {Kn} of sections of A4S , obtained by gluing
Kn (for 0 ~ n ~ 24) to the deep holes in A24 . The determinant K n of Kn is
shown in Table 6.2 and Fig. 1.5 of Chap. 1, and MOG coordinates for
n ~ 24 are given in Fig. 6.3.
Relations between densities. The sequences of determinants {~n} and {K n}
possess a number of curious palindromic properties (see Tables 6.1, 6.2),
namely
24-n
~4S-n = K 4S-n =.!.
[ ] (0 ~ n ~ 48), (10)
~n K n 2
~24-n K 24-n
-- = -- = 1 (O~n ~24),
~n K n

~16-n = 2S-n (0 ~ n ~ 16), K 12-n = 36 - n (0 ~ n ~ 12), (12),(13)

~n K n
166 Chapter 6

These imply the useful relations

An K n
A24+n -- -2n ' K
24+n -- -
2n (0 ~ n ~ 24).

These identities are partially explained by the following theorem.

Theorem 4. Let Ln be a (not necessarily integra!) lattice in Rn , and let
E = Ln n S be a k -dimensional section of L n, where S is a k-
dimensional subspace of Rn. Let F = L: n S J., an (n - k) -dimensional
section of L:. Then

det F = det E . (I7)

det Ln

In particular, if det Ln = 1,
det F = det E . (1s)

Proof Let el, ... ,en be an orthonormal basis for Rn, chosen so that
el, ... ,ek is a basis for Sand ek+l, ... , en is a basis for SJ.. Since E
is a section of Ln, we can find an integral basis VI, . . • ,Vn for Ln so that
v I, . . . ,Vk is an integral basis for E. Let WI, . • . ,Wn be the dual basis

1
to VI. ... , V n , with Vi . Wj = oij. Then WI, . . . ,Wn is an integral basis for
L: and Wk 1 ' . . . ,Wn is an integral basis for F. There is a matrix

M = (A 0
B C ' with inverse
(A-* I
0 )
C-I, such that

=M

We now compute determinants (see Eq. (5) of Chap. 1):

det Ln = (det A)2 (det C)2, det E = (det A)2, det F = (det C)-2, and
(17) follows.

Corollary 5. Let Ln be either (j) an integral unimodular lattice or (ij) a

lattice which is equivalent to its dual and has been scaled so to have
determinant l. Then the smallest determinant of any k-dimensional
section of Ln is equal to the smallest determinant of any (n - k)-
dimensional section, for 0 ~ k ~ n.
Proof In case (i) Ln = L:, and in case Oi) the scaled version of Ln is
congruent to L: (cf. §lA of Chap. 1). In both cases det Ln = I, so
det E = det Fin (17).
Corollary 6 [LeelO, Th. 4.5.11. Let Ln be one of the lattices D 4, E 8, K 12 ,
A 16, A 24 , P 48p or P 48q , scaled so as to have determinant l. To every k-
Laminated Lattices 167

dimensional section of Ln there is a corresponding (n - k) -dimensional

section with the same determinant. The same result applies to the
particular A4S constructed in §7.
Proof The lattices mentioned all satisfy the hypotheses of Corollary 5.
Corollary 7 (cf. [LeeIO, Th.4.5.2]). (a) Each of Ao.· ... As.
A 16 • . . . • A 23 has the smallest determinant of any section of the Leech
lattice in its dimension. (b) For n ~ 23. An has the smallest determinant
of any n-dimensional section of A n+1. (c) For n ~ 23. Kn has the
smallest determinant of any n-dimensional section of Kn+l containing or
contained in K 12.
Proof (a) Ao •...• As are the densest possible lattices in their dimensions
(Theorem 2), which proves the first assertion, and the second follows from
Corollary 6. (b) Corollary 6 also implies that A 9 • . . . • A 15 have the
smallest determinant of any section of A 16. (c) Similarly K 7 • . . . • K II
have the smallest determinant of any section of K 12, and K 13 • . . . • K 17
have the smallest determinant of any section of A24 containing K 12. For
o ~ n ~ 6 and 18 ~ n ~ 24, Kn = An.
Remark. We are unable to exclude the possibility that there are sections
of A24 in dimensions 9 to 15 with smaller determinant (and hence higher
density) than either the An or Kn sequences.
Mordell's inequality is also a consequence of Theorem 4. This states that
if On denotes the highest possible center density of any lattice in Rn, then

on-l ~ 1.. 0 (n -2)/n

".. 2 n

([Cas2, Chap. X, Th. IV], [Mor4], lOppl)).

Proof Let 0 (L) and !L (L) denote the center density and minimal norm of
L, respectively. Let Ln be a densest lattice, with det Ln = 1. Then
detL~ = 1, so o(L~) ~ O(Ln), !L(L~) ~ !L(Ln). Then !L(L~) =
determinant of densest I-dimensional section of L~ = determinant of
densest (n - I)-dimensional section of Ln ~ !L (Ln), and now use (2).
Equality holds in (19) for n = 1, 2, 4, 8. If, as we suspect, K 12 , A 16 ,
A24 and P 4Sp are the densest lattices in these dimensions, equality also
holds for n = 12, 16, 24 and 48. Some sections of P 48p are described in
the following corollary.
Corollary 8. In dimension 48 - n the lattice P 48p has a section of density
o= (348-2n /248-n D) 1/2 where D is given by the following table:

n=O 1 2 3 4 5 6 7 8 9 10
D = 1 2 3 4 4 4 4 4 32
9
28
9
24
9

Moreover in dimensions 43 and above these lattices have the smallest

possible determinant of any section of any even unimodular 48-
dimensional lattice with minimal norm 6.
168 Cha p te r 6

Proof. It follows from Table 5.3 that coordinates for P 4Sp can be chosen
so that the minimal vectors include all the vectors
(20)

and the particular vector

(2J)

It therefore contains scaled copies of Dn (all the v ± i. ± j supported in the

first n coordinates), An (the part of the above Dn + I with zero coordinate
sum), and the lattices we shall call Sn (generated by v· together with the
above Dn -I). On the scale at which P 4Sp is unimodular, its sections
orthogonal to the lattices

have the same densities as those lattices by Corollary 5; the first assertion
now follows by rescaling . The second assertion of the corollary follows
from the optimality of Ao . ... . Ds (Theorem 2).

3. Properties of Ao to As
The lattices Ao to As (which were identified in Theorem 2) may be
specified as the sections of the Leech lattice shown in Fig. 6.2. For
instance the diagram for A2 indicates that A2 may be defined as the set of
all points of the Leech lattice in which the first three coordinates add to
zero and the rest are equal to zero. This lattice is spanned by (4, -4, 0 22 )
and (4, 0, - 4, 021 ), and clearly is a scaled version of A 2.
For use in the next section we remark that, for any r, Ar/2Ar is an
elementary abelian group of order 2r. The usual coordinates for Dn and
En are more convenient for describing the congruence classes of Ar/2Ar.
First, the nonzero classes of D3/2D3 have the structure of a projective
plane of order 2, as shown in Fig. 6.4.

101

110 011
Figure 6.4. Representatives for classes of D) /2D) . Three classes on a
line add to zero.
Laminated Lattices 169

Representatives for the sixteen classes of D4/2D4 may be taken to be

0000 2000 1111 1111
- -
0011 0011 1100 1100
-
0101 0101 1010 1010
-
0110 0110 1001 1001

where the bars indicate minus signs, and there is an automorphism of D4

rotating the last three columns.
The classes of Ds/2Ds and typical representatives are:
(i) 1 class (00000)
(ii) 20 classes (± 12, 03), (minimal vectors)
(iii) 1 class (±2,04)
(iv) 5 classes (± 14, O)even
(v) 5 classes (± 14 , O)odd.

The elements of E 6 are described by elements of E 8 in which the last three

coordinates are equal. The classes and representatives for E6/2E6 are:
1 class (08)
16 classes «± 1/2) S; (± 1/2) 3), (minimal vectors)
20 classes (± 12, 03 ; 03), (minimal vectors)
1 class (±2, 04 ; 0 3)
16 classes (P/2, (± 112)4; (± 112) 3)
10 classes (± 14, 0; 0 3).

The last 27 classes are each represented by ten norm 4 vectors forming a
5-dimensional coordinate frame (in these coordinates the minimal vectors
have norm 2). Similarly the classes of E 7 /2E 7 are:

1 class represented by (08)

63 classes represented by a pair of minimal (norm 2) vectors
63 classes represented by 12 norm 4 vectors forming a 6-dimensional
coordinate frame
1 class represented by 56 vectors of norm 6, namely (± 16 ; 02)
with an odd number of minuses, and ± (± 2 1, OS; 12).

Finally E 8 /2E 8 contains:

1 class represented by (08)
120 classes represented by a pair of minimal vectors
135 classes represented by 16 vectors of twice the minimal norm
forming an 8-dimensional coordinate frame.

Each deep hole in E 8 is an element of V2 E 8 that is congruent (modulo E 8)

to 16 vectors which are halves of one of these coordinate frames (see
 170 Chapter 6

Chap. 21). In MOG notation Gn which the minimal norm of Eg is 32),

these 135 types of deep holes are the following:
1 frame f 0 consisting of 16 vectors of the form

o 0
±4 0
o 0
o 0

35 positive tetrad frames fatcdlefgh consisting of all vectors like

±2 0
±2 0
±2 0
o ±2

where the support is a, b, c, d or e,f,g, h and the product of the nonzero

entries is positive,
35 negative tetrad frames f.;bcdlefgh defined similarly, and
64 odd frames consisting of vectors like

±1 ±1
±1 ±1
+3 ±1
±1 ±1

In the following section these 135 types will be referred to as the frames in
Eg (or Ag).

4. Dimensions 9 to 16
In this section we establish the assertions of Theorem 1 for 9 ~ n ~ 16.
Let A g + r (I ~ r ~ 8) be any laminated lattice in R g+r . By definition
Ag+ r contains Ag';;; E g. From Theorem 3 there is a gluing map
8: Ar --+ RAg such that Ag+ r consists of the vectors
u + vB + cv,
If v E Ar has the minimal nonzero norm, 4, then from (6), N(v B) ~ 2.
But the biggest hole in Ag has norm 2, so we conclude that minimal
vectors in Ar must be glued to deep holes in Ag.
We saw in the previous section that the deep holes in Ag are vectors in
lf2 Ag; and the value of vB is only important modulo Ag. Also the minimal
vectors of Ar generate An so 2Ar is mapped into Ag. Therefore 8 induces
a well-defined map

(22)
Laminated Lattices 171

The classes of Ar/2Ar that are represented by minimal vectors of Ar must

be mapped by e to frames in As (see the previous section).
Proposition 9. For 1 ~ r ~ 8, any class of Ar/2Ar not represented by a
minimal vector of Ar is mapped by e either to 0 or to a frame (but not to
a class represented by half a minimal vector of As).

Proof For r ~ 2 the hypothesis is vacuous. For r ~ 3, r ¢ 7, we reduce

to the case r = 3 by observing that any class of Ar/2Ar is contained in a
subgroup A3/2A3.
The case r = 3. Suppose the nonzero classes of Di2D 3 , shown in Fig. 6.4,
are mapped by e into the elements A, ... ,G of V2As/As shown in
Fig. 6.5, where A , ... , F are frames. Without loss of generality A is the
frame f o. B cannot be an odd frame, represented say by (-3, 17), for then
C == B + A would be represented by (1S), which is not a deep hole. Thus
B, and similarly C, D, E, are tetrad frames, and therefore G is
represented by a vector with even coordinates. So if G is represented by
half a minimal vector we can assume it is (22, 06). But then F == A - G
is represented by (2, -2, 06), which is not a deep hole.
The case r = 7. The previous argument must be modified because there is
one class of A7/2A7 which is represented by norm 6 vectors. However we
may choose a basis VI, ••• , V7 for A7/2A7 such that all the Vi and Vi ± Vj
are represented by vectors of norm 2 or 4. By the previous argument no Vie
or (Vi ± Vj)e may be represented by half a minimal vector. There is a
quadratic form Q defined on V2 As/As which takes the value 0 on the 0
class and on the frames, and the value 1 on the other 120 classes. Then
the quadratic form eo Q maps the Vi and Vi ± Vj to zero and so is
identically zero on A7/2A7. Thus the class of A7/2A7 represented by norm
6 vectors is also mapped to 0 or to a frame bye.

c E

Figure 6.5. Image of Fig. 6.4 under e.

172 Chapter 6

Proposition 10. If VI, •.• ,VI are minimal vectors of Ar that are
independent modulo (ker e, 2Ar) then we may assume that v? , ... , VIS

[]O []2 []2

are represented by an initial segment of

0 0o 20 22 00
[
0] 0' 20'00' 22
(23)
00 20 00 00

and so t :.;; 4. Thus the possible gluing maps e are completely

determined by t and ker e.

Proof Without loss of generality VI is mapped to the framefo. As in the

proof of Prop. 9, no minimal vector can now be mapped to an odd frame,
so without loss of generality vr
is represented by the second element of
(23), and so on. This completes the proof.
We can now determine all the possible laminated lattices A9 to A 16 •
We choose the kernel of e in A r /2Ar to have dimension k ~ r -4, and so
that no class in the kernel is represented by a minimal vector of A r . Let
VI, ••• , Vr be a basis for A r /2Ar such that VI, ••• , Vk is a basis for ker e
and Vk+I, ..• , Vr are represented by minimal vectors. Then we map the
t = r - k vectors Vk+l, . . . , Vr onto an initial segment of (23).
(A 9 ). Here r = t = 1, k = 0 and e maps VI to the framefo. Thus there
is a unique A9 , which may be described in MOG coordinates as being
spanned by

* * o0 o0
As: * * o0 o0
* * o0 o0
* * o0 o0
and
4 0 4 0 o0
vf + CVI o0 o0 o0
:
o0 o0 o0
o0 o0 o0
This leads to the definition of A9 given in Fig. 6.2. Also N(cvI) = 2 = 'Irs,
and >-9 = 'lrs>-s = 512, as stated in Table 6.1.
(A IO ). Here r = t = 2, k = 0 and e maps VI and V2 to the first two
elements of (23). Thus there is a unique A 10 which is spanned by A9 and
2 0 2 0 o0
2 0 2 0 o0
V = v~ + CV2 :
2 0 2 0 o0
2 0 2 0 o0
Laminated Lattices 173

(see Fig. 6.2 ). The projection of v onto RA9 is

2 0 2 0 o0
2 0 o0 o0
2 0 o0 o0
2 0 00 o0
which is therefore a typical deep hole in A 9, of norm h§ - 5/2, and the
projection onto (RA 9)..L is

o0 o0 o0
o0 2 0 o0
o0 2 0 o0
o0 2 0 o0
which has norm 11"9 = 3/2. Also from either (4) or (9) we have ;\ 10 - 768.
(A II). There are two possibilities, depending on whether the central class
in Fig. 6.4 is or is not in the kernel of e: (a) t = 2, k - I, ker e
= < (200) >, leading to A!'rx (see Fig. 6.2); (b) t -= 3, k = 0, e maps
v" v2, v3 to the first three elements of (23), leading to Arlin. We shall see
at the end of this section that these two lattices are distinct, as are the
A\2's and Al3's constructed below.
(A 12). The kernel of e is a subgroup of the first row of the table of classes
of D4/2D4 (see §3), and there is a unique subgroup of each of the orders I,
2 and 4. Thus there are (at most) three possibilities for Al2 : Ar1ax (when
k = 2), shown in Fig. 6.2, A r1id (when k = 1), and A r1in (when k = 0).
(A l3 ). The kernel must be a subgroup of classes m,
(iii), (iv), (v) of
Ds/2Ds, and have dimension ~ 1. However two classes from (iv) or (v)
with distinct support have a difference of type (ii), and cannot both be in
the kernel. Thus the kernel contains at most three nonzero classes. There
are (at most) three possibilities: (a) k = 2, ker e - {(OOOOO),
(20000), (11110), (-11110)}, leading to Al'3ax as shown in Fig. 6.2; (b)
k = I, ker e = < (20000) >, giving Al'3id ; (c) k = I, ker e = < (11110) >,
giving Al'3in .
(A I4 ). Similar reasoning shows that there is a unique A I4 , with ker e=
< (20000; 000), (1111 0; 000) >.
(A Is ). There is a unique A IS , with ker e= «200000;00), (111100;00),
(110011;00) >.

(A I6 ). There is a unique A I6 , with ker e = «20000000), (11110000),

(I 1001100), (10101010) >. The covering radius of AI6 will be determined
in the following section.

To see that these lattices are distinct we compute their kissing numbers
T. The minimal vectors fall into three classes: those in As (240 in
number), those gluing As to cAr (16T(A r ) of them), and those in cAr (the
174 Chapter 6

number of representations of elements of ker e by vectors of norm 4). For

example
r(Al1ax ) = 240 + 16·40 + (10+8+8) = 906. (24)

The results appear in Table 6.3, and the inclusions between lattices of
adjacent dimensions are displayed in Fig. 6.1. When n ~ 15 there are
very few n -dimensional sections of An+l that contain As, so all such
inclusions are easily found.

Table 6.3. Kissing numbers for laminated lattices.

Ao Al A2 A3 A4 As A6 A7
0 2 6 12 24 40 72 126

A8 A9 AIO A min A max AffilO Amid

12 Amax
11 II 12 12
240 272 336 432 438 624 632 648

A min Amid A max

13 A14 AI5 AI6 AJ7 AI8
13 13
888 890 906 1422 2340 4320 5346 7398

A 19 A 20 A21 A22 A 23 A24 A 2s

10668 17400 27720 49896 93150 196560 196610-196656

5. The deep holes in A 16

Having proved that A 16 is unique, our next step is to find the deep holes in
it, which we do by Simon Norton's method (see Chap. 22). First we
enumerate the congruence classes of V2 A 16/A 16.
Proposition 11. The 2 16 classes ofV2A 16 /A 16 consist of
(i) the zero class,
(ii) 2160 classes each represented by a pair of vectors ± V2 v, where
v E A 16 has norm 4,
(iii) 30720 classes each represented by a pair of vectors ± V2 v, where
v E A 16 has norm 6,
(iv) 135 classes each represented by 32 vectors ± V2 VI, . . . , ± V2 V16,
where the Vi are mutually orthogonal norm 8 vectors (this accounts for
4320 of the norm 8 vectors),
(v) 32400 classes each represented by 16 vectors ± 1/2 VI, . . . , ± 1/2 vs,
where the Vi are norm 8 vectors (this accounts for the remaining 518400
norm 8 vectors - see Table 4.12 of Chap. 4), and finally
(vi) 120 classes each represented by the halves of a certain set of 512
norm 12 vectors in A 16. These representatives have the property that they
Laminated Lattices 175

can be extended to minimal vectors of the Leech lattice by supplementing

them with (scalar multiples oj) minimal vectors of As. Typical
representatives are shown in Fig. 6.6, where the supplementing vectors are
enclosed in broken lines. In Fig. 6.6a the 512 representatives are formed
from the symmetric difference of any row and any column, with an even
number of minus signs, while in Fig. 6.6b the - 3 may be in any of the 16
positions and there are 25 possible sign combinations.
The proof of this proposition is straightforward and is omitted.

0
2
2
0
2 2
0 0
2
0
-2:
01
-3 1I
1I
,
1 I
2 0 0 0 0 01 1I
1 I
2 0 0 0 0 01 1 __
1 J1
...J

(8) (I)

Figure 6.6. Typical deep holes in A 16. The broken line encloses (a scalar
multiple of) the associated minimal vector of Ag •

Theorem 12. The covering radius of AI6 is .J3, and the deep holes in AI6
are those vectors of Y2 A 16 which are congruent modulo A 16 to one of the
classes of type (vi) in Prop. 11. Thus apart from permutations and sign
changes the deep holes are congruent modulo A 16 to either Fig. 6.6a or
Fig. 6.6b. There is a 1-1 correspondence between the congruence classes
of deep holes and the pairs ± u of minimal vectors of As.
Proof Let the covering radius of AI6 be.Jd. By compactness there is a
point x which is at distance .Jd from 0 and at distance ~ .Jd from all
other points of A 16 . The existence of the vectors shown in Fig. 6.6 proves
that d ~ 3. The covering radius of Y2 AI6 is Y2 .Jd, so we can write
x = u + x where u E Y2 AI6 and N(x) ~ d/4. From Prop. 11, u belongs
to one of the classes (j) to (vi). In fact u must be in class (vi), for if
x = lh v, + x where v, E A 16 has norm r then N (x ± Y2 v,) ~ d, so

d = N(x) ~ N(Y2 v,) + N(x) ~ r:d , (25)

which implies r ~ 9 and hence r ~ 12 by Prop. 11. Thus x = Y2 Vl2 +x

where Vl2 E A 16 , N(v12) = 12, NCx) ~ d/4.
Let us move the origin of coordinates to the point Y2 V12' From
Prop. 11 there are 512 points fjJ I, . . . , fjJ 512 in Y2 A 16 surrounding the new
origin, with N(fjJ;) = 3, which without loss of generality we may assume
consist of the point shown in Fig. 6.6a and the 511 other points obtained
from it by permutations and sign changes. We define the map
T : RI6 -+ RI6 by
512
T(y) = ~ (fjJ; .y) fjJ;. (26)
;-1
176 Chapter 6

Then
512
T(y) . y = ~ (cjJj . y)2 (27)
j-I

is a quadratic invariant of the orthogonal group which fixes the origin and
sends A 16 to itself. Since this is an irreducible group, T (y) . y must be a
scalar multiple of the quadratic form y . y, say
T(y) . y = k y . y . (28)

(This result is an instance of a theorem of Hadwiger, and expresses the

fact that the cjJj form a eutactic star [Cox20, §13.7], [Hadl].) By taking y
to have a single nonzero coordinate we find k = 96. Then with y = x we
obtain
512
~ (cjJj . x)2 = 96 N(x) ~ 24d . (29)
i-I

From (29) there is some cjJj, say cjJ, with

1/2
A... - ~ { 96 N(x) }
(30)
'I' x ~ 512

Let us write d = 3+0, 0 ~ 0, and N(x) = a 2, a ~ O. Since x is a deep

hole, N (cjJ-x) ~ 3+0. From this and (30) we obtain
a2 - a../3/2 - 0 ~ 0, and so either

a ~ 1 {I - [1 + 1~0 ]"2 } (31)

or

If (31) holds then from a ~ 0 we have a = 0, 0 = 0, x = 'Iz V12 and

d = 3, as required. We complete the proof of the theorem by showing that
(32) leads to a contradiction. From (32) and a 2 ~ (3 + 0)/4 it follows
that () = 0, d = 3, and a = N (x) = ../3/2. Then, for all i, IcjJi . xl ~ 3/8
x
and, from (29), cjJj . = ± 3/8. It is now straightforward to use the
explicit coordinates for the cjJj to show that this is impossible.

6. Dimensions 17 to 24
We next establish the assertions of Theorem 1 for 17 ~ n ~ 24. If
A 16+, (I ~ r ~ 8) is any laminated lattice then, arguing as in §4, we see
that there is a gluing map
e : A,/2A, --+ V2 A16/A16 . (33)
Laminated Lattices 177

The images of the minimal vectors in Ar under e are congruence classes of

deep holes modulo A 16. From §S there is a monomorphism

associating pairs of minimal vectors in Ag with classes of deep holes in A 16.

Lemma 13. If a, {3, 'Yare congruence classes of Ag/2Ag represented by
±a,±b,±c. where a,b,c are minimal vectors of Ag. and a + (3 = 'Y.
then the labeling can be chosen so that a + b = c.

Proof Without loss of generality a and b can be chosen so that

a . b ~ O. Therefore N(a+b) ~ 2N(a), so N(a+b) = N(a) or 2N(a).
If the latter, then from §3 the minimal representatives of a + bare 16
vectors of norm 2N (a), contradicting the existence of c. Thus a + b is a
minimal vector which we denote by c. This completes the proof.
We observe that Ar (1 ~ r ~ 8) has a strong generating set. By this
we mean a set S of minimal vectors VI • . . . • Vr with the property that if
we take the closure of S under the operations of (i) adjoining - v if v E S,
and (ij) adjoining v + v' if v, v' E S and v + v' is a minimal vector, then
we obtain all the minimal vectors of A r .
Now consider the images fwd ..... [wrl of [vd ..... [vrl under e,
where VI • . . . • Vr is a strong generating set for An and the square brackets
denote "congruence class containing ... ", and let [xd be the image of [Wi 1
under cJ?-I. Whenever we adjoin a vector v + v' to S to reach another
minimal vector of An by Lemma 13 one of the corresponding vectors
± x ± x' in Ag is a minimal vector. Proceeding in this way, when we
have found the closure of the strong generating set in A" we have obtained
an equal number of minimal vectors in Ag. Since those vectors belong to
the Z-span of XI • . . . • x" the latter must be a copy of A r •
Thus we have proved that what Ar is glued to in A 16 by e is the cJ?-
image of a sublattice <XI • . . . • x r > of Ag which is a copy of A r • Since
Aut (Ag) is transitive on sublattices An we conclude that A 17 • • • . • A24
are unique (and in particular that A24 is the Leech lattice). Their
determinants are given by (9), and the subcovering radii (except that of
A 24 ) by (3) and (4). Finally, the covering radius of A24 is known from
Chap. 23 (;;;; [Con20]).

7. Dimensions 25 to 48
In this section we establish the remaining assertions of Theorem 1. First
we construct a sequence of lattices L 25 k L 26 k ... k L4g containing A 24 ,
and then prove that these are laminated lattices.
Let i denote the element of Aut (A 24 ) shown in Fig. 6.7, satisfying
i2= - 1. Then (I + ;)/2 is a similarity of R 24 which halves the norms of
vectors. The lattice L 24+s (I ~ s ~ 24) consists of the vectors
u + vB + cv, where u E A 24 , V E As, () = (I + ;)/2 and c = 1/.J2. The
determinant of L 24+s agrees with the value of A24+s given in Table 6.1.
 178 Chapter 6

Figure 6.7. The element i of Aut(A 24 ). In the first column the 1st and
3rd components are to be exchanged, and the 2nd and 4th, and then the
2nd and 3rd components are negated. Similarly for the other columns.

We claim that L 24+s is a A 24+s. For 1 ~ s ~ 8 this follows

immediately from Theorem 3, since all L 24+s contain A 24 .
For 9 ~ s ~ 16, we first consider any lattice M 32+r (1 ~ r ~ 8)
containing any A 32 . Thus A24 k A32 k M 32+r • Projecting everything
onto (RA 24)..L we obtain 0 k cAs k M s +r , or 0 k As k c- 1 MS+r ,
where c = 1/.J2 and MS+r = M 32+r /A 24 . We apply Theorem 3 to
c- 1 Ms+ro deducing det (c- 1 M s +r ) ~ ASAr . T r , det M 32+r =
det A24 . det MS+r ~ Ar . 2-2r . From Table 6.1 the lattices L 32+r meet
this bound and therefore are laminated lattices, since no lattice containing
A32 can have a smaller determinant.

For 17 ~ s ~ 24 a similar argument shows that any lattice M 40+r

containing any A40 has determinant det M 40+r ~ Ar . T S- 3r , and since
L 40 + r meets this bound it is a A40+r •

The A 2S 's. We claim that there are exactly twenty-three A2S'S, one for
each type of deep hole in A24 (cf. Chap. 23). From the definition of a
laminated lattice it follows that there is at most one type of A2S for each
type of deep hole in A 24 . The kissing number of a A 2S is 196560 + 2V,
where V is the number of vertices of the hole (see Table 16.1 in Chap. 16).
This already shows that there are at least nine distinct A2S'S, the highest
kissing number being 196560 + 96 for the A 2S constructed from the hole of
type Af4.
To show that all the 23 holes produce distinct A2S'S, it is enough to
prove that there is a unique A24 inside a A 2S . In fact it is not difficult to
show that in each A 2S the original A24 is characterized by the following
property: it is spanned by all minimal vectors of A2S that are orthogonal to
at least 93150 other minimal vectors. We remark that the automorphism
group of the A 2S corresponding to a deep hole of type D24 has order 4
(whereas IAut (An) I > 1000 for 4 ~ n ~ 24).

The A 26'S. Finally we sketch a heuristic argument suggesting that the

number of distinct A26'S is very large. Any A 26 is obtained by gluing a A2
to A 24 . Let a, b, c be minimal vectors of A2 with a+b+c = 0, and let us
just consider the special class of A26'S in which all of aD, b O, CO are deep
holes in A24 of type A f2' with a O+ bO+ cO = o.
Laminated Lattices 179

The subgroup of Aut (A 24) fixing an A [2 hole has order 52

(Table 16.1), so the number of triples a, {3, "I of A[2 holes (modulo A24) is
(g/52)3, where g = IAut(A 24)1. The sum a + {3 + "I is in (1/13) A24, a
lattice in which A24 is subgroup of index 13 24, so the sum will be zero for
about (g/52)3/ 13 24 triples, falling into at least g2/ (52 3'13 24) orbits under
Aut (A 24). Taking account of the fact that -a, -{3, -"I and any
permutation of a, {3, "I lead to the same A26 , we can expect about
g2/ (12' 52 3'13 24) > 75000 distinct A26'S of this restricted type.
Some calculations of Simon Norton concerning norm-doubling
endomorphisms of A24 suggest that the number of distinct A4S'S may also
be large, and we believe that there are many distinct An for all
n = 26, ... ,48.
Acknowledgements. We are grateful to Simon Norton for informing us of
these calculations and for determining the covering radius of A 16, and to
E. S. Barnes and John Leech for their comments.

Appendix: The best integral lattices known.

Table 6.4, which is taken from [Con36] and also incorporates results of
Plesken and Pohst [Ple61, [Poh21, shows the densest presently known
integral lattices with minimal norms /L = 2, 3 and 4.
For minimal norm /L = 2 the smallest possible determinant is known for
all n (see the second column of the table). It is equal to 1 in all
dimensions n ~ 14. The third column gives examples of lattices having
these determinants, taken from Chap. 16. The lattices are specified by
their components (see §3 of Chap. 4) with a + to indicate the presence of
additional glue vectors of norm greater than 2.
For minimal norms 3 and 4 the entries in general only give upper
bounds on the smallest possible determinant. An {/L} indicates an integral
laminated lattice of minimal norm /L, An {3}.L denotes the orthogonal lattice
to A n {3} in A23 {3} for 0 ~ n ~ 12, and Kn{3} = Kn+dKI'
The unique A23 {3} is the shorter Leech lattice 0 23 , This is the unique
23-dimensional unimodular lattice of minimal norm 3 (see Chaps. 16, 19),
and consists of those vectors of A24 that have even inner product with a
fixed minimal vector v E A24 , projected onto v.L. Its theta series is given
in Eq. (7) of Chap. 19.
We have now explained all the entries in Table 6.4 except 0 24 , which is
the odd Leech lattice. This is the unique 24-dimensional unimodular
lattice of minimal norm 3 (Chap. 17), and may be defined as follows. Let
hA24 be the lattice constructed in §3.4 of Chap. 5. Then
0 24 = hA24 u «Y2)24) + hA 24 , (35)

whereas the Leech lattice itself (§4.4 of Chap. 5) is

A24 = hA24 U (-1 Y2, (\12)23) + hA 24 .
0 24 was first found by O'Connor and Pall in 1944 [O'Con
180 Chapter 6

Table 6.4. Smallest determinant of any integral lattice presently known

with minimal norm Jl = 2, 3 or 4, and examples of typical lattices having
these determinants.

J..L=2 J..L = 3 J..L=4

n
Det Lattice Det Lattice Det Lattice
0 I Ao 1 Ao{3} 1 Ao
1 2 Al 3 Ad3} 4 Al
2 3 Az 8 A z{3} 12 Az

3 4 A3 16 A 3 {3} 32 A3
4 4 D4 32 A 4 {3} 64 A4
5 4 Ds 48 As {3} 128 As
6 3 E6 64 A6{3} 192 A6
7 2 E7 64 A 7{3} 256 A7
8 1 E8 128 A8{3} 256 A8
9 2 E8 A l 192 A9{3} 512 A9
10 3 E8 A Z 243 K lO{3} 768 AlO
II 2 (DlOA 1)+ 256 A l d3} 972 Kll
12 1 Dtz 256 A IZ {3} 729 K12
13 2 (D6E 7)+ 243 K lO{3}-L 972 K13
14 1 Er+ 192 A9{3}-L 768 A14
15 1 Ais 128 A8{3}-L 512 A 1S
16 1 E§ 64 A7{3}-L 256 A16
17 1 (A llE6)+ 64 A6{3}-L 256 Al7
18 I A~+ 48 As{3}-L 192 A18
19 1 (ArDs)+ 32 A4{3}-L 128 A19
20 1 E8 D iz 16 A3{3}-L 64 Azo
21 1 Aj+ 8 A z{3}-L 32 AZI
22 1 E8E r+ 3 Ad3}-L 12 An

23 1 E8A is 1 OZ3 4 A23

24 1 Ei 1 OZ4 1 A24
 7
Further Connections Between
Codes and Lattices
N. J. A. Sloane

This chapter contains further investigations of the connections between

codes and sphere packings. Constructions A and B of Chapter 5 are
analyzed in greater detail and are generalized to complex lattices. We also
study self-dual codes and lattices and their weight enumerators and theta
series.

1. Introduction
We already saw in Chap. 5 that codes and sphere packings are related. In
this chapter we analyze Constructions A and B of Chap. 5 in greater detail
and generalize them to complex lattices. (More about Constructions A
and C will be found in the following chapter.) We also study the theta
series of the packings obtained from these constructions.
We shall see that the connections between codes and packings become
stronger when we consider self-dual linear codes and self-dual lattice
packings (§§3, 4), and strongest of all when we compare self-dual codes in
which the weight of every codeword is a multiple of 4 with self-dual
lattices in which the norm of every vector is even (§§6, 7). In the latter
case there are parallel theorems giving upper and lower bounds on the best
codes and lattices (Cor. 21 and Theorems 24, 25), and parallel theorems
characterizing the weight enumerator of the code and the theta series of
the lattice (Theorems 16, 17).
The weight enumerator of a self-dual code and the theta series of a
self-dual lattice are very strongly constrained, for the former must be
invariant under a fairly large group of transformations, and the latter must
be a modular form for a fairly large subgroup of SL 2 (R). As a
consequence these weight enumerators and theta series must lie in certain
simply described rings, sometimes free rings with a small number of
generators (see Theorems 6, 7,16,17,28-33).
182 Chapter 7

Extremal codes and lattices have the highest minimal distance or norm
permitted by the preceding theorems (the precise definition is given in §4).
We describe what is known about the classification of self-dual codes and
lattices, and in particular of extremal codes and lattices. In general only
finitely many extremal codes or lattices exist, and in certain cases they are
all known (Theorems 11-13, 20, 34, 36). The extremal weight
enumerators and theta series can (at least in principle) be determined
explicitly. This makes it possible to show that certain coefficients do not
vanish (Theorems 9, 20, 34), leading to upper bounds on the minimal
distance or norm (Corollaries 10, 21 35), and that other coefficients are
negative, showing that, for all n sufficiently large, extremal codes and
lattices do not exist. (Extremal lattices are unrelated to extreme forms
(§2.2 of Chap. 2).)
Sections 2-7 deal with binary codes and real lattices, and §§8-1O with
nonbinary codes and complex lattices. This chapter is based on [SI06]-
[SlolO1. Broue and Enguehard [Bro5]-[Br07] have also observed the many
parallels between codes and lattices (see also [MahI]-[Mah3], [TasI]).
Notation. We use the notation for codes established in §2 of Chap. 3, and
the theta functions defined in §4.1 of Chap. 4. In particular we recall from
Chap. 3 that Type I, II, III, IV codes are respectively self-dual codes over
F 2, F 2 , F3 , F4 in which the weight of every codeword is a multiple of 2, 4,
3, 2. As in §2.4 of Chap. 2, Type I or II lattices are real self-dual (i.e.
integral unimodular) lattices in which the norm of every vector is a
multiple of 1 or 2 respectively. If I J, 12,'" are algebraically independent
polynomials or power series then elf I,J 2 , .. J denotes the ring of
polynomials in I I, 12,'" with complex coefficients (a free ring).

2. Construction A
We repeat the construction from §2 of Chap. 5, changing it very slightly in
order to clarify the parallels between codes and packings.
Construction A. Let C be an (n,M,d) binary code. We assume the zero
codeword 0 is in C. A sphere-packing MC) in Rn is obtained by taking as
centers all x = (XI, . . . ,xn ) in Rn with
J2 x (mod 2) E C.

Thus the centers consist of all vectors which can be obtained from the
codewords of C by adding arbitrary even numbers to the components and
then dividing by J2. The origin is always a center.
Example 1. The lace-centered cubic lattice D 3. Let C =
{000,01I, 101, 1I0} with n = 3, d = 2. Some of the centers closest to the
origin are r l / 2 (± 1, ± 1,0), rl/2 (±2,0,0) ,.... This is the face-centered
cubic lattice D 3, the densest lattice sphere-packing in R3 (§6.3 of Chap. 4).
In general, if d < 4, the centers closest to the origin are the 2d Ad (0)
vectors of shape r l / 2 (±l) d on-d obtained from the codewords of weight
d. Such centers are at squared distance d /2 from the origin. If d > 4,
Further Connections Between Codes and Lattices 183

the centers closest to the origin are the 2n centers of type (± .Ji) lon-I, at
squared distance 2. Finally, if d = 4, both sets of centers are at the same
distance from the origin. Therefore the radius of the spheres should be
taken to be
p = r3/2 d 1/ 2 if d ~ 4, or rl/2 if d ~ 4, (2)

and the number of spheres touching the sphere at the origin is

T = 2d Ad (O)if d < 4, 2n + 16A 4 (O)if d = 4, 2n if d > 4. (3)

«3), (4) of Chap. 5). The center density of A(C) «2) of Chap. 5) is
{j = M pn 2- n / 2 •

(These formulae differ superficially from those in Chap. 5 because of the

.J2 inEq. (J).)
Theorem 1. A(C) is a lattice packing if and only if C is a linear code.
Proof. Elementary.

Theorem 2. Suppose C is a linear code of dimension k. Then

(i) det A (C) ,. 2n -2k ,

(ii) A (C*) = A (C)*,

(iii) A (C) is integralif and only if C ~ C*,
(iv) A (C) is of Type I if and only if C is of Type I,
(v) A (C) is of Type II if and only if C is of Type II,
(vi) Aut (A (C» contains a subgroup 2n .Aut(C).
Proof. Without loss of generality assume C has generator matrix [J B].
Then

I [I B] (5)
.J2 0 21

is a generator
matrix for A(C). (i)-(v) now follow easily.
(vi) Aut(A (C» certainly contains all the coordinate permutations in
Aut(C), all sign changes Xj - -Xj, and possibly other symmetries.
Remarks. Compare Theorem I of Chap. 8. If the .J2 is omitted from (1),
Construction A always produces an integral lattice.

Theorem 3 ([Ber?], [Bro6]). Suppose C is linear, with weight enumerator

Wc(x,y). Then the theta series of A(C) is given by
0 A (C)(z) = WC (8 3 (2z),8 2 (2z», (6)

where 82 (z), 83 (z) are Jacobi theta functions defined in §4.1 of Chap. 4.
184 Chapter 7

Proof Consider a codeword u = (u 1, ... , Un) in C. The corresponding

centers in A (C) consist of the set

A(u) = {(Yl,'" ,Yn):Y, E ~ u, + -J2z, 1 ~ r ~ n}.

From Eq. (44) of Chap. 4 we have

El.J2z(z) = El z (2z) =03(2z), El 1 .J2 (Z) =02(2z) .
.J2 + 2Z

Therefore

ElA(C)(Z) = ~ ElA(u) (z) = Wc (03(2z), 02(2z)).

u EC

Example 2. The cubic lattice zn. The universe code Fq has d = 1,

W(x,y) = (x + y)n. This produces the lattice ~ zn, which has theta-
function 0J(V2 z)n, and establishes the identity «22) of Chap. 4)
03(2z) + 02(2z) = 03(V2Z) . (7)

Example 3. The square lattice Z2. The code CC 2 = too, II} has weight
enumerator
(8)

and A (CC 2) == Z2, giving the identity «25) of Chap. 4)

OJ(2z)2 + 02(2z)2 = OJ(Z)2 .

Example 4. The checkerboard lattice Dn. As already observed in §2.4 of

Chap. 4, if C is the [n, n -1,2] even-weight code we obtain the lattice Dn.
In the standard version of Dn the minimal norm is 2, so in this case we
omit the -J2 in (1). The weight enumerator of C is
Wc(x,y) = V2{(X+y)n+(x-y)n}, (\0)

so from (6) the theta series of Dn is (allowing for the new scale)
ElD/Z) = Wc (OJ (4z), 02(4z))

= V2 {(OJ(4z) + 02(4z))n + (03(4z) - 02(4z))n}

= liz {03(Z)n + 04(Z)n}, (12)

using (22) of Chap. 4. On this scale Theorem 2(i) reads

det A (C) = 22n -2k ,
Further Connections Between Codes and Lattices 185

so det Dn = 4 (cf. §7.1 of Chap. 4).

Example 5. The Gosset lattice Eg. Let.Yfs be the [8, 4, 4] Hamming
code and V;g its weight enumerator (§2.4.2 of Chap. 3). As observed in
§2.5 of Chap. 5, applying Construction A to .Yfs produces the lattice Eg.
From (2)-(4) and Theorem 2 we find that Eg has p = l/.J"i, T = 240,
o = r4, det = 1, and is a Type II unimodular lattice. The theta series of
Egis (from (6»
flEs (z) = 8 3 (2z)g + 14 8 3(2z)4 8 2(2z)4 + 82 (2z)g (14)
= 'h [{8 3(2z)2 + 82(2z)2}4 + {8 3(2z)2 - 8 2(2z)2}4 (15)

+ 16 8 3(2z)4 82(2z)4]
(16)

in agreement with Eq. (40) of Chap. 2. (To go from (15) to (16) we use
Eqs. (24), (25) of Chap. 4.) The lattice E7 can be obtained in the same
way from the Hamming code :Yt, (§2.5 of Chap. 5).
Example 6. Best's lO-dimensional nonlattice PlOc. Best's (10,40,4) code
leads to a nonlattice lO-dimensional packing P lOc with 0 - r7 . 5 (§2.6 of
Chap. 5). Using the weight distribution given in Chap. 5 we find that this
packing is distance invariant and has theta series
8 3 (2z)1O+ 228 3(2z)68 2(2z)4 + 128 3(2z)48 2(2z)6 + 58 3(2z)28 2(2z)g
= 1 + 372q2 + 768 q 3 + 5684q4 + 6144q 5 + ... , (17)

verifying that the kissing number is 372. The generalized Jacobi formula
for periodic packings given in [OdI6] (which replaces z by -l/z and
multiplies by an appropriate constant) transforms (17) into

sq + -5-
1+ 4 516 q 2 + -2048
5- q
5/2 + 1728 3 +
-5- q
(18)

Question. Is there a formal "dual" to Best's packing, a nonlattice packing

with average theta series equal to (18)?
We do not even know if the hexagonal close-packing (§6.5 of Chap. 4)
has a "dual" in this sense. If so, its average theta series would be, from
(73) of Chap. 4 (cf. [OdI6D
1 + l.q4/3 + 2 q 3/2 + 9q41/24 + 3q17/6 + 6q4 + 9 q l\3/24 + ... (19)
2

3. Self-dual (or Type n codes and lattices

Throughout this section C will denote a code of Type I and A a lattice of
Type I. We shall see that the weight enumerator Wc(x,y) and the theta
series fl A(T) are strongly constrained, and in similar ways.
186 Chapter 7

In fact from the MacWilliams identity for codes (Eq. (50) of Chap. 3)
and Jacobi's formula for lattices (Eq. (] 9) of Chap. 4) we deduce:
Theorem 4.
W e (V2(X+Y), V2(X-Y)) = We(x,y), (20)
We(x,-y) = We(x,y). (21)

Theorem 5.
0 A (-I/z) = (z/j)n/20 A (z), (22)

0 A (z +2) = 0 A(T). (23)

Let G be an arbitrary finite (multiplicative) group of m x m complex
matrices. Then a polynomial f (x) = f (x I, . . . ,xm ) is said to be
invariant under G if f (Bx tr ) = f (x) for all BEG. (see §4.2 of
Chap. 3). Theorem 4 implies that We(x,y) is invariant under the group
G I generated by the matrices

Jz [~ -~ ], [~ -~ ]. (24)

It is easy to see that G I is isomorphic to the dihedral group of order 16 (a

reflection group - see §2 of Chap. 4). There are now several ways of
proceeding; the one we like best uses invariant theory (cf. [Blil], [Fla21-
[Fla6], [Huf21, [MiI21, [Sta3]) and is described in some detail in [S1071
and [Mac6, Chap. 19]. (See also [Ber7], [Br061, [Glell, [Mac3], [Toll].)
Whichever method is followed, the final result is:
Theorem 6 (Gleason). If C is a binary self-dual code then
(25)

where >/;2 is given by (8) and

~8 = x 2y2(X 2_ y2)2. (26)

In words, We (x,y) can be written uniquely as a polynomial in >/;2 and ~8'

or equivalently as a polynomial in the weight enumerators of C{j2 and .1fs.
Generalizations are given later in this chapter and in [Ber7], [Hufl],
[Le07], [Mac3], [Mac4], [Mal2], [MaI41, [MaI5], [Que61.
On the other hand, let G be an arbitrary subgroup of finite index in
SL 2 (R), and let X be a character of G. A complex valued function fez) is
said to be a modular form of weight w for G with respect to X if
(i) f (z) is holomorphic for 1m (z) > 0,
(ij) f [ az + b ) = [cz + d ] W f (z)
cz+d x(u)

for every u = [; ~] E G, and

Further Connections Between Codes and Lattices 187

(iii) f (z) is holomorphic at every cusp of G.

(References on modular forms are [Cas21, [Gun I), [Har41, [Hec11,

[Hec21, [Kit31, [KnoI], [Lan91, [Lew1], [MahI]-[Mah3], [ModI], [Oggll,
[Pet41, [Rad 1], [Ran61, [Sch6], [Sch7], [Ser11, [ShiI], [Sie1 ]-[Sie3J.)
Since many different definitions of weight appear in the literature, it is
worth saying that in our notation the cusp form ~24 (z) which generates
the Ramanujan numbers is
00

~24(Z) = q2 II (l_q2m)24, where q = e"iz, (28)

m-l

(Eq. (37) of Chap. 4); this is a modular form of weight 12 for SL 2 (Z)
with respect to the character X = 1.
Theorem 5 implies that e A (z) is a modular form of weight Vzn for the
group generated by

U= [~ n: z-+z+2, (29)

S= [~ ~1) 1
z--+-- , (30)
z

with respect to the character X(U) = 1, X(S) = i. Let be the

complex vector space span'!,ed by all modular forms of weight Vzn with
""n
respect to X, and let"" = $ ""n. Then"" is a graded ring.

""n
n-O

Theorem 7 (Heeke). (a) dime = 1 + [n/8]. (b) If A is a self-dual

lattice then

where ~8(Z) is the cusp form

-&; 8 2(z)48 4(z)4

r
~8(Z) =

= r 88 2 [ z ;1
II
00

= q ((l_q2m-l) (l_q4m)J8

Proof (a) follows from [Ogg 1, Th. 4, p. 1-411, since the hypotheses imply
e A (z)E "" (2, V2 n, 1) in the notation of [Ogg 1, p. xiv 1. (b) Since the
Poincare series for"" is

(35)
188 Chapter 7

we expect to find a basis for AI consisting of two algebraically independent

modular forms, one in All and one in Als. We already know from
Eq. (17) of Chap. 4 that fh(z) E All' For the form in Als we can either
use 8 Eg(Z) from Eq. (I 6), or the simpler form
16 .:ls(z) = 03(Z)S - 8 Eg (z)
= {03(2z)2+02(2z)2}4 - {03(2z)s+ 140 3(2z)40 2(2z)4+0 2(2z)g}

r
from (9), (14),

= 40 3(2z) 20 2(2z)2{03(2z)2_ 02(2z)2F = 02(Z) 40 4(Z)4 = -h- O2 [ z; I

by (24), (25) of Chap. 4,

= 16q II {(1_q2m-l) (1_q4m)}S,

m-l

by (34) of Chap. 4. It remains to show that the Poincare series for

C[03(Z), .:ls(z)] is equal to (35). For this it is enough to prove that for
any dimension n = 8a + ", 0 ~ " ~ 7, all a + I products
03(Z)n-Sr .:ls(z)' (O~r ~a)

are linearly independent. But this is immediate, for the rth product begins
qr + ....
Remark. From the proof of Theorem 7, in particular Eq. (35), we see that
the theta series of an n -dimensional Type I lattice can be written as

8 A (z) = ~ a r 03(Z)n-Sr .:ls(z)',

r-O

if n = 8a + ", 0 ~ " ~ 7, where the ai are integers. Thus 8 A (z) is a

polynomial of a particular form (an isobaric polynomial) in 0 3 and .:ls, or
equivalently in the theta series of Z and E s.
Example 5 (cont.). It is well known (see e.g. [Ogg 1]) that the Eisenstein
series

E 4(z) = 1 + 240 ~ (J3(m)q2m (37)

m-l

belongs to Als. Since both series begin 1 + 240q2 + ... , it follows that
E 4 (z) is equal to the theta series of the lattice E s.
Theorem 8 [SloIO]. If we expand .:lS(T) = T-arqr, the coefficients ar
alternate in sign and are multiplicative in the sense that

Iar I . Ias I = ~ d 3 1ars /d 21. (38)

dl(r,s)
dodd

(Compare Th. 21 below.)

Further Connections Between Codes and Lattices 189

Classification. Type I codes have been classified for length n ~ 30

[PleIO], [PleI2], [PleI7], and Type I lattices for dimension n ~ 25, as we
discuss in §2.4 of Chap. 2, and Chaps. 16, 17.

4. Extremal Type I codes and lattices

It follows from Th. 6 that if n = 2j - 8a + 2v, where 0 ~ v ~ 3, then
a
Wc(x,y) = ~ a,y.,2 j - 4,H" (39)
,-0

for uniquely determined integers ao •...• aa. Suppose we begin by

choosing the aj so that the right-hand side of (39) is

(40)

contammg no powers of y between 0 and 2a + 2. We call (40) the

extremal weight enumerator w* (x ,y) of length n, and a code having this
weight enumerator (if there is one) an extremal code W of Type I.

If such a code exists its minimal distance is 4a + 4, unless A:a +4 - 0,

in which case d may be greater than 4a + 4. But Theorem 9 shows that
such "accidents" do not happen. Thus an extremal code (when such a
code exists) has the greatest minimal distance of any Type I code.
Similarly for lattices: let n = 8a + v, where 0 ~ v ~ 7, write e A (z) in
the form (36), and choose ao • ...• aa so that the right-hand side of (36)
becomes the extremal theta series in dimension n:

(41)

A lattice with this theta series is called an extremal lattice(!) of Type I.

Theorem 9 ([Mall], [Mal3], [Sie3]). In the extremal weight enumerator
(resp. extremal theta series) the leading coefficient A;a+2 (resp. A:+ 1 ) is
positive.
The proof (of Theorem 9 and also Theorems 19 and 34) gives explicit
formulae for the extremal weight enumerators and theta series - see
[Mall], [Mal3].
Corollary 10. The minimal distance of a self-dual binary code satisfies
d~2[n/8]+2. (42)

The minimal norm of a self-dual lattice satisfies

J.L ~ [n /8] +I . (43)

0) A code or lattice which is "extremal of Type I" might accidentally have

Type II!
190 Chapter 7

In an extremal code or lattice equality holds in (42), (43). Eq. (43)

implies an upper bound on the density (see for example Eq. (2) of
Chap. 6). For large values of n the JPL bound (see [McE4], and Eq. (2)
of Chap. 9) implies the stronger result that any code with dimension
k = nl2 has

!!... ~ 0.178 ... , (44)

n

and the Kabatiansky-Levenshtein bound (Eq. (45) of Chap. 1) implies

l!:.. ~ 0.102 .... (45)

n

In view of (44), (45) it is not surprising that we have:

Theorem 11 [Malll. The next coefficient A;a+4 (resp. A:+ 2 ) in the
extremal weight enumerator (resp. theta series) is negative for all
sufficiently large n. Therefore the corresponding extremal codes or
lattices do not exist.
In both cases the first negative coefficient appears when n = 32. E.g.
the extremal theta-function for n = 32 is
e*(z) = 1 + 4700160q5 - 8094720q6 + ....

Theorem 12 ([Mal3], [Ple10], [Ple17], [War2]). Extremal self-dual codes

of Type I exist if and only if n = 2, 4, 6, 8, 12, 14, 22 or 24.
Theorem 13 (see Chap. 19). Extremal self-dual lattices of Type I exist if
andonlyifn = 1-8,12,14,15,23 or 24.
The Nordstrom-Robinson code. The extremal weight enumerator of
length 16 is

although no linear code with this weight enumerator exists. But there is a
nonlinear code with this weight enumerator, the (16, 256, 6) Nordstrom-
Robinson code (see §2.12 of Chap. 3). Similarly the extremal theta series
in dimension 16 is
e* (7) = 1 + 7680q3 + 4320q4 + 276480q5 + 61440q6 + .... (47)
If a corresponding packing were to exist, it would set new records for the
density and kissing number in 16 dimensions. No such lattice exists
(Chap. 19), and we conjecture that there is also no such nonlattice
packing. Inspection of (47) suggests that such a packing might be a union
of A 16 and 15 of its translates. However, this is impossible, for we can
prove:

Theorem 14. Suppose P is a packing in R16 with minimal norm 3,

Further Connections Between Codes and Lattices 191

consisting of a union of t translates of A 16• Then t ~ 9, and 9 can be

attained.
The proof is omitted. Question. Are there packings analogous to the
Kerdock or Preparata nonlinear codes (§2.l2 of Chap. 3)? (The analogous
codes or lattices of Type II do not exist, as we shall see in §7.)

5. Construction B
We repeat the construction from §3 of Chap. 5, again with slight changes.
Construction B. Let C be an [n,k,d=8] binary linear code in which the
weight of every codeword is a multiple of 4. A sphere packing !£(C) is
obtained in Rn by taking as centers all x = (x I, . . . , x n) in Rn with
(j) J2 x (mod 2) E C, and (48)
n
(iO 41 J2 ~ Xi. (49)
i-I

(The construction may be applied to other codes, but this version covers
the most important cases.)
Theorem 15. !£(C) is an integral lattice with minimal norm, kissing
number, center density, determinant and theta series given by
IL = 4, (50)
T = 2n(n-1) + 128 As, (51)
0= 2k-I-n/2, (52)
det !£(C) = 2n+ 2- 2k , (53)
(Hz) = V2 Wc (03(2z), 02(2z» + V2 04(2z)n. (54)

Example 5 (cont.) The [8, 1, 8] repetition code produces the lattice Es

again. In this case we obtain the theta series directly in the form (16).
Example 7. The Barnes-Wall lattice A 16 . The [16, 5, 8] first-order
Reed-Muller code produces the lattice AI6 (§1O of Chap. 4, §3.4 of
Chap. 5), with IL = 4, T = 2·16·15 + 128·30 = 4320,0 = r 4 , det = 256,
and we obtain its theta series (Eq. (132) of Chap. 4) from Eq. (54).
Example 8. The [24, 12, 8] Golay code CC 24 produces the integral lattice
hA24 (§3.4 of Chap. 5), for which IL = 4, T = 98256,0 - r l , det - 4.

6. Type II codes and lattices

We continue the study of self-dual codes and lattices from §4. In this
section C will denote a Type II code and A a Type II lattice. It is here
that we find the strongest analogies between codes and lattices.
Equations (21) and (23) can now be replaced by
192 Chapter 7

We(x,iy) = We(x,y), (55)

8 A (z+I) = 8 A (z)' (56)

Therefore We(x,y) is invariant under the larger group G 2 generated by

~ [~_ ~] and [~~]. This is the complex reflection group 4[6]2 of

order 192 (see [She2]), and we obtain (cf. [SI07]):
Theorem 16 (Gleason). If C is a Type II code then
(57)

where 1/;s is the weight enumerator of .Yfs (§2.4.2 of Chap. 3) and

b4 = X 4 y 4(X 4 _ y 4)4. (58)

Equivalently, We (x,y) can be written as a polynomial in the weight

enumerators of.Yfs and C6'24.
Similarly 8 A (z) is a modular form of weight V2 n for the full modular
group SL 2 (Z) generated by S (Eq. (30» and T: z - z + 1, with respect
to the character x(T) = I, x(S) = i. Let Jln and JI be as in §3.
Theorem 17 (Heeke). If A is a Type II lattice then

(a) dimcJln=I+[2:] if 8In,=0 if 8~n.

(b) 8 A (z) E C[E 4 (z ), .:124 (z ) 1, (59)

where E 4 (z) is the theta series of the lattice E s, given by (I 4), (16) or
(37), and .:1 24 (Z) is given by (28) (see also (37)-(39) of Chap. 4).
Equivalently, 8 A (z) can be written as an isobaric polynomial in the theta
series of Es and A24 .

Proof (a) follows from [OggI, Th. 3, p. 1-231. (b) The Poincare series is

~
n-o
(dimcJln)Xn = (
I-X
s) I(
I-X
24)' (60)

and we expect to find a basis for JI consisting of two algebraically

independent modular forms, one in Jls and one in Jl24 . We already have
E 4 (z) E Jls from (16), and .:1 24 (Z) = r S{8 2 (z)8 3 (z)8 4 (z)}S (Eq. (38) of
Chap. 4) E Jl24 • The remainder of the proof follows that of Theorem 7.

Remark. The proof of Theorem 16 shows that the theta series of a

Type II lattice is (j) an isobaric polynomial in E4 and .:1 24 , and (ij) a
symmetric function of 8 2 (z)S, 8 3 (z)S and 8 4 (z)s.

Corollary 18. A Type II code or lattice exists if and only if n is a

multiple of 8.
Further Connections Between Codes and Lattices 193

Example 8 (cont.) The Leech lattice A Z4 ' The Leech lattice AZ4 is
obtained by taking the union of hAz4 and its translation by (-1 V2, (V2)Z3)
(§4.4 of Chap. 5). This doubles the density (without changing the minimal
norm), so AZ4 has f..L = 4, 0 = 1, det = 1, and therefore is a Type II lattice.
We can write down its theta series immediately from Theorem 16. Since
there are no vectors of norm 2, it is E 4(z)3 - 720 LlZ4(Z) (see Eqs. (138)-
(141) of Chap. 4).
Theorem 19 (Ramanujan [RamIl; see also [MorIl, [OggiU If we expand
Llz 4(z) = ~::::;", r(m)qZm, the function r(m), the Ramanujan function,
is multiplicative in the sense that
r(rh(s) = ~ d11r(rs/d Z). (61)
d I ("S)

We have already mentioned estimates for the magnitude of r(m) in

§2.5 of Chap. 2.
Comparing (57) and (59) we observe that C[1fs, b4] and
C[E 4(z), LlZ4(Z)] are isomorphic graded rings - both have Poincare series
(6). In fact an isomorphism between them is given by Eq. (6), sending
(62)

Unfortunately this does not tell us how to find a lattice with given theta
series, even when an analogous code is known. In fact (62) does not even
map the weight enumerator of the Golay code onto the theta series of the
Leech lattice. Broue and Enguehard [Br06] give an interesting discussion
of the isomorphism between these rings.
Classification. Type II codes have been classified for length n ~ 32
[Con21], [PleIO], [PleI2], and Type II lattices for dimension n ~ 24, as
we have discussed in §2.4 of Chap. 2 (see Chap. 16).

7. Extremal Type II codes and lattices

The results are quite similar to those in §4, so we omit the details. Let
n = 8j = 24a + 8v, 0 ~ v ~ 2. For codes, let us choose ao, ... ,aa so
that
a j-3,
~ a,1fs H4 = x" + A:a+4 x"-4a-4 y4a+4 + ... . (63)
,-0

This is called an extremal weight enumerator and a code having this

weight enumerator is an extremal code of Type II and length n. Similarly
for lattices: choose ao, ... , aa so that

i
,-0
a,E 4(z)j-3, LlZ4(Z)' = 1 + A;a+2 q2a+2 +

This is an extremal theta series and a corresponding lattice is an extremal

lattice of Type II in dimension n. The 48-dimensional extremal theta
series was given in Eq. (57) of Chap. 2 (see Example 9 below).
194 Chapter 7

Theorem 20 ([Mall], [Ma13], [Sie3]). In the extremal weight enumerator

(63), A:"+ 4 > 0 for all n, but A:a+8 < 0 for all sufficiently large n. In
the extremal theta series (64), A;a+2 > 0 for all n, but A;a+4 < 0 for
all sufficiently large n.
This coefficient first goes negative at around n = 3720 for codes and
41000 for lattices. For example there is no extremal code of length 3720,
since the extremal weight enumerator of that length is
W*(x,y) = X 3720 + A~24 x3096y624 + A~28 x3092y628 +

where A~24 = 1.16 .... 10 170, A~28 = -5.84 .... 10 170 [MaI3].

Corollary 21. The minimal distance of a Type II code satisfies

d ~ 4[n /24] +4. (65)

The minimal norm of a Type II lattice satisfies

!L ~ 2[n/24] + 2. (66)

In [Mall] it is shown that for any constant b

d (C) ~ ~ - b, d (A) ~ In2 - b , (67)

for all sufficiently large n. Note that (65), (66) are stronger than (44),
(45).

In an extremal code or lattice equality holds in (65) and (66).

Extremal Type II codes are known of length 8 (one code, .Yfs [PlelO]),
16 (two codes, for example .Yfs ED .Yfs [PlelO]), 24 (one code C€24 [De1l3],
[Mac6, Chap. 20], [Ple8], [PleI7]), 32 (five codes [Con2l], [Koc3]), 40
(at least nine codes [lorl], [Oze3], [Oze4], [Ton2]), and 48,56,64,80,88,
104, 136 (at least one each - see [Mac6, Fig. 19.2], [Mo03], [HuB]).
Most of the codes of length ~ 48 are quadratic residue or double circulant
codes. The first open case is n = 72 (see [Con22], [Huf4], [PleI3],
[PleI4], [PleI9], [SI03]). The extremal weight enumerators for n ~ 200
are given in [Ma131

Extremal Type II lattices are known in dimensions 8 (£8 [Wit4]),

16 (two lattices, £8 ED £8 and Dt, [Wit4]), 24 (one lattice, A24 [Con4] ==
Chap. 12, [Lee5], [Nie2]), 32 (BW32 [BarI8] and §8.2f of Chap. 8, and at
least one other lattice [Bayl], [Br07], [Che4], [Ozel], [Ven3], [Ven6]
(although A32 of the previous chapter has determinant 1 on the standard
scale, it is not integra!), 40 (M 40 [McK2] and §5 of Chap. 8, and at least
two others [Bayl], [Oze3], [Oze4]), 48 (at least two lattices, P48p and
P 4Sq , see Example 9 below, and also [Oze2], [Petl]), 56 (at least one
lattice [Hsi6aD, and 64 (at least one lattice Q64, see [Que5] and §2c of
Chap. 8). See also the Postscript to Chap. 5. The extremal theta series
for n = 48, 56, 64 and 72 (the first open case) are respectively
Further Connections Between Codes and Lattices 195

I + 52416000q6 + 39007332000q8 +
1 + 15590400q6 + 36957286800q8 + ... ,
1 + 2611200 q6 + 19524758400q8 + ... ,
I + 6218175600q8 + 15281788354560ql0 + (68)

Example 9. The lattices P48p and P 48q . In this section we summarize the
properties of these two very similar lattices. Both are even unimodular
(Type II) extremal lattices and were defined in §5.7 of Chap. 5. P 48q is
obtained by applying Construction B3 (§5.4 of Chap. 5) to the [48, 24, 15]
ternary quadratic residue code, and then doubling the density by adjoining
a translate by (-5/2,1 47 ). P 48p is obtained in the same way from the [48,
24, 15] Pless code. Since the two codes have the same c.w.e. (Table 3.2),
the vectors in P 48q and P 48p have the same shapes. The two lattices are
not equivalent, however, since they have different automorphism groups.
For P 48q we label the coordinate positions 00, 0, 1 ..... 46 and let Q
denote the set of nonzero quadratic residues modulo 47. Then P 48q is
spanned by the vectors
47 vectors, supported on a translate {{O} U Q} + i,
o~ i ~ 46,
c(-5,1 47 ), a single vector, and (69)
c(±6 2, 046 ), 2·48·47 vectors,
where c = l/m. For P 48p we label the coordinate posItIons by 00,
O•...• 22, 00', 0' ..... 22', and let Q denote the set of nonzero quadratic
residues modulo 23 and N the nonresidues. Then P 48p is spanned by:
c(2 12 , (-2)12,0 24 ),23 vectors, with 2's at i and {Q +;}' and -2's at 00'

and {N+;}', 0 ~ i ~ 22,

c(-5,1 47 ), a single vector, and
(70)
c(±6 2, 046), 2·48·47 vectors.
P 48q and P 48p have det = 1, minimal norm = 6, T = 52416000, their
minimal vectors are described in Table 5.3, p = .J372,
~ = 0.00000002318... (0 = (3/2) 24 = 16834.112 .. '), and the covering radii
R are not known, but R ~ 2, corresponding to putatively deep holes
c(2 12 ,036). We have Aut(P 48q ) = 22. L2 (47), and Aut (P 48p )
= 2·L 2 (23) x S3 [Tho7l. (It is because there is no suitable group
containing both L2 (23) and L2 (47) that we know these lattices are
inequivalent.) Theta series: see Eq. (68) and Eq. (57) of Chap. 2,
Table 7.1. For cross sections see Corollary 8 of Chap. 6.
It is natural to ask if there are any nonlinear extremal Type II codes or
nonlattice extremal Type II packings (analogous to the Nordstrom-
Robinson code). In fact such codes or lattices do not exist, as J.-
M. Goethals (personal communication) has remarked. For the code or
lattice would be distance invariant, and then the argument of [DeIl3] (or
196 Chapter 7

Table 7.1. Theta series of 48-dimensionallattices P48q, P 48q .

m N(m) Factors of N(m)

0
6 52416000 29 32 53 7' 13
8 39007332000 25 37 53 7 3 13
10 6609020221440 2 3 5.7'13'23'47
11 8

12 437824977408000 2 12 33 53 7'13'31'103'109

[Mac6, p. 646]) shows that it must be linear.

Extremal codes whose length is a multiple of 24 are especially
important in view of:
Theorem 22 (Assmus and Mattson [Ass3], [Mac6, Chap. 6]). If C is an

r
extremal Type II code and n is a multiple of 24 then the code words of
any nonzero weight form a 5-design (cf §3.1 of Chap. 3).
The parameters of these designs can be determined from [Mal3]. For

:':m:l: ::::~':::' :'. ::::a~ 0[~:~~2 ht 'Mm a ',]"'go with

There is a similar result for lattices.
Theorem 23 (Venkov [Ven7]). If A is an extremal Type II lattice and n
is a multiple of 24 then the vectors of any nonzero norm form a spherical
II-design (cf §3.2 of Chap. 3).
The number of minimal vectors in an extremal Type II lattice in
dimension n = 24a can be found from Eq. (7) of [Mall]; it is
• 65520 a
A 2a + 2 = - - - - - - ~
~ ( r+l ) 1T11 ( r+l ) (a+ j)
Pa-r ,
691 a+ 1 r-O

where p~a+j) is the coefficient of qm in the q-expansion of

II
00

(I - qS)-24(a+ j) .

s-l

The numbers A;a+2 for n = 24, 48, ... ,216 and their prime
factorizations are given in Table I of [S108l
Corollaries 10 and 20 give upper bounds on self-dual codes and lattices.
There are corresponding lower bounds, which state that for large n self-
dual codes meet the Gilbert-Varshamov bound of [Mac6, Chap. 17], and
self-dual lattices meet the Minkowski bound of Chap. 1, Eq. (29). The
following theorems are proved in [Mac6, Chap. 19], [Mac8J, [Mil7,
pp. 46-471, [Tholl We state the result only for Type II codes.
Further Connections Between Codes and Lattices 197

Theorem 24. Let {j (n) be the largest multiple of 4 such that

[:) + [;) + ... + [Hn;-4) < 2n / 2 - 2 + 1.

Then there exists a Type II code of length n and minimal distance

d ~ Hn). For large n there exist Type II codes with

!£.
n
~ Hi! (lh) = 0.110 ... ,

where

(72)

is the binary entropy function.

Theorem 25. Let m (n) denote the nearest integer to

! {~ r [ ~ + 1) rn (73)

Then there exists an n -dimensional Type I lattice with minimal norm

J.t ~ m (n). For large n there exist Type I lattices with

(74)

or equivalently
1
-10g2.:l ~ -1. (75)
n

If m (n) is defined as the nearest even integer to (73), the same assertions
hold for Type II lattices.
The proofs use generalizations of the Minkowski-Siegel mass formulae
for lattices and their analogs for codes (see Chap. 16).

8. Constructions A and B for complex lattices

We now consider nonbinary codes and complex lattices, using the notation
of §2.6 of Chap. 2. In §§8-10 J denotes either the Eisenstein integers 8 or
the Gaussian integers t'fJ; §4 of Chap. 8 discusses the case J = Z[~],
~ = (1 +;)/.Ji. The constructions may also be applied to other rings.
Let be a prime in J(= 8 or t'fJ) of norm 'll"7r = q (say), so that
'II"

J/'II"J==Fq. Then there is a map O':J-+J/'II"J-+Fq (a ring

homomorphism) with 0' (a) E Fq for all a E J. We define 0' on n -tuples
byO'(a) =O'(al, ... ,an ) = (0' (a I) , ... ,O'(an ».
Construction Ac. Let C be an [n,k] code of length n over Fq • Then the
J -lattice A (C) consists of the points '11"-1/2 0'-1 (C) in en.
198 Chapter 7

Stated another way, if we abuse notation and regard Fq as a subset of

J, then A(C) consists of the points 7r- l/2 (c+7rx) for all c E C and all
x E r.If C has generator matrix [h B] then A (C) has generator
matrix

G = _1_
~
[h0 (76)

and
(77)

The corresponding 2n-dimensional real lattice A (C) real (§2.6 of Chap. 2)

has determinant
(7S)

The minimal squared norm /.L of A (C) or A (C) real and the packing radius
p and kissing number T of A (C)real are determined by the codewords of
minimal Euclidean norm in C. The center density of A (C)real is
(79)

The theta series of A (C) (or equivalently of A (C) real) is determined by the
c.w.e. of C, and in some cases just by the Hamming weight enumerator.
Construction Be. A' (C) consists of the points 7r -1/2 (c + 7r x), for all
c E C and all x E r
such that ~Xi == 0 (mod 7r). On order for A'(C) to
be a lattice C must satisfy certain obvious conditions') Then
det A'(C) = qn/2-k+l, and A'(C)real has center density
(SO)

Example 10. The case J = Eisenstein integers, 7r = 2. Let J = tS, 7r = 2,

so that 7r7r = 4, tS/2tS;;;; F 4 • We partition tS into 2tS, I + 2tS, w + 2tS,
w+ 2tS, and identify F 4 with {O, I, w, w} c tS (Fig. 7.1). Construction Ae
produces an tS-lattice from a code Cover F4 .
Theorem 26 [S1091. If C is an [n,k,d] code then A(C) has minimal
norm /.L = min{2,V2d} and detA = 2n - 2k . Furthermore A (C)real has
p = V2.J;, () = 3- n / 2 4k p2n, and theta series

(S 1)

where Wc(x,y) is the Hamming weight enumerator of C and 1>o(z),

1> 1 (z) are given in (I 2), (I3) of Chap. 4. If C is a Type IV self-dual
code then A is a self-dual tS -lattice.
Example lOa. The Coxeter- Todd lattice K 12. For example, when C is
the [6,3,4] hexacode C(J6 (§2.S.2 of Chap. 3), A (C(J6) is a 6-dimensional
self-dual tS-lattice, for which A (C(J6) real ;;;; K12 (§9 of Chap. 4). The theta
series given in (131) of Chap. 4 follows from Theorem 26. (This is the
Further Connections Between Codes and Lattices 199

ow ow ow

---0
1

ow ow
Figure 7.1. The Eisenstein integers 8 are represented by small circles,
and 28 by double circles, showing how 8/28 == F4 = (a, 1,w,w).

"2-base" for K12 in the notation of [Con37]. The "4-base" is given in

(I28) of Chap. 4.)
Example 11. The case J = Eisenstein integers, 11" = H. Instead of 2 we
may use the prime () = H = w - w
in 8, obtaining 8/(}8;;;; F 3 • We
partitIOn 8 into () 8, 1 + () 8, -1 + () 8, and identify F3 with
{O,I,-I} c 8 (Fig. 7.2). Construction Ac produces an 8-lattice A(C)
from a code Cover F3.

o------@-----o
-I 0 -I 0

~
0 -I o -I

0 @r---O o------@-----o
-I 0 1 -I 0 1

~ o------@-----o 0---
0 -I 0 -I

o------@-----o o------@-----o
-I 0 -I 0

Figure 7.2. The double circles represent 1J8, where IJ = H, showing

how 8/1J8 == F3 = (O,I,-I).

Theorem 27 [SlolO]. IfC is an [n,k,d] code, A(C) has minimal norm

Il = min{.J3, d/.J3) and det A(C) = 3n / 2 - k . Then A (C) real has
200 Chapter 7

det = 3n - 2k , (82)
0= 2n 3k - 3n / 2 /n, (83)

r=3dAd ifd";;2, 6n+27A 3 ifd=3, 6n ifd~4, (84)

and theta series

(85)

Example Ila. The Gosset lattice E 6 . If C is the repetitIOn code

{(OOO), (I I 1), (222)}, A (C) real is E 6 (see (120) of Chap. 4).
Example lIb. Eg again. The [4,2,3] tetracode C(J4 (§2.5.1 of Chap. 3)
produces E g.
Example Ilc. K\2 again. By applying Construction Be to the repetition
code of length 6 we obtain K \2. (This is the "3-base" for K \2, in the
notation of [Con37].)
Example 12. The complex Leech lattice. The Leech lattice A24 may be
constructed as a complex 12-dimensional &'-Iattice from the [12, 6, 6]
ternary Golay code C(J12 (§2.8.5 of Chap. 3). We start by applying
Construction Be to C(J 12. The lattice .JO A' (C(J 12) consists of the points
12
c + (J x, where c E C(J12, X E &' and ~ Xi == °
(mod (J). The complex
Leech lattice consists of the union of this lattice and two translates. Thus
(after multiplying by (J) we define the complex Leech lattice to consist of
the vectors
0+ (Jc + 3x, 1 + (Jc + 3y, -1 + (Jc + 3z, (86)

where 0= (0 12), 1 = (112), -1 = «_1)12), c E C(J12> and x, y, z E &,12

satisfy
~ Xi == 0, ~ Yi == 1, ~ Zi == -I (mod (J).

It is easily checked that this is an &'-lattice. The minimal norm is 18, and
the 196560 minimal vectors are given in Table 7.2. In this table the

Table 7.2. Minimal vectors in the complex Leech lattice.

Coordinates Number
(J «(i,06) = (Jc+3x 264'3 5 = 64152
({3,a ll ) = !+(Jc+3y 2'3 6 '12'2 = 34992
or -1+(Jc+3z
«2a)2,a lO) = !+(Jc+3y 36 '66'2 = 96228
or -1+(Jc+3z
(Oa)2,0IO) = 3x 66'2'3 2 = 1188
Total = 196560
Further Connections Between Codes and Lattices 201

symbols a stand for (possibly distinct) numbers of norm 1. i.e. ± 1. ± "'.

± ",2. and {3 is a number of norm 7. i.e. one of ± (1 + 3",). ± (1 + 3"'2) •....
The centers described in the first row of the table are obtained by taking
one of the 264 codewords of weight 6 in ~12. say
e = (0.1.0.0.0.0.-1.0.1.-1.-1.1). and choosing x so that
Oe + 3x = (O.o",a.O.O.O.O.-O",b. O.O",c.-O",d.-o",e.o",f)

with a. b. e. . .. E {0.1.2} and ~Xi == 0 (mod 01/). There are three

choices for each of the first five nonzero components of x. after which x is
uniquely determined. and so there are 264· 35 such centers. A typical
center 1 + Oe + 3y in the second row is
(1 1 1 1 1 1 1 1 1)
+0 (0 0 0 0 0 -1 0 -1 -1 1)
+ 3 (0 0 0 0 0 0 ",2 -1)
'" '" '"
",2
= (1 ",2 ",2 2",-0
'" '" '"
The first vector may be 1 or -1. e may be any codeword of ~ 12. and then
there are 12·2 choices for y: the element of norm 7 may appear in any of
the 12 positions and can be chosen in two ways. Thus the number of
centers of this type is 2· 36. 12·2. The other centers may be counted in a
similar manner.
The corresponding real lattice. when multiplied by ..fi/3. is the Leech
lattice A24 - compare Eqs. (135). (136) of Chap. 4. Table 7.3 shows
some of the parallels between the two lattices. The automorphism group of
the complex Leech lattice will be discussed in Chap. 10. See also [Wil3].
[YoslJ.
It is natural to ask if there is an analogous construction of A24 as a 6-
dimensional quaternionic lattice. using the hexacode ~6 over F 4. The
MOG construction of A24• described in detail in Chap. 11. is essentially
this analog. although the construction is not quite what one would expect
just from a comparison of the real and complex constructions. A
construction using the Cayley numbers is described in §2 of Chap. 8.

Table 7.3. Analogies between the real and complex Leech lattices.
Real Leech lattice Complex Leech lattice
Binary Golay code ~24 Ternary Golay code ~12
2 O=H
{+I.-l} h.",.",2}
Steiner system S (5. 8. 24) Steiner system S (5.6. 12)
Mathieu group M 24 Mathieu group M 12
Automorphism group contains Automorphism group contains
diag«-1)C\ ...• (_1)c ll ) diag(",CI •...• ",C 12 )
for any e E ~24 for any e E ~12
202 Chapter 7

Example 13. The case J = Gaussian integers, 7r = 1 + i. Let J = rg,

7r = 1 + i, so 7r7r = 2, rg/7r rg;;;; F 2. Constructions Ae and Be produce
Gaussian lattices from binary codes. It is best to omit the factor 7r -1/2 for
these lattices. Then Construction Ac produces a lattice with theta series
W(03(2z)2, 02(2z)2), where W(x,y) is the weight enumerator of the code.
The determinant of the real lattice Areal corresponding to a Gaussian lattice
A is
det Areal = (det A)2. (87)

Example 13a. D4 again. Construction Ae applied to C(j2 = {DO, II}

produces a two-dimensional Gaussian lattice ~2 with generator and Gram
matrices

M =
[01 1 +1]i ' A =
-
M M lr =
[2
1+i
I-i]
2 . (88)

The 24 minimal vectors have the form (± I,± 1), (± I,±j), (±i,±j),
(0, ± 1 ± j). The corresponding real lattice is D 4; in this form the theta
series is (cf. (92) of Chap. 4)
8 D /z) = 02(2z)4 + 03(2z)4 = 1 + 24q2 + 24q4 + 96 q 6 + .... (89)

Example 13b. E 8 again. Construction Be applied to {DODO, Il produces III

a Gaussian version of E 8 , consisting of all vectors x = (Xb X2,X3,X4) E rg4
satisfying x (mod 1 + j) == 0000 or 1111, and ~ Xi == 0 (mod 2). The
corresponding real lattice is the Hamming code version of E 8 (§8,I of
Chap,4).

9. Self-dual nonbinary codes and complex lattices

There are analogous results to Theorems 6 and 7.
Theorem 28 (Gleason). IfC is a Type III code then Wc(x,y) is invariant
under a group G 3 of order 48, and belongs to C[f4, ~121. where f4 is the
weight enumerator of C(j4 (Eq. (61) of Chap. 3) and
(90)

Equivalently, We (x,y) can be written as a polynomial in the weight

enumerators of C(j4 and C(j12'
To describe the c,w,e. we first define some further polynomials, Let
a = x 3 + y3 + z3, P = 3xyz, (91)
b = x 3y3 + y3 z3 + z3 x 3, (92)
(36 = a 2 - 12b = x 6 + y6 + z6 - 1Q(x3y3+y3z3+z3x3), (93)
7r9 = (x 3 _ y 3)(y3_ z 3)(z3_ X 3), (94)
(3) (6) (3) (3)
a12=a(a3+8p3)=~ X12+4 ~ X 9 y 3+6 ~ X 6y 6+228 ~ X 6y 3z 3. (95)
Further Connections Between Codes and Lattices 203

Theorem 29 [MaI5]. If C is a Type III code which contains the all-ones

vector then the c.w.e. is invariant under a group G4 of order 2592, and
belongs to
R E9 fJ6 7r§ R, (96)

where
(97)

In other words the c.w.e. can be written uniquely as a polynomial in fJl,

a12 and 7r3, plus fJ6 7r§ times another such polynomial.
Theorem 30. If C is a Type IV code then Wc(x,y) is invariant under a
group G s of order 12 and belongs to e[~2" ~61, where ~2' = x 2 + 3y2 is
the weight enumerator of the code CC2' = { 00, 11,ww,ww} and
~6 = y2(x 2_ y2)2. (98)

Equivalently, Wc(x,y) can be written as a polynomial in the weight

enumerators of CC2' and CC6.
Theorem 31 [Mac4]. If C is a Type IV code which contains the all-ones
vector then the c.w.e. is invariant under a group G6 of order 1152, and
belongs to elf 2,f6,f s,f 12] where (using standard notation for symmetric
functions of w, x, y, z)

h = (2) + x 2 + y2 + z2,
= w2 (99)
f6 = (6) + 15(222) = w + ... + 15(w x y2+ . . . ),
6 2 2 (100)
fs = (8) + 14(44) + 168(2222), (101)
f12 = (12) + 22(66) + 330(6222) + 165(444) + 330(4422). (102)

Here h is the c.w.e. of CC~, f 6 is the c.w.e. of the version of the

hexacode defined by Eq. (64) of Chap. 3, fs is the c.w.e. of the F4 span of
~, and f 12 is obtained from the c.w.e. of a certain code of length 12 (see
[Mac4]). For proofs and generalizations see [Ber7], [Glel1, [Leo71,
[Mac3], [Mac4], [Mac6, Chap. 19], [MaI2], [MaI4], [MaI5], [Slo6],
[Slo7]. G 3 is generated by ~ U.31 ], [6 ~] and is the reflection
group 3[6]2. G 4 is generated by all permutations, diag{I, 1,w} and

1r:: [1lww,
1 ~
,,3 1 ww

has a center Z (G 4) of order 12, and G 4/Z (G 4) is the Hessian group of

order 216 [Cox28]. G s is generated by V2 D
!1]' [6 _~\] and is a
dihedral group. G 6 is generated by all 4 x 4 monomial matrices and the
matrix H4 of Chap. 4, §7.1, and is the reflection group [3, 4, 3].
204 Chapter 7

r r
Theorem 32 [S1091 If A is a self-dual 8-lattice then
8 A(z) E C[Q>o(z/2), ~6(Zn where

~6(Z) == 116 0/) [~ 0/) [ 3; (03)

= q II 0_qm)6 0_q3m)6 (104)

m-)

Equivalently, 8 A (z) can be written as an isobaric polynomial in the theta

series of 8 and of the complex (8 -lattice) version of K )2, If we write
~6(Z) =}; amqm, the coefficients am are multiplicative, and in fact satisfy

(105)

Theorem 33. If A is a self-dual tf}-lattice then 8 A(z) E C[03(z)2, ~8(Zn

Equivalently, 8 A (z) can be written as an isobaric polynomial in the theta
series of tf} and of the complex (tf} -lattice) version of E 8·
Proof From (67) of Chap. 2, det Areal == (det A)2 = 1, so Areal is a Type I
lattice of even dimension. A and Areal have the same theta series, and the
result follows from Theorem 7.
Classification. Type III codes have been classified for length n ~ 24
[Con24], [Le08], [MaI2], [PleI8], Type IV codes for length n ~ 16
[Con24], [Mac4], and self-dual codes over F5 for length n ~ 12 [Leo71.
Self-dual 8-lattices have been classified for dimension n ~ 12 [Fei2], and
self-dual tf}-lattices for n ~ 7 Uyal1 (it would now be possible to extend
the latter classification using the results of Chap. 16). See also [Hsi6a],
[Ple201
To illustrate how rapidly the numbers of these codes and lattices grow
as n increases, Table 7.4 shows Nt(n), the total number of distinct
Type IV codes of length n, as well as Ne(n), the number of inequivalent
codes and N j (n), the number of inequivalent indecomposable codes.
Table 7.4. The numbers of Type IV codes.

n Nt(n) Ne(n) Nj(n)

2 3 1 1
4 27 1 0
6 891 2 1
8 114939 3 1
10 58963707 5 2
12 120816635643 10 4
14 989850695823099 21 10
16 32436417451427131131 55 31
24 _4'1043 > 108
Further Connections Between Codes and Lattices 205

10. Extremal nonbinary codes and complex lattices

Extremal weight enumerators, codes, theta series and lattices can now be
defined from Theorems 28, 30 (just using the Hamming weight
enumerators), 32 and 33, exactly as in §§4 and 7. There are analogs of
Theorem 9 and Corollary 10.
Theorem 34. In an extremal Type III or IV weight enumerator, or an
extremal fff - or l§ -lattice theta series, the first coefficient that is not zero
by definition is always strictly positive, and for all sufficiently large n the
next coefficient is negative.

Corollary 35. The minimal distance of a Type III code satisfies

d~3[n/12]+3, (106)

and for a Type IV code

d ~ 2[n /6] + 2 . (07)

The minimal norm of a self-dual fff -lattice satisfies

,.,. ~ [n /6] + 1, (08)

and for a l§ -lattice

,.,. ~ [n /8] +1. (09)

An extremal code or lattice satisfies (06)-009) with equality.

Extremal Type III codes exist for n = 4 (~4)' 8 (~4E9 ~4)' 12 (~12)'
16, 20, 24, 32, 36, 40, 44, 48, 56, 60 and 64, do not exist for n - 72, 96,
120, 144, ... , and and other values are undecided (see [Beel1, [Dawl1,
[Mac6], [MaI2], [Ma13], [Ple18] and the Postscript to Chap. 5). Extremal
Type IV codes exist for n = 2 (~2')' 4 (~2'E9 ~D, 6 (~6)' 8, 10, 14,
16, 18, 20, 22, 28, 30, and do not exist for n - 12, 102, 108, 114, 120,
122, 126, 128, .... The other values of n are undecided [Con24], [Mac4].
Since there are [24, 12, 8] Type II and [24, 12, 9] Type III codes, it
would be nice to know whether an extremal [24, 12, 10] Type IV code
exists ([Con22], [Con23], [Mac4]).

Extremal fff-lattices exist for n = 1 to 5 (8"), 6 (K 12)' 8 to 11, but do

not exist for n = 7, 12 and sufficiently large n [Fei2].
Theorem 36. Extremal l§-lattices exist if and only if n = 1 to 4, 6, 7 or
12.
Proof This follows from Theorem 12.
Acknowledgements. We thank E. Bannai (who independently discovered
Theorem 23) for drawing our attention to Venkov's paper [Ven7], and J.-
M. Goethals and C. L. Mallows for their helpful comments.
 8
Algebraic Constructions for Lattices
J. H. Conway and N. J. A. Sloane

In this chapter we construct dense lattice packings in dimensions 32, 36,

40, 48, 64, 96, ... , 65536, ... using a variety of techniques. The main
constructions used are Constructions Ar (a more abstract version of
Construction A), D (which uses a nested family of codes) and E (a
powerful general construction which can be applied recursively). Other
methods construct dense lattices from ideals in algebraic number fields.
We also construct E s and the Leech lattice using icosians.

1. Introduction
This chapter makes use of a variety of constructions, generally more
technical than in previous chapters. One common theme is that they use
more complicated rings than Z, so that more knowledge of algebra is
required. The lattices constructed include Es (§2), the Leech lattice (§§2,
3, 5, 7.3, 7.5), Quebbemann's lattices Q32 and Q64 (§§3, 4), the lattices C 32
(§8.2h) and B 36 (§8.2d), McKay's lattice M 40 (§5), the lattices P 4Sq (§7.5)
and P 64c (§8.2e), the Barnes-Wall lattices BWn , n = 2m (§8.2f), the
lattices Bn , n = 2m , obtained from BCH codes (§8.2g), Craig's lattices
An(m) (§§6, 7.3), and the infinite trees of lattices 71 (A) obtained from
Construction E (§IQ). The latter give reasonably dense lattices in
dimensions up to at least 1010000 (see §10h, Table 8.7, and Tables 1.3, 1.4
of Chap. 1), although eventually their density is not very good.
The constructions used include two general versions of Construction A
(§§3, 4), a version of Construction C that always produces lattices
(Construction D, §8), and a powerful generalization of most of the previous
constructions (Construction E, §§9, 10) that may be applied recursively.
Section 7 gives a number of constructions using ideals in algebraic number
fields. One of the most interesting of these is the result, already mentioned
in §1.5 of Chap. 1, that towers of algebraic number fields of the type
exhibited by Golod and Shafarevich correspond to infinite sequences of
extremely dense lattices (§7.4).
Algebraic Constructions for Lattices 207

All the pac kings mentioned in this chapter are lattices. We shall
sometimes take the more general point of view mentioned in §2.2 of
Chap. 2, and regard a lattice A as a discrete subgroup of a real vector
space V in which norms and inner products are defined by means of a
symmetric bilinear form j: V x V - R. The norm of x is given by
j(x) = j(x,x), and we have
j(x,y) = V2(f(X+Y) - j(x) - j(y». (I)

j (x) is assumed to be a positive definite quadratic form.

We begin (in §2) by introducing certain quaternions called icosians,
which are equivalent to the E 8 lattice equipped with a multiplicative
structure. This leads to a simple construction of the Leech lattice as a
three-dimensional icosian lattice (§2.2), a direct analog of the Turyn
construction of the Golay code.

2. The icosians and the Leech lattice

2.1 The icosian group. The icosian group is a multiplicative group of
order 120 consisting of the quat ern ions
V2 (±2,0,0,0) A, V2(± I, ± I, ± I, ± 1) A, V2(O, ± I, ±u, ±T)A,

where (a,{3,'Y,o) means a + (3i + 'Yj + ok,

u = V2(I -,/5), T = V2(I +,/5) ,

and the superscript A means that all even permutations of the coordinates
are permitted. We use the particular names
w = V2(-I,I,I,n, iH = V2(0,I,u,T).

There is a homomorphism from this group to the alternating group A5 on

five letters {G, H,I,J, K}, defined by
i = 1/2 (0,2,0,0) (H,[)U,K),
j = Y2(O,O,2,O) --+ (H ,J)(K,I),
k = V2(0,0,0,2) (H,K)(I,J),
w = V2(-I,I,I,n (J,J,K),
iH = V2(0,I,u,T) (G,[)U,K),

in which the kernel is {± I}. Table 8.1 below gives much more information
about this homomorphism. Abstractly the icosian group is the perfect
double cover 2. A5 of A5, and is sometimes called the binary icosahedral
group.
The icosian ring ~ is the set of all finite sums ql + ... + qn, where
each qt is in the icosian group. Elements of the icosian ring are simply
called icosians ([Con161, [DulJ, [WiI9]).
The typical icosian q has the form q = a + {3 i + 'Y j + 0 k, where the
coordinates a, (3, 'Y, 0 belong to the golden field Q(T), and so have the form
208 Chapter 8

a + b J5 where a, b E Q. The conjugate icosian (cf. §2.6 of Chap. 2) is

q = a - {3i - "(j - ok, and qq = a Z + {3z + "(z + oz.
We shall work with vectors v = (q[,qz, ... ) whose entries are icosians.
The typical scalar multiple of v is AV = (Aq[,Aqz, .. J, with scalars (which
are also icosians) on the left. Correspondingly the congruence
q == r (mod A), where q, r, A are quaternions, means that q - r = AS
for some icosian s. Two vectors v and w = (r[,rz, ... ) have a quaternionic
inner product

We shall use two different norms for such vectors, the quaternionic norm
QN(v) = (v, v),

which is a number of the form a + bJ5, with a, b E Q, and the Euclidean

norm
EN(v)=a+b. (3)

(It is easy to show that EN(v) ~ 0.) The icosians of quaternionic norm 1
are the elements of the icosian group.
With respect to the quaternionic norm the icosians belong to a four-
dimensional space over Q(J5); with the Euclidean norm they lie in an
eight-dimensional space. In fact under the Euclidean norm the icosian ring
~ is isomorphic to an E8 lattice in this space. Table 8.1 (taken from
[WiI9]) displays the particular isomorphism we shall use.
Table 8.1 has 60 entries, one for each pair of elements ±q of the
icosian group. The typical entry in this table:
-i
W = WIG --+ (HJK)
-1 0 0
-1 0 0
+ + -2 2
+ + 0 0

should be read as follows. The top line gives name(s) for q (in this case
wi = W /G = V2 (-I - i + j + k» and indicates the corresponding even
permutation of {G,H,I,J,K}. The four quaternionic coordinates of 2q
appear in the first column, followed by two columns giving the E 8 vectors
representing 2q and 2u q. As usual - stands for -I and + for + 1. The
formulae given in Eq. (74) of Chap. 2 are helpful for manipulating these
quaternions.
The naming system for q = V2(a,{3;y,0) is as follows. The letters i, j,
k indicate that a = 0, W indicates a = -1, S indicates a = -u, t indicates
a = - 7 , and the subscripts indicate the signs of a, {3, "(, 0:
Algebraic Constructions for Lattices 209

Table 8.1. Icosians and corresponding elements of A5 and E 8.

I_identity i=jG~(HI)(JK) j-jG~(HJ) (Kl) k=kG-(HK)(IJ)

2 2 2 - + o 0 0 - + o 0 0 + o 0 0 - +
o 0 + - - + + o 0 0 +
o o + - o + - - + o - +
o + - - + o + -. +

w-woB-(IJK) wl-wG,-(HKJ) cqI""WGJ-(H1K) oI'=wG.~(HJI)

-1 - - - + -1 - - + - -1 - - + -1 + -
1 + + - + 1 + + - + -1 - - - + -1 - +
+ + . . :. . + -1 - - + - 1 + + - + -1 - +
1 + + - + -1 - - - + -1 - - + + + - +

W-wBG-UK}) iii::zwlG-{HJK) Wl-wJo-(HKl) w'=w'G~(H/J)

-1 2 -2 -1 - - 0 0 -1 - 0 0 -1 - 0 0
-1 - - 0 -1 - - 0 1 + + 0 1 + + -2 2
-1 0 1 + + -2 -1 - - 0 1 + + 0
-1 1++0 1++-2 -1 - 0

iR~(G/)(JK) iH~(GJ)(K/) kH~(GK)(IJ) i,-(HG)(JK) i,-(HG)(K/) k.~IHG)(/J)

o 0 0 - + o 0 0 - + o 0 0 - + o 0 0 0 o 000 000 0
2 - + , 2 0 (J 0 0 + + 1 2 0 0 -t-2 0 0 -q 0 0 0- 2
q 0 + + o + -q 0 0-2 1 2 0 0 --t-2002
, 2 (1 0 + + - + -t-2 -q 0 0-2 1 2 0 0

i,~(HI)(GK) J.~(HJ)(GI) k,~(HK)(GJ) i.~(HI)(JG) i,~(HJ)(KG) k,~ (NK) (16)

0+-0 0+-00 o+ - 0 0 o - + - + 0-+-+ 0-+
1 + + 0 0 t + + -2 0 -(I + - 0-2 1 + + - + -t - + + - + + +
+ - 0 -2 1 + + 0 f + + -2 0 (1 - + + + I+ + - + -t - - + +
r + + -2 -(1 + - 0-2 + + 0 0 - + + " - + + + 1 + + - +

wHI--(G1K) wHJ-(GKI) wH,,·-oIGIJ) wJK---<o(GHI) w",--(GHJ) w/J--(GHKI

.- \ 0 -\ 0 - \ 0 - \ 0- 2 + \ 0 - 2 + . \ 0 2 +
0+0 + f + + - 2 o . q 0 r 2
T + + -2 o + - 0 - + , 2 o + - a 0
- + r + + -2 0 o + - t 2 0 - - +

w'H-(GK}) wJH--+(GIK) w .. J---<o(GIH) w,,,---<o(GJH) wn--(GKH)

- I - + -1 - - + - -\-2 0 0 \ -2 0 0 - \ 2 0 0
o - + - + " +
o 0 q -2 0 o
+ + + + q + - - t - 2 o 0 q 0
-(1 + - + + + q 0 0 ., 0 2

IH1-(HGIJK) lHJ ...... (HGJKI) (HI( ..... (HGKIJ) 'a--(JGKHI) t",--(KG/HJ) t lJ -(1GJHK)
r - 2 0 0 ·2 0 0 r 2 0 0 -- r - 2 0 + + t - 2 0 + + 2 0 + +
o 0 o (J 0 0 0 o 0 0 + .. 1 -2 + a 0 0 + +
q 0 o o a 0 0 + + 0-2 +
2 o q 0 o - 1 0 -2 + q 0 + + o 0 +

t,H-(lGHKJ) IJH-(lGHIK) ("H----(KGHll) t n .---<o(1GKJH)

2 0 2 2 0 + + + + + +
0+-00-\- OO-a+ 0+ + ++ + a+
-{1 + - 0 - 2 0 + 0 0 1 a-+ 0+ + 1++-+
-I 0 0 -(] + - 0-2 0 + o 0 I + + + -a + - 0 + - +

su-(KGJHI) SKI-(lGKHJ) SlJ-(lGIHK) '~HI-(IGHJK) sHJ---<o(JGHKl) sH,,---<o(KGHlJ)

o 0 - - -q 0 0 - -0 0 0 . q + . q + q +
00 -+-r--20+ + I 0 + o - + + r + + I + + - +
1 0 - + 0 + -T -2 + + I + + - + - + + r + +
-T-2 + + \ 0 + 0 0 + t + + - + + + + - +

s,,~(JGKIH) s,,~(KGIHJ) sJI-(IGJKH) s'H-(HGIKJ) s'H~IHGJ/K) su-(HGKJI)

-(1 + - 0-2 -(1 + - 0-2 - (J + 0 -2 -q 0 0 0-2 -q 0 0 0-2 q 0 0 0-2
- + 0 r + +-2 -\ - 0 o 0 0 -,-2 0 - 1 -2 0 0
-1 - + 0 r + + -2 .. \ - 2 0 0 , - 2
t++-2 -I o - + 0 -t -2 -\ -2 o 0
210 Chapter 8

signs subscript signs subscript

0(+00) G 0+-- I
-+++ GH -0++ HI
HG -0-- IH
-+-- GI -0+- JK
--++ IG -0-+ KJ
0+++ H

For example WXY corresponds to the permutation (Z, T,U), where X, Y,

Z, T, U is an even permutation of G, H, I, J, K. The triple (ix,jx,k x )'
where X is any of G, H, I, J, K, forms a system of unit quaternions (with
i}=-I, ix}x =kx , etc.).

The 240 minimal vectors of this version of E 8 have Euclidean norm I,

and quaternionic norm either I or 1I2. They consist of the elements q and
II q, where q is any element of the icosian group.

Using the map in Table 8.1, a vector v = (q] • ...• qn) with n icosian
coordinates is identified with a vector having 8n rational coordinates.

2.2 The icosian and Turyn-type constructions for the Leech lattice. Let L
be the three-dimensional lattice over the icosians consisting of all vectors
(x,y,z), where x, y, z E ~ satisfy

x == y == z (mod h),

x + y + z == 0 (mod h) , (5)

and h is the quaternion

h = W -!- II = V2 (-.J5, I, 1,1) ,

with hh = 2. Then the Euclidean norm (Eq. (3» converts L into a copy
of the Leech lattice. This will be proved in §3, Example (c). Ot is also
established in [WillI, [WiI9J.)
Let G be the automorphism group of L (regarded as a three-
dimensional icosian lattice). G is a double cover 2J 2 of the Hall-lanko
group J 2 = HJ [ConI6, p.421, [WiI91. In fact it follows from the
definition of L that G contains a subgroup S (of order 27.3.3!) generated
by all permutations of x, y, z and by right multiplication of (x,y,z) by
the diagonal matrices {],i, i}, {i,}, k), {w, w, w}. The minimal Euclidean
norm of L is 4. Modulo scalar multiplication (on the left) by elements of
the icosian group, there are four orbits of vectors in L of quaternionic
norm 4, namely
(2,0,0), (O,h,h), (h, 1, 1), (I ,TW,IIW) , (6)

with 3, 24, 192 and 96 images, respectively, a total of 315. If r is one of

these 315 "root" vectors and vEL, then (v,r) E 2~ [WiI9, p. 1631, and
so the reflection in r,
Algebraic Constructions for Lattices 211

2(v,r)r
v--+v-
(r ,r) ,

preserves L (so is in G), and preserves quaternionic norms and inner

products. These 315 quaternionic reflections generate 2J 2.
The minimal vectors in this version of the Leech lattice are the
elements of L having quaternionic norm 4, 4u 2 or 2 + 2u 2• There are
120· 315 = 37800 vectors of shape qr (where q is in the icosian group and
r is a root) and quaternionic norm 4, an equal number of shape uqr and
norm 4u 2, and 120· 1008 = 120960 of shape qt and norm 2 + 2u 2, where
t is one of the images of (h,uwh,O), (l,l,uwh), (uw,uw,h) or
(l,uw,uw i +w k ) under S. (These have 48, 192, 192 and 576 images
respectively.) For further information about this lattice and its group see
[Wil9].
We remark that there is an easy and well-known construction of the
Leech lattice from E 8 , analogous to Turyn's construction of the Golay code
from the Hamming code given in §12 of Chap. 11. In the usual MOG

ITG.
notation, the vectors v for which

(I)

is in A24 form two E8 lattices in the same space, say EJO and EJ2}. We
can now define A24 to be the lattice generated by the vectors
(a,b,c), a +b +c = 0, a, b, c EEJo ,
(x,x,x), x EEJ2} .

The icosian construction of A24 is merely this Turyn-type construction

equipped with extra structure. (See also [Coh2], [CosU, [Lep2], [Tit6],
[Tit7J.)

3. A general setting for Construction A, and Quebbemann's

64-dimensional lattice
The following fairly general version of Construction A will be used in
several examples. (It does not completely replace the versions of Chaps. 5
and 7, however, since it applies to linear codes and real lattices.
Construction E of §9 is an even more general version.)
Let L be a lattice in Rn with an associated symmetric bilinear form f.
Let p be a prime number which does not divide det L, and set
i = L/pL.

We denote the image of tEL in L by t == t + pL. L has induced

~u~dratic and bilinear forms given by j (i) == f (t) mod p,
f(t,;n) = j(t,m) mod p. (Note that if p = 2, there is no analog of
Eq. (1) for 1.)
212 Chapter 8

Let C be a subspace of L. (C plays the role of the "code" in this

constructionJ
Construction A f. We define a new lattice A (C) in Rn by
A (C) = {eE L : e E C} ,

equipped with the bilinear form l' (e,m) = p-'f(e,m).

Clearly pL ~ A (C) ~ L. The dual lattice A (C)* is given by

A(C)* = {x ERn: ~f(x,A(e» ~z}. (8)

The dual of C in L, e*, is defined with respect to 1:

C* = {x EL :j(x,e)=o}.

Then it is immediate that

A (C)* = A (C*). (9)

As usual the dual of L is defined with respect to the original form f:

L* = (x ERn :f(x,L) ~Z}. (10)

By abuse of notation we can regard e* /e as a subgroup of

A (C*)/A (e) = A (C)* /MC)'

Theorem 1. If C ~ C* then
(A (e)* /A (C»/(C* /C) ;; L * /L.

This implies

det A (C) = det A . I C* /Ci. (12)

If L is unimodular and C = C*, A (C) is unimodular.

Proof j(c,c') = 0 for c, c' E C, so f(e,e') E pZ for e,e' E A(C).
Therefore A (C) ~ Mc)*. Similarly A (C) * ~ L *, so

We define a homomorphism
(J: A(C)*/A(C) -+ L*/L

by (J (v + A (C» = v + L. It is easy to check that (J is well-defined and

(using pi det L) that (J is onto. Also ker (J = (v + A (C) : v + pL E CO).
Since (A(C)/pL) n C* = C, ker (J ;; C* /C, which completes the proof.
Examples (a). Taking L = zn, p = 2 we obtain the linear version of
Construction A (Chap. 7, §2).
Algebraic Constructions for Lattices 213

(b). Orthogonal geometries. Let M ~ Rk be a lattice with a symmetric

bilinear form I, let p be a prime not dividing det M, and let M = M /pM.
Suppose M has the structure of an orthogonal geometry, in particular that
M = V$ V',

where V and V' are subspaces of M such that j(V} = j(V'} = 0, and
V = V', V' = V' * (with respect to j).
Let L = M n ~ R kn with the bilinear form
I(e,m} = I«e l • ...• en}, (mI •...• m n »= ~/(ei,m).

We construct a self-dual "code" C as follows. Let B be an arbitrary

subspace of yn and set
B' = V'n n B.l. = {v E (V'}n : j(v,B) =O},
C =B $ B'.

Then C = C· (with respect to j). From Theorem 1, A(C} (with the form
p-II) is integral and
(13)

(c) The Leech lattice. The Turyn-type construction of the Leech lattice
given in §2.2 follows by taking M '"" E 8, I to be usual Euclidean inner
product on E 8, and p = 2. Then M = M /2M is a 4-dimensional space
over F2 which may be written V $ V', where (in the icosian notation)
V = <h,ih,jh,wh >, V' = <h,ih,jh,wh>.

We apply Construction AI with n = 3, taking B to be the repetition code

consisting of all (x,x,x), x E V, and B' to consist of all (x,y,z) E (V'}3
with x + y + z = O. Then A (C) is the lattice L defined in Eqs. (4), (5).
It is now straightforward to check that this corresponds to an even
unimodular 24-dimensional lattice of minimal norm 4 (compare the proof
of Theorem 1 below), and so by Chap. 12 is the Leech lattice.
(d) The lattice Q64' Quebbemann [Que5] used Construction AI to obtain
two important lattices in 32 and 64 dimensions. We postpone
consideration of the 32-dimensional lattice to §4. The 64-dimensional
lattice is obtained bvaking M (in Example (b» to be E 8 , k = 8, P = 3
and n = 8. Then M = E8/3E8 can be decomposed as V $ V', where
V = <ehe2,e3,e4>, V' = </1./2./3./4> and the ei' Ii are given in
Table 8.2. The coordinates in Table 8.2 are labeled to show the
identification with the Hamming code version of E8 (§8.1 of Chap.4).
The cosets ei + 3E8 and Ii + 3E8 (i = 1..... 4) have minimal norm 6
(on the scale at which the minimal norm of E8 is 2). Each pair
{ehe2} ....• {!J'/4} generates a copy of the tetracode C(f4 (Chap. 3,
§2.5.1}. Aut (C(f4) contains elements of order 8, and so there are
automorphisms of E 8/3E 8 of order 8, such as
11" = {(l2 00 4} (3506) then negate 2,5}
 214 Chapter 8

(with the coordinates labeled as in Table 8.2L

Let B be the subspace of VS spanned by the vectors
(x,x,x,x,x,x,x,x), x E V and (y,1ry,1r 2y •...• 1r 7y), Y E V. From
(11), B' consists of the vectors (ZO,ZI • . . . • Z7) E (V') IS satisfying

Zo + ZI + ... + Z7 = 0, (14)
Zo + 1r- I ZI + ... + 1r-7Z7 = o. (15)

We take C = B e B', and let Q64 = A (C). Thus Q64 is spanned by the
vectors
0[,0 7), where [ runs through a basis for £s, (16)
(xS), x E {e"e2,f"h}, (17)

(y,1rY, ... • 1r 7y), Y E {e"e2,f"h}, (18)

(ZO,ZI, . . . • Z7)' Zi E {e3,e4,/J,f4}, satisfying (14), (15). (19)

Table 8.2. A basis for £8/3£8, an 8-dimensional vector space over the
field F3 (1 denotes - J).
1 2 4 00 3 5 6 0
V 1 1 0 0 0 0 0
el
-1
e2 0 1 1 1 0 0 0 0
e3 0 0 0 0 1 1 1 0
-
e4 0 0 0 0 0 1 1 I
V' 1 1 0 0 0 0 0
fl
-1 -
0 0
f2 1 1 1
-01 0 0
h 0 0 0 0 1
-1 0
-
f4 0 0 0 0 0 1 1 1

Theorem 2 [Que5t Q64 is an extremal even unimodular 64-dimensional

lattice with minimal norm 6. kissing number 2611200 and center density
0= 0/2)24.
Proof (13) implies det Q64 = 1, and the lattice is even by construction.
Suppose there is a vector v = (vo, ...• V7) E Q64 with 1/3f(v,v) ~ 4, i.e.
f(v, v) ~ 12. Let v = (vo • ...• V7) where v;. = x + p'y + z" x, Y E V,
z, E V'. Since f (Vi' Vj) is even, there are at least two coordinates i, j for
which Vi = Vj = 0 (otherwise f (v, v) ~ 14). Therefore 1ri y = 1r j y, so
y = 0, x = 0 and v;. = z,. Since z, E V', the norm of any element of
z, + 3£ S is divisible by 6. Therefore at most two of the z, are nonzero.
(14), (15) now imply that all z, = 0, and V E 3£l, Since 3£H has
minimal norm 18, we conclude that V = 0 and the minimal norm of Q64 is
6. The kissing number is then determined by Theorem 17 of Chap. 7.
Quebbemann's construction is slightly different from the one given here,
and it is likely that there are several inequivalent lattices with these
Algebraic Constructions for Lattices 215

parameters. The existence of at least one such lattice was established in

[Que3].

4. Lattices over Z[e'lri/4], and Quebbemann's 32-dimensional

lattice
Sections 8-10 of Chap. 7 describe lattices over the Eisenstein and Gaussian
integers and their connections with codes over the fields F 2, F 3 and F4.
Quebbemann [Que6] has developed a similar theory relating lattices ove
the ring of cyclotomic integers zB"1. where ~ = e"i/4 = (1 +j)/.J2, to code;
over F9• The results are not exactly parallel to those in Chap. 7, however,
and it seems best to treat this case separately. The most interesting
example is an 8-dimensional Z[~]-lattice which produces the 32-
dimensional lattice Q32' This is obtained from a complex analog of
Construction AI, replacing the symmetric bilinear form f by a Hermitian
form.
The lattice Z[~] itself. We begin by showing that Z[r] can be regarded as
a version of the two-dimensional Gaussian lattice !!)2 (Example 13a of
Chap. 7) or of the four-dimensional real lattice D4 • It is plausible that
Z[~] corresponds to a four-dimensional real lattice, since the corresponding
field Q(~) is an extension of the rationals of degree 4, with basis
1, i, .J2, i.J2. (20)

The cyclotomic number ~ = e"i/4 satisfies ~2 = i, ~ = -1, ~8 = 1, and

has minimal polynomial X4 + 1 (over Z). Elements of Z[~] may he
written either as
a =,8 + 'Y~, ,8, 'Y E Z[d,

or as

To identify Z[~] with [!)2 we regard Z[~] as a two-dimensional lattice

over z[d with basis 1, ~, and simply define a Hermitian inner product cf> on
Z[r] by specifying that the Gram matrix of the basis is

~ ~
[1 i 1 i ) . (23)

Thus if a =,8 + 'Y~, a' =,8' + 'Y'~, where,8, ,8', 'Y, 'Y' E Z[;],
cf> (a,a') = ,8F cf> (1, 1) + ,8i cf> (1,n + 'YF cf> (~, 1) + 'Yi cf> (~,~), (24)

where cf> (1, 1) = cf> (~,~) = 2, cf> (1,~) = cf> (~, 1) = 1 - i. We define norms
in Z[~] by N(a) = cf> (a, a).

Since (23) is the same as Eq. (88) of Chap. 7, Z[~] with this inner
product is [!)2' The minimal norm is 2, and the 24 elements of norm 2 are
216 Chapter 8

rv, rv(I - nand r(I - .J2) = rV(I - r+ r 3) for v = 0, I, . . . ,7 (forming

three orbits of size 8 under multiplication by r). Although not as simple as
the familiar representation of the minimal vectors of D4 (the permutations
of (± I, ± 1,0,0», this set of 24 points forms a pleasantly symmetric
configuration in the complex plane (Fig. 8.]) . To recover the familiar
coordinates we map
in z!r1 in z[iF in Z4
I to I I or I I 0 0
r oI I 0
r2 o0 I I
r3 -I -I 00 I

The inner product in Z4 is the real part of the inner product in Z!r1 or
z[iF. Multiplication by r in z!r1 corresponds to interchanging the two
coordinates of Z[i F and multiplying the second coordinate by i, and to the
operation "(0123) then negate 3" on Z4. The vectors of norm 4 in z[rl
are obtained by multiplying the vectors of norm 2 by + r 1 = J2. The r
theta series of z!r1 is given by Eq. (89) of Chap. 7.
We can write 4> in another way. The trace map Tr from QW to Q(i)
sends a + bi + cJ2 + diJ2 (with a, b, c, d EQ) to 2a + 2bi. For
a = (3 + -yr E Q(r), (3, -y E Q(i), we define
T(a) = Tr(a(] +rl/2» = Tr«{3 + -yr)(1 +rl/2»
= 2{3 + (I + i)-y, (25)

a map from Q(r) to Q(i). Then

4> (a,a') = T(a,d), N(a) = T(aa). (26)

- 1

Figure 8.1. A projection of the polytope {3,4,3l whose vertices are the 24
minimal vectors of D 4 , here represented as elements of Z[l"J.
Algebraic Constructions for Lattices 217

The field F9• By a slight abuse of notation we can also regard the 8-th
roots of unity r' as the nonzero elements of the field of order 9. To be
more precise we let F9 consist of the elements 0, 1, t, t 2 "" 1 - t,
t 3 - - I - t , t 4 =-I, t 5 =-t, ~=-I+t, t 7 -I+t, where
t 2 + t - 1 = 0, t 8 = 1 lMac6, p. 951. Now Zl;)/3Zl;) is identified with
F9 and Zlt1/3Zlt1 with F9 x F9. The homomorphism u: zlt1 -
Zlt1/3Zlt1 - F9 x F9 is given by
u(t) = (t, - t) (27)

or, using the representation (22), by

u (a) =- (ao+al t+a2t2+a3t3, ao-al t+a2t2-a3t3). (28)

The following construction is the appropriate analog of Construction A/.

Construction At/>. If Band C are codes of length n over F9 with C !;;; B·,
we define
(29)

with the hermitian inner product

1
i(q,(V\oWI)+ ... +q,(vn,wn ». (30)

The corresponding real lattice A (B, C) real is obtained by regarding

A (B, C) as a sublattice of ~~!;;; C2n , and writing down real and
imaginary parts (cf. Eq. (66) of Chap. 2). Then
det A (B,C)real = 4n (detA(B,C»4. (31)

Next we shall relate the norms in A (B, C) to the weights of the

codewords in Band C. The map T of (25) induces a map
t : F9 x F9 - F9; it is given by

(see Table 8.3). There is a corresponding "norm" N: F9 x F9 - F3

defined analogously to (26). For a = (b,c) E F9 x F9, let a
= (c3,b 3).
a
The component-wise product a is (bc 3,cb 3), and we define the norm of a
by
(32)

which takes values in F 3 - to, 1, -1l.

Theorem 3 lQue6J. The elements in zltl of norm 2 map I-to-1 under u
onto the elements of norm -1 in F9 x F9; the elements of norm 4 map I-
to-I onto the elements of norm 1; and the elements of norm 6 map 3 -to-I
onto the nonzero elements of norm 0.
Sketch of proof This comes from the fact that D 4 /3D 4 consists of 81
congruence classes: the zero class, 24 classes represented by single vectors
218 Chapter 8

Table 8.3. db,c) for b, c E F 9 .

b\c 0 I ~ ~2 ~3 ~ ~5 ~6 ~7

0 0 0 0 0 0 0 0 0 0
I 0 -I I I 0 I -I -I 0
~ 0 0 I -I -I 0 -I I I
~2 0 -I 0 -I I I 0 I -I
~3 0 I I 0 I -I -I 0 -I
~4 0 I -I -I 0 -I I I 0
~5 0 0 -I I I 0 I -I -I
~6 0 I 0 I -I -I 0 -I I
~7 0 -I -I 0 -I I I 0 I

of norm 2, 24 classes represented by single vectors of norm 4, and

32 classes each containing precisely three vectors of norm 6. The vectors
in Z[~] ;;;; D4 of nonzero norm (mod 3) map into vectors of nonzero norm
in F9 x F9.
Theorem 3 makes it possible to find the minimal norm in A(B,C). We
see that for each pair of codewords b = (b l • . . . • bn ) E B,
C = (CI • . . . • cn) E C, we must determine the number of coordinates i
with (bj,cj) ¢ (0,0) for which N(bj,cj) takes the values -I, I and 0: let
these numbers be PI(b,d, P2(b,d and P3(b,d respectively. Let Po(b,c)
be the number of i for which (bj,cj) = (0,0), so that
3
n = ~ p" (b, d . (33)
"-0

If v E Z[~]n is such that IT (v) = (b, d ¢ (0,0), then, from Theorem 3,

N(v) ~ 2PI(b,d + 4p2(b,c) + 6p3(b,d, (34)

and the number of v for which equality holds in (34) is 3P #.cl. We call
the quadruple (Po(b,d, PI (b, d, P2(b,d, P3(b ,d) the type of (b, d, and
set

Furthermore one can show (as in the proof of Theorem I) that if

b I CI+ ... + bn c;, = 0 then P (b, c) is a multiple of 3, which justifies the
division by 3 in (30), and proves that A (B, C) is an integral Z[~]-lattice.
Since P (b, c) is obviously even, A (B, C) is in fact an even integral Z £r)-
lattice. Also A(B,C)* = A(C*,B*), so if C = B* then A is an even
unimodular Z[~]-Iattice.
The elements v E Z£r)n such that (J (v) = (0,0) belong to 3z[~]n, a
sublattice of A (B, C) with minimal norm 6 (using (30». Therefore (from
Algebraic Constructions for Lattices 219

(34» the minimal norm of A (B, C) is

min {d/3, 6},

where
d = min {p (b, c) : b E B, c E C, (b, c) ;z:! (0,0) } (37)

is a kind of "minimal distance" for the pair of codes (B, C). We

summarize the preceding results:
Theorem 4 [Que61. A (B, C) is an even integral Z[~]- lattice and is
unimodular if C = B*. The minimal norm is given by (36), (37). The
kissing number is given by
T = ~ 3P3(b,cl, if d < 18, (38)

the sum being over all b E B, c E C with p(b,c) = d, or by

T = 24n , if d > 18, (39)

or by the sum of these two expressions if d = 18.

Quebbemann also gives a result analogous to Theorem 3 of Chap. 7 for
the theta series of A (B, C) .
Example. The Barnes- Wall lattice A 16 again. Let B be the [4,1,4]
repetition code over F9 and let C = B* be the [4,3,2] zero-sum code. If
b ;z:! 0, b has four non-zero components and (from Table 8.3) p (b,c) ~ 8.
If b = 0, c ;z:! 0, p(b,c) ~ 18. Since 6Ip(b,c) we conclude that d, the
minimal value of p (b, c) for (b, c) ;z:! (0,0), is 12. Therefore A (B, C) real
is a 16-dimensional lattice with minimal norm 12/3 = 4 and determinant
44 (from (31)).
To determine the kissing number we must find all (b, c) E B x C with
p(b,c) = 12. The possible types (PO,PhP2,P3) are (from (33), (35»
(0,2,2,0), (0,3,0,1) and (2,0,0,2). Consider type (0,3,0,1), and assume
b = (1,1,1,1). Then to have PI (b,c) = P2(b,c) = 2 we see from
Table 8.3 that c must contain two elements from {I,~5=-~,~6=-1 +~}
and two from {~4 = -1,~, ~2 = 1 - d, with the sum of all four equal to 0.
There are 162 possibilities, and the total number of pairs (b, c) of this type
is 8· 162 = 1296. A typical pair is b = (III 1), c = (I,-~, 1-~, 1-~),
and the unique v E Z[~]n with u(v) = (b,c) and N(v) = 12 is
(1,~3-~2,-1 +2~-~2,-1 +2~_~2) .

The corresponding vector in RI6 is

(1 °1 0, °°-1 -1, 1 -1 -1 1, 1 -1 -1 1).
The number of (b, c) of type (0,3,0, 1) is 864, and the number of type
(2,0,0,2) is 48. From (38) the kissing number is
1296 + 864 . 3 + 48 . 32 = 4320. A (B, C) real is the Barnes-Wall lattice
Al6 (§1O of Chap. 4).
220 Chapter S

There is also an analog of Theorems 17 and 32 of Chap. 7, with a

similar proof.
Theorem 5 [Que6]. If A is an even unimodular Z[\l-lattice then

0 A(z) E C [0 D4 (Z), 6 16 (Z) 1 (40)

where

6 16 (Z) = :6 {0 A1/Z) -0D /Z)4} = q2 II ((J _q2m)(I_ q 4m)J8

m-I

Remark. The theta series in Theorem 17 of Chap. 7, in the preceding

theorem, and in Theorem 32 of Chap. 7 are respectively modular forms for
the groups H(]) == SL 2 (Z), H(2) and H(3), where the Hecke group H()...)
is generated by Z -+ Z +)... and z -+ - I /z ([Hec3, p. 609], [Oggl, p. xiii),
[Que6]).
As in Chap. 7 we may now define extremal theta series and lattices.
An n-dimensional extremal Z[S-)-Iattice has minimal norm 2[n/4) + 2, and
[Que61 gives examples for n ~ 11. The initial examples are z[\ln for
n = 1,2,3, and then A(B,B') where B is the repetition code of length n,
for 4 ~ n ~ 7. The next example is the most interesting.
Example. The 32-dimensional real lattice Q32. For n = 8 Quebbemann
[Que6] takes B to be the [8,2,7] Reed-Solomon code over F9 (Chap. 3,

° °
§2.5) generated by (ill...]) and (Iww 2 ... w 7 ), so that C = B* is an
[8,6,3] Reed-Solomon code. If b ;: then p (b, c) ~ 2 . 7 = 14, and if
b = 0, c ;: then p (b, c) ~ 2 . 3 . 6 = 36. Thus d = 18, so A (B, C) has
minimal norm 6, and is extremal. Its theta series is
0 D4 (Z)8 - 1920 D4 (z)46 16 (Z) + 5766 16 (Z)2
= 1 + 261120q6 + + 535818240qlO
18947520q8
+ 8320327680ql2 + 83347937280q14 + ... , (42)

Quebbemann found that there are 448 pairs (b,c) E B x C of type

(5,0,0,3),36288 of type (J,6,0,]), 108864 of type (1,5,2,0) and 31104 of
type (0,7,1,0), and so from (38) + (39) the kissing number is 261120, in
agreement with (42).

The corresponding real lattice is a 32-dimensional lattice Q32 with

determinant = 2 16 , minimal norm = 6, kissing number = 261120, center
density 0 = T 24 3 16 = 2.5658 ... and theta series (42). This is the densest
packing known in 32 dimensions. Q32 was originally constructed in [Que5]
by the method described in §2. That method also yields lattices Q30 in R 30
with 0 = T22313.5 = 0.6584 ... and Q31 in R 31 with 0 = T23531S = 1.2095 ...
(see Fig. 1.5). A slightly less dense 32-dimensional lattice is constructed in
§8.2(h) below.
Algebraic Constructions for Lattices 221

5. McKay's 40-dimensional extremal lattice

McKay [McK2] describes a construction for the Leech lattice that also
produces a 40-dimensional even unimodular lattice of minimal norm 4.
Let p == -1 (mod 4) be prime and let H be a Hadamard matrix (§2.13 of
Chap. 3) of order p + 1 of the form H = 8 - I, where 8 is skew-
symmetric. (This implies 8 2 = -pl.) Let n = 2(P + 1) = 8k. Then an
n -dimensional lattice Mn is defined by the generator matrix

1 [(k+lH 0]
.Jk + 1 8 - 21 I' (43)

The Gram matrix is

(k+1)1 -8-21]
[ (44)
8-2/ 41 '

so Mn is an integral even unimodular lattice if k is odd. It is a

straightforward but tedious calculation (which we omit) to verify that if
k = 3 or 5 then Mn has minimal norm 4. For k = 1, 3 and 5 there is
essentially only one way to choose 8 [Lon11. The resulting lattices are
Ms ,;;; Es, M 24 ';;; A24 and M 40 , an extremal Type II lattice.
A slightly nicer description of M 40 is obtained if the first column of
(43) is negated. Thus as generator matrix for M 40 we take

~ [6~ ~], (45)

.
where B is constructed as follows. Let the rows and columns of B be
labeled {co,O, 1 •...• 18}. Then Boooo = 2, Bii = -2 and Boo; = B;oo = 1
for i ~ 0, Bij = 1 if j - i is a quadratic residue modulo 19 and B;j = -I
if j - i is a nonresidue, for i ¢ j, i,j ~ O. M 40 has det -= I, minimal
norm = 4, kissing number = 39600 (the minimal vectors being given in
Table 8.4), packing radius = 1, center density 0 = 1 (to be compared with
o= 4 for the lattice A40 of Chaps. 5 and 6), and theta series
E 4(z)5 - 1200 E4(z)2~24(z)
= 1 + 39600q4 + 87859200q6 + 20779902000qS + (46)

To describe the minimal vectors and the automorphism group we represent

M 40 as a 20-dimensional complex lattice, a typical vector being written
x = (XooXOXI",XIS)' Then the 20 real parts give the left-hand half of
(45), and the 20 imaginary parts the right-hand half.
Let Q = {1,4,5,6,7,9,1l,16,17}, N = (2,3,8,1O,12,13,14,15,18}, and
set 1I'(t} = +1 if t E to} U Q, 1I'(t} = -I if t E {co} UN. Then
Aut(M 40) has a subgroup isomorphic to 2.PGL 2 (19) of order 13680 (this
may in fact be all of Aut(M4O» consisting of the transformations
UA : X t - .J1I' (ad -be) 11' (ct +d)x at+b, (47)
ct+d
222 Chapter 8

for A = [: !j, a, b, c, d E F 19 , ad - bc = ± I, where either square root

may be taken. For example, if A = [~:], aA : Xl -- Xl+l; if A = [-~ ~ j,
aA : Xl -- iX-I> which swaps the two halves of (45). In this complex
notation M 40 is generated (over Z) by the vectors of type (6,0 19), (6i,019),
and the images of (2 + i, 119) under the transformations (47).
In Table 8.4 we write the components of the minimal vectors in the
order X~XO[XIX2X4X8 . . • XIO], where the brackets enclose the cycle of
length 18 under Xl -- X 21' It seems that there are seven orbits of minimal
vectors, and for each orbit we give a representative vector, the order of the
subgroup of 2 . PGL 2 (I9) fixing it, and the size of the orbit. However, we
have not been able to exclude the possibility that the group is slightly
larger than indicated, and that some of the orbits fuse. Note that this is
not a Gaussian lattice in the sense of Chap. 2, §2.6, since it is not closed
under multiplication by i.

Table 8.4. The minimal vectors of M 40 (a=l+i, b=l-i, c=-I+i,

d = -1-;). The coordinates are in the order x ~XO [Xl X2X4'" X5X101.

Coordina tes Stabilizer Number

2 + i, 1[118] 171 80
a, 2i [(a,0)9] 9 152O
i, 2 + i[U, 1)9] 9 152O
2 + i, I [(I,i,i,i,-i,-1)3] 3 4560
2i, b [(c,0,a,0,0,b)3] 3 4560
2i - I, -1 [1,-1, 12,-i,i 5,1 2,-I,-i,I,-I,-i,i] 13680
2, -2i [c,0,0,d,d,b,a,0,b,b,a,07] 13680

6. Repeated differences and Craig's lattices

Let A be an n -dimensional lattice in Rm, with m ~ n. A simplicial basis
for A is a set of n + I vectors Vo, VI, . . . ,Vn which span A and satisfy
Vo + VI + ... + Vn = 0. Then the difference lattice ~A (with respect to
this basis) is spanned by the vectors Vi - Vj. Clearly ~A ~ A.
Let if be the (n + I) x m matrix with rows Vo, • . . ,Vn . By omitting
the last row we obtain a generator matrix M for A. If A and ~ denote the
matrices

° °° ° °° °°
° °
I -I -I

A=
-I
° °, ~=
-I

° °° °° I -I
°°° -I

of sizes (n
-I

+ I) x (n + I)
°
and n x n respectively, then the rows of
2

Aif
Algebraic Constructions for Lattices 223

form a simplicial basis for ~A, and ~M is a generator matrix for ~A.
Although ~A depends on the particular simplicial basis used, its
determinant does not.
Theorem 6.
det ~A = (n + 1) 2det A. (48)

Proof This follows since det ~ = n + 1.

Examples (a). For A~ we may take
if = (n+1)-1/2 «n+l)Jn+I-J), (49)

where J is the all-ones matrix. Then ~A~ = .In+T An. The minimal
norm has increased from n to 2n + 2.
(b) Craig's lattices. The following lattices were first described by Craig
[Cra5] using the construction of §7.3c. As a simplicial basis for An we
take the rows of ,i itself, and define An(m) = ~m-I An for m = 1,2, ....
Thus the rows of ,im form a simplicial basis for An(m). For example AJ3) is
spanned by the rows of
1 -3 3 -1 0 0 0
o -3 3 -1 0 0

3 -1 0 o 0 +1 -3
-3 3-1 000

(These lattices should not be confused with Coxeter's lattices A~ [CoxlO];

Coxeter's A~ is our An [s 1, where rs = n + I,)
Theorem 7 [Cra5]. Adm) has determinant (n + I)2m-l. If n + 1 = P is a
prime and m < V2n. Adm) has minimal norm at least 2m.
Proof The first assertion follows from Theorem 6. Since An(m) ~ An, the
minimal norm is even. To find the minimal norm we represent vectors
u = (uou I ... un) E An(m) by polynomials U ex) = Uo + U IX + ...
+ unxn in the ring R = Z[X]/exp - 1). Then Adm) is represented by the
principal ideal (ex - 1)m) in R. Suppose U E An(m) has norm ~ 2m - 2.
Then there are two subsets S, T ~ {a, 1 ....• n} of size m - 1, possibly
containing repeated elements, such that
U ex) = ~ Xs - ~ Xl .
sES lET

Now U ex) = ex) ex - 1)m in R, for some a ex), so

a
U ex) = a ex) ex - 1)m + b (x) exp - 1) in Z[X1.

By repeatedly differentiating this expression we find that

u(i)(I) == 0 (mod p) for i = O..... m - 1.
Hence
224 Chapter 8

~ Si == ~ t i (mod p), i = 0, ... ,m - 1.

sES lET

Therefore by Newton's identities [Mac6, p.24S], [Wae2, I, p. 81], the

elementary symmetric functions of Sand T of degree < m agree, and so
the polynomials
II (X-s), II (X-t)
sES lET

coincide over Fp. Since a polynomial of degree m - lover a field has at

most m - I zeros, we conclude that S = T, and u = O. This completes
the proof.
It can be shown that the Gram matrix for A;m) is obtained from ~m
(~m)lr by omitting the last row and column, and has (j ,j) entry equal to

(-1) (;+ j) [2km] , (SO)

where k = m + Ii - j I (mod n + 1).

When n + 1 is a prime, Theorem? gives a lower bound on the density
of A;m). The lower bound is maximized by taking m = mo, the nearest
integer to Y2 n/loge (n + 1). Examples are given in Table 1.3, where it can
be seen that these are the densest pac kings known for 148 ~ n ~ 3000.
For large n the density ~ satisfies
I I
- log2 ~ ~ - - log2 log2 n + 0 (n) ,
n 2

which (assuming that the lower bound is close to the actual value) is
inferior to the lattices constructed in §9 (compare Table 1.4).

7. Lattices from algebraic number theory

7.1 Introduction. The connections between lattices and algebraic number
theory have been studied by many authors from Minkowski onwards. In
this section we describe some of the principal ways in which lattices can be
obtained from ideals in algebraic number fields. This account has been
influenced by the work of Lekkerkerker [Lekl, §4], Craig [Cra3]-[CraS],
Thompson [Tho?] and Litsin and Tsfasman [Lit3] (see also [Bayl], [Feil],
[Quel], [Taul]). For background information in algebraic number theory
see for example [CasS], [Coh6], [Hasll, [HecO], £Jan?], [Lan?], [LeVI],
[Narl], [Ribll, [Wae2], [Wei2].
7.2 Lattices from the trace norm. Let K = Q(O), of degree n, be an
algebraic number field which is Galois with primitive element O. Let the
conjugates of a typical element a E K be a = a(ll, ... , a(n), so that
(1), ... ,(n) are names for the elements of the Galois group
G = Gal (K / Q). The trace and norm of a E K are
(S1)
Algebraic Constructions for Lattices 225

(52)

We apologize for the confusion caused by the multiple uses of the word
"norm". We use N for the norm of a lattice vector (§2.1 of Chap. 2) and
Norm when the word is used in the sense of algebraic number theory.
The two most interesting cases for our purpose occur when K is either
totally real (all the conjugates ()(j) are rea!) or totally complex (none of
the ()O) are rea!). If K is totally complex then n = 2s, say, and we label G
so that
a(S + J) = a(J), ... , a(2s) = atn

and "(s + 1)" is complex conjugation. Elements a E K may be represented

as vectors in Rn by the mappings
a - v (a) = (a(\) , a(2), ... , a(n» (53)

if K is totally real, or
a - v (a) = (Rea(J), Ima(\), ... , Rea(s>, Ima(s» (54)

if K is totally complex. The (lattice) norm of v (a) is

n
N(v(a» = ~ a(j)2 = TrKIQ (a 2) (55a)
j-I

if K is totally real, or

(55b)

if K is totally complex. The trace norm defined by (55a), (55b) is just one
of many possible ways of defining a norm on K; another is given in §7.5.
If we define c = 1 if K is totally real and c = V2 if K is totally complex,
(55a) and (55b) may be combined into
N(v(a» = c TrKIQ <laI 2). (56)

Let (!J denote the ring of algebraic integers in K, and let WI, ••. , W n
be an integral basis for (!J. The matrix n = (w?» (1 ~ j, k ~ n) has the
property that the (j, k)th entry of n n is
TrKIQ (w j Wk),

and (det n)2 = Idisc(K) I is the absolute value of the discriminant of K.

Finally, let d ~ (!J be an integral ideal. A real lattice A (.91) is
defined by
A (.91) = {v (a) : a Ed}. (57)

We first discuss the case when d = (!J.

226 Chapter 8

Theolt,.iI 8. If K is totally real or totally complex, A ({0 is an n-

dimensional real lattice with det A (@) = c n 1 disc(K) 1 and minimal norm
at least cn.
Proof If K is totally real then n is a generator matrix for the lattice, so
det A({0 = (deL n)2 = Idisc(K) I. For a E @,a;:e 0, we have
N (v (a» = TrKIQ (a 2) ~ n {Norm (a 2) ]lIn,

by the arithmetic-mean geometric-mean inequality. This is at least n since

Nonn (a 2) is a positive integer. The proof in the totally complex case is
similar.
Remark. If K is totally real, or if K is totally complex and complex
conjugation is in the center of G, then c-1N(v(a» E Z and c- 1/2 A({0 is
an integral lattice.
TbeOll:ill 9. For any ideal d = {Si, A (,r;1) is an n-dimensional sublattice
of A ({Si), and
det A (,r;1) = c n Norm (,r;1)2 1 disc(K) I. (58)

Proof (Cf. [LeVI, 11, p. 681.) We may choose an integral basis for d of
the form

(59)

where aij E Z. Then det A (,r;1) = (a 11 a 22 . . . ann) 2 det A ( (0, and one
can show that all' .. ann = Norm (,r;1) = number of residue classes of @
modulo d.
For a E @ let M (a) be the n x n integral matrix defined by

= M(a) (60)

Then
n diag {a(I), ... , a(n)} = M (a) n , (61)

so det M(a) = Norm(a). This immediately implies the following theorem.

Theorem 10 [Cra3]. If d = (a) is a principal ideal, A (.r;1) has generator
matrix M(a)n. If si= (a, a', a", .. .), A(,r;1) has generator matrix Mn,
where M is a generator matrix for the sublattice of zn spanned by the
rows of M(a), M(a') ,....
Algebraic Constructions for Lattices 227

7.3 Examples from cyclotomic fields. Craig [Cra3]-[CraS] has obtained a

number of interesting examples by applying this construction to ideals in
cyclotomic fields.
For any integer m ~ 3 let ! = !m = e2-r:i/m. The cyclotomic field
Q(!m) is an abelian extension of Q of degree n = cp(m), and is totally
complex when m is odd. The ring of integers (!) = Z[!m], with integral
basis !, !2. . . . • !", and

disc Q(!m) - (_1)11/2 mil II p-II/(P-ll. (62)

plm

The corresponding lattice A (z[!mD has Gram matrix A - (Ajk), where

([Cra3D
A - jL(d)cp(m) d = m (63)
jk- cp(d) , (m,k-j)'

If m is an odd prime, A (z[!mD ;;; A~_I'

(a) The lattice E 6 • When m = 9 and d = ((I -!9)2), Craig [Cra3]
shows that A(.9I) == E 6 • He is able to construct five of the six 6-
dimensional extreme forms [Bar6], [Bar7] in this manner.
(b) The Leech lattice. Suppose m = 39, n = cp (39) = 24. In Q(!39)
the ideals (3) and (13) have prime factorizations
(3) = (f!lJl f!lJ1 f!lJ2 f!lJ2)2, (13) = U1l2ll) 12,

where f!lJJ, f!lJ2 have Norm 33 and 211 has Norm 13. Let d = f!lJ1 f!lJ2 211.
Then Craig [Cra4] shows that A (.91) = A24 . However, this is not a very
simple way to construct A24, and the proof that A (.91) is A24 is fairly
complicated. A more natural cyclotomic construction of A24 is given in
§7.5.
(c) The lattices A;m). Craig's lattices A;m) described in §6 were
originally constructed in [CraS] as the lattices A(.9I), where d is the ideal
((I _!p)m+l) in the cyclotomic field Q(!p) and p ... n + 1 is a prime.
7.4 Lattices from class field towers. As pointed out by Litsyn and
Tsfasman [Lit3], the infinite class field towers found by Golod and
Shafarevich [GoI9], Martinet [Mar2], [Mar3] and others [RoqI1 produce
infinite sequences of very dense lattices. For example, these authors show
that an infinite sequence or tower of totally complex fields
(64)

may be obtained by starting with a suitable KJ, and defining Ki+l to be

the maximal unramified extension of Ki whose Galois group GaHKi+dKi)
is an abelian 2-group. Martinet [Mar2] proves that if k is totally real field
of degree n ~ 10, and q is a prime number that totally decomposes in k,
then the totally complex field K 1 = k (.r-tj) may be used in (64). For
example, we may take k = Q (cos 211' /11, .J2) and q = 23, obtaining
228 Chapter 8

a field of degree 20 over Q with discriminant 230 11 16 2310. The subfield

Q (cos ~~ , .J-46) , (66)

of (65), having degree 10 over Q and discriminant 2 15 11 8 23 5, may also be

used for K I.
In the tower (64) let [K; : K;-d = hi, [K; : Ql = n; and disc(K) = D;.
Then
(67)

and, since the extensions are unramified,

IDi+11 = ID;lh i +1 (68)

[Hasl, Chap. 25, §51. Applying the construction (57) to the ring (!); of
algebraic integers in K;, we obtain a lattice A «(!) of dimension n;. From
Theorem 8 the packing radius and center density satisfy p ~ .Jil;/2.J2,

(69)

and so the density of A «(!);) satisfies

- 1 log2 Ll ~ - - 1 log ID; I - log2

n; 2n;
-J!i +-
7re
0 [IOg2
--
n
n; )
.

However, from (67), (68),

IOg2I Di+11 log2 I D;I
2 n;+1 2n;

So the sequence of lattices A «(!);) satisfies

1 1 2ID I II / n l [IOg2 n;)

- log2 Ll; ~ - - log2 +0 -- . (70)
n; 2 7re n;

Choosing the field (66) for K b we obtain

1
- log2 Ll; ~ -2.218 ....
n;

Unfortunately at present there does not seem to be any practicable method

known for explicitly finding these lattices (or even their dimensions). In
view of Odlyzko's bounds on the discriminants of number fields [Odlll-
[Odl3], lattices constructed in this way always satisfy
1
-log2 Ll < -1.193,
n
Algebraic Constructions for Lattices 229

and, assuming the generalized Riemann hypothesis,

1
-log2 d < -1.694, (73)
n

so they can never reach Minkowski's bound (Chap. I, Eq. (29».

7.5 Unimodular lattices with an automorphism of prime order. The lattices

Es, Dt., A24, P4Sq are even unimodular lattices in dimension n "" p + 1
where p is a prime, having an automorphism of order p. The following
analysis, due to Thompson [Th07], makes it possible (at least in principle)
to find all such lattices. (A somewhat similar approach has been used in
[Bayl1, [Cra5], [FeB], [Quel].) Such a lattice arises naturally as a 2-
dimensional sublattice r glued to a (p - O-dimensional sublattice d, where
d is defined by an ideal in a cyclotomic field, using a trace form that
generalizes the one used in (56).
Let A be an n-dimensional even unimodular lattice, where
n == p + 1 == 0 (mod 8) and p is prime, containing an element q of order
p in its automorphism group G. Let r - ker(l- q), the sublattice fixed by
q, and let

d - (l-q)A == {v EA: V'u-O for all u E r}.

Thus r has dimension 2 and d has dimension p - 1.

We show that d has a natural representation as an integral ideal .91 in
the cyclotomic field K =- Q(r), where r"" e 21fj / p • The ring of integers in K
is Z[n Now r satisfies
1 + r + r2 + ... + rp - I "" 0,

and the element v = 1 + q + ... + qP-1 in the group ring Z[q] satisfies
vd- (l-qP)A=O.

Therefore we may regard d as a Z[rJ-module, or more precisely there is an

isomorphism

(74)

where .91 is an ideal of Z[n

Theorem 11 ([Tho7J, cf [Feil, Theorem 6.11). (a) With A, d as above,
there is a totally positive element l' E K (i.e. l' and all its conjugates are
positive) such that
u .v = TrK/Q (4) (u)-y4> (v» (75)

for all u, v E d, and

(b) l' E .sfl .91- 1 ~-I, (76)

230 Chapter 8

where f!) is the different of K/Q. (c) Conversely, given any ideal .91 of
Z[~] and a totally positive element,), E Z[~] satisfying (76), .91 becomes a
positive definite integral lattice if we define
(77)

Sketch of proof (a) The inner product u . v on d extends in the obvious

way to d ® Q; but from (74) d ® Q;;;; .91 ® Q = K. So we have an inner
product ( , ) on K satisfying u· v = (c/>(u), c/>(v» for u, v Ed.
Furthermore c/> (au) = ~c/> (u) for u E d, which implies (a,{3) = (~a,N3)
for a, (3 E d. Therefore the inner product on K is determined by the
p - 1 numbers
(1,~i) = ai E Q, 0 ~ i ~ P - 2. (78)

Now r-I. ... ,r(p-l) is a basis for Kover Q, and there exists a dual basis
(J I, . . . , (Jp -I satisfying

[Lan8, p. 212]. If we set

p-I
')' = ~ ai(Jj,
i-I

then
(79)

and (75) follows from (78), (79). Furthermore, since A is positive definite,
TrK/Q(a')'a> = (a,a) >0
for all a E K, a ;II! O. This implies that')' is totally positive. (b) We also
know that u . v E Z for u, v E d, so
TrK/Qha~) E Z for a,{3 E .91,

whic.!t means that')' is in the complementary ideal_to f!jJ = .91.91, denoted

by f!jJ (cf. [Ribl, p. 204], [Wei2, p. 107]). Now f!jJ is the fractional ideal
f!jJ = f!jJ-I f!)-I, where f!) is the different of K/Q, and this implies (76).
We omit the proof of the converse.
The next step is to consider how the sublattices d and r are glued
together to form A (cf. §3 of Chap. 4). From Theorem 1 of Chap. 4 we
have
d* /d;;;; r* Jr. (80)

(For, by construction, d = d n (A ® R), r = r n (A ® R), so there is no

self-glue.)
Eq. (80) is an isomorphism of Z[a]-modules. But a acts trivially on
r* Jr, and therefore acts trivially on d * / d. If
Algebraic Constructions for Lattices 231

sl = {x E K : TrK/Q (0l'Y x) E Z for all 01 E d} (81)

is the ideal corresponding to .:l*, t acts trivially on d* , i.e.

(82)

Also d* ~ d (otherwise .:l would be a direct summand of A, and A

would not be even, since dim .:l is not a multiple of 8).

Theorem 12 [Th07]. (a) With A. .:l. r as above. the ideal d representing

.:l is such that
(83)

is a principal ideal with a totally positive generator 'Y. and

(b) (84)

(c) Conversely. if d is an ideal of ZW satisfying (83) where 'Y is totally

positive then the inner product (57) makes d into an even integral lattice
with determinant p.
Sketch of proof (a), (b). From (81), d* is the complementary ideal to
d~,so
(85)

From (82), (85),

(86)

Also dd'Y C ~-I, where the containment is proper since d* ~ d.

Thus

-I -I
But (1- t) ~ has index p in ~ , so finally we may conclude that
-I -
(1-t) ~ = dd'Y, (87)

(1-t)d* = d. (88)

Therefore d has index p in d*, which proves (84). In this cyclotomic

field we know that ~ is the principal ideal «I - t)p-2), and
~-I = «1-t)2-p). Then (83) follows from (87).
Examples. The Leech lattice arises by taking n = 24, P = 23, and letting
d be a prime ideal dividing 2. Now X 23 + 1 factors modulo 2 into
(x + I)fl(x)h(x), where
fl(x) = Xli + x lO + x 6 + x 5 + X4 + x 2 + I,
(89)
232 Chapter 8

Therefore in Z[f23], (2) factors into (2) = fJJJ1 fJJJ2, where

.0J>1 = (2'/1 (f», fJJJ2 = (2'/2 (f» .
We take d = fJJJ], d = fJJJ2, and

(90)

which defines A. We must take r to be a 2-dimensional even lattice of

determinant 23 and minimal norm 4, and so r may be chosen to have
Gram matrix [~~]. One can now show that there is a unique way to glue
A to r to produce an even unimodular lattice, which is A24 .
The lattice P 48q (Chap. 7, Example 9) arises in an exactly similar way.
X 47+ 1 factors modulo 2 into (x + 1) f 1(x) f 2(x), where f 1 and h have
degree 23. We take d = fJJJ1 = (2'/1 (f», a prime ideal dividing 2, and
r = [~~]. Again there is a unique way to glue A to r to produce P 48q.
8. Constructions D and D'
8.1 Construction D. This construction, first given in [BariS], uses a
nested family of binary codes to produce a lattice packing L in Rn. It
generalizes a construction of Barnes and Wall [Barl8], and has some
features in common with Construction C of Chap. 5, §6.1, although
differing from it in always producing lattice packings. (On the other hand
Construction C can be applied to codes that are not nested and to
nonlinear codes.)
Let 'Y = 1 or 2, and let Co :2 C 1 :2 .. , :2 Ca be binary linear codes,
where C; has parameters [n, k;, d;], with d; ~ 4; h for i = 1, ... , a, and
Co is the [n, n, 11 universe code Fg. Choose a basis C 1, ... ,Cn for Fg such
that (j) Cl, • . . ,Cki span C; for i = 0, ... ,a, and (ji) if M denotes the
matrix with rows C], ... ,Cn , some permutation of the rows of M forms an
upper triangular matrix. Define the map II; : F 2 -+ R by II; (x) = x li- 1
(x = 0 or 1), for i = 1, ... , a, and let the same symbol II; denote the map
Fg -+ R n given by
II; (x 1, . . . , X n) = (II; (x 1)' . . . , II; (x n» .

Also let ka+l = O. Then the new lattice L in R n consists of all vectors of
the form
a k;
I + ~ ~ aj
(;) ()
II; Cj , (91)
;-1 j - I

where I E (2z)n and ayl = 0 or 1.

Theorem 13 [BariS]. L is a lattice, with minimal norm at least 4h,
determinant
Algebraic Constructions for Lattices 233

n-1; k i
det L .. 4 i-I (92)

and hence center density

Q

1; ki- n
~ ~ 'Y- n / 2 2i - 1 (93)

This is a consequence of Theorems 15 and 16 below.

An integral basis for L is given by the vectors Ui (Cj) for i -= 1, ... , a,
j = ki +,
+ I, ... , k i , plus n - k, vectors of the shape
(0 .... ,0,2,0, ... ,0).
8.2 Examples. (a) Construction A. When a - 'Y ... 1, Construction D
reduces to the linear case of Construction A (Chap. 5, §2.0. A typical
example is obtained when C, is the [8,4,4] Hamming code 318. We may
take
c, = 0,1,1,1,1,1,1,0,
C2 = (0,1,0,1,0,1,0,0,
C3 == (0,0,1,1,0,0,1,1),
C4 == <0,0,0,0,1,1,1,0,
Cs - (0,0,0,1,0,0,0,0),
c6 = (0,0,0,0,0,1,0,0),
C7 = (0,0,0,0,0,0,1,0),
C8 = (0,0,0,0,0,0,0,1).

Then L is a version of the E 8 lattice (see Example 5 of the previous

chapter), is spanned by the rows of the matrix

1 1 111
010 010
001100
o 0 001 111
o0 0 2 0 0 0 0 '
00000 200
o0 0 0 0 0 2 0
o0 0 0 0 0 0 2
and has minimal norm 4, determinant 28, and ~ ... r4.

(b) Construction B. If a == 'Y - 2, C, is the [n, n - 1,2] even weight

code, C 2 has minimal distance 8 and the weight of every codeword of C 2 is
a multiple of 4, then Construction D reduces to the version of
 234 Chapter 8

Construction B given in §5 of Chap. 7. For example, as in Example 7 of

Chap. 7, if C 2 is the [16,5,8] Reed-Muller code we obtain A16 == BW 16 •
(c) Construction C. Whenever Construction C of Chap. 5 is applied to
codes that are linear and nested, Construction D produces a lattice packing
with the same density.
(d) Rao and Reddy's code and a lattice in R36. Figure 8.2 shows a [48,
31, 8] code C found by Rao and Reddy [Raoll The dual code C' has
minimal distance 12; let a E C' have weight 12. The codewords c E C
which vanish on the support of a (with these twelve O's deleted) form a
[36, 20, 8] code C 2• The maximal weight occurring in C 2 is 32; let b E C 2
have weight 32. By applying Construction D to the codes C 1 = [36,35,2],
C 2, C 3 = {O,b} we obtain a 36-dimensional lattice B36 with h = 4 (see
Table 1.2 and Fig. 1.5). Similarly one can obtain h = .J2, 2 and 2.J2 in
dimensions 33, 34 and 35. This is an improvement of Bos' earlier use of
Construction C to obtain nonlattice packings with the same densities
(personal communication).

MI 0 0
~
MI=~
M=
M2 M2 0
,
M2

M3
0

M3
M2

M3
M- 5EJ
2-~

Figure 8.2. Generator matrix M for the [48, 31, 81 Rao-Reddy code.
Here u - (11111111), and .YEs (Eq. (57) of Chap. 3),88 , M], M2 are
generator matrices for [8, 4, 41, [8, 7, 21, [16, 5, 8], [16, 11,41 codes
respectively.

(e) A 64-dimensional lattice. For a second example we use the codes in

Eq. (13) of Chap. 5 to obtain a lattice P64c in R64 with 0 - 222.
(f) The Barnes-Wall lattices. Generalizing example (b), we apply
Construction D to the Reed-Muller codes used in §6.5 of Chap. 5. This
yields the Barnes- Wall lattices BWn for n = 2m , m ~ 2. BWn has center
density
(94)

(as in §6.5 of Chap. 5), density

1 1
- log2 ~ - - - log2 n as n --+ 00, (95)
n 4

and kissing number given by Eq. (17) of Chap. 5. Also BW4 == D4 == A4,
BWs == Es == As and BW 16 == A 16 . After multiplication by 1/.J2, BW32 is
Algebraic Constructions for Lattices 235

an extremal Type II lattice with det = 1, ~ = 1, T = 146880. See

Table 1.3.
(g) Lattices from BCH codes. Similarly the nonlattice packings obtained
from BCH codes in §6.6 of Chap. 5 may now be converted to equally dense
lattice packings Bn, for n = 2m. The dimensions of these codes may be
estimated from the BCH bound [Mac6, Chap. 7], and the densities of the
corresponding lattices are shown in Tables 8.5, 1.3. Except for small n the
actual densities of these lattices may be slightly greater than is shown.
(Tables of BCH codes of modest length are given in [Mac6], [PeO].)
Asymptotically the density satisfies
1
;; log2 a - -"21 log2 log2 n , (96)

as shown in §6.7 of Chap. 5. The kissing numbers are not known.

(h) Let C 1 = [32,31,2], C 3 = [32,1,321, and let C 2 be the [32,17,8] code
given in [Che3]. Then Construction D produces a 32-dimensional lattice
C 32 for which ~ - 2 and T - 249280.

Table 8.5. Center density 6 of n-dimensional lattices B. obtained from

BCH codes.

n log2 ~ n log2~ n log2~

4 -3 256 250 16384 57819

8 -4 512 698 32768 130510
16 -4 1024 1817 65536 290998
32 0 2048 4502 131072 642300
64 19 4096 10794 ... ...
128 85 8192 25224

8.3 Construction D'. This construction generalizes another of the

constructions in [BarI8], and converts a set of parity-checks defining a
family of codes into congruences for a lattice, in the same way that
Construction D converts a set of generators for a family of codes into
generators for a lattice.
Let 'Y == 1 or 2, and let C ~ C 1 ~ .•• ~ Ca be binary linear codes,
where Ci has parameters [n,ki,di ] and d i ~ 'Y. 4i for i = 0, ... ,a. Let
hi, ... ,hn be linearly independent vectors in F~ such that (i) for
i == 0, ... , a, the code Ci is defined by the ri .. n - k i parity-check
vectors hi, ... , h,/, (in some rearrangement of hi, ... , h n forms the rows
of an upper triangular matrix, and let r -I - o. Considering the vectors hj
as integral vectors in Rn, with components 0 or 1, we define the new lattice
L' to consist of those x E zn that satisfy the congruences
hj . x == o(mod 2i + l ) (97)

for all i = 0, ... ,a and ra-i-I +1 ~ j ~ ra-i.

236 Chapter 8

Theorem 14 [BarIS]. The minimal norm of L' is at least 'Y . 4a , and

a
log2 det L' = ~ rj. (98)
j-O

Proof (98) holds because for each i = 0, ... ,a the lattice satisfies
ra _j - r a _j -I independent congruences modulo 2j + '. The evaluation of
the minimal norm is straightforward and is omitted.
By changing the scale and relabeling the codes, Construction D may be
restated in such a way that the norms and determinants of Land L' agree.
Then if the Cj are Reed-Muller codes, as in Example 8.2f, the two lattices
coincide. In general however the two constructions produce inequivalent
lattices with the same density.

9. Construction E.
The following rather general construction includes many of the earlier
constructions as special cases. It was first given in [BariS] and [Bos3].
The ingredients for the construction are a lattice A in Rm, an
endomorphism D of A that satisfies certain conditions, one of which is that
AIDA is an elementary abelian group E of order pb, say, and a family of
codes Co ;;2 C 1 ~ •.• ;;2 Ca of length n over E; the result is a family
Lo ~ L 1 ~ . . . ~ La of lattice packings in Rmn.
Hypotheses.
(i) Let A be a lattice in R m with minimal non-zero norm M.
(ij) Let D be an endomorphism of A which is also a similarity (i.e. a
constant times an orthogonal transformation) and which satisfies
,
pD- 1 = ~ ajD j (99)
j-O

for integers p ~ 1, r ~ 0, ao, ... , a,. Let T = D- 1, so that, from (99),

p T is also an endomorphism of A.

It follows that AID A has the structure of an elementary abelian group

E of some order pb (b ~ n.
This implies that there is a b-dimensional
sublattice K ~ A, spanned say by vectors v" ... ,
Vb E A, such that
K I (D A n K) ;;;;; E. The assumptions also imply that

pK ~ K ~ A, (loo)

pK ~ pA ~ DA ~ A, (101)

and therefore that

K I pK = K I (D A n K) ;;;;; E . (102)

Note that (100) and (101) imply pK ~ DA n K, so

K I pK ;;2 K I (D A n K). But both sides have order pb and so must be
equal, and (l02) follows. Furthermore
Algebraic Constructions for Lattices 237

Idet TI = p-b = tm , say,

and T multiplies norms by t 2•

(iii) Assume that all pb - 1 non-zero congruence classes of T A/A have

minimal norm at least t 2 M.
(iv) Let 1> denote the obvious map from Z/pZ to Z which takes the
congruence class x to x, for x E {O, 1, .. ...:..' P - J}. The elements of E
may be identified with the b-tuples X = (XI,'" ,Xb), where all
X; E Z/pZ. Then
X- V<X) = 1> (XI)v1 + ... + 1> (Xb)vb

is a map from E into A, and

X = <XI, ... ,Xn) - V<X) = (V<X I ) , ... , V<Xn»

maps En into An.

(v) Let Co :2 C I :2 ... :2 Ca be additive codes over E of length n
(i.e. abelian subgroups of En), where C; contains j
lk
codewords and has
minimal distance d; (we indicate this by saying that C; has parameters
[n,k;,d;D, and suppose that Co is the trivial [n,n,11 code. Let
C I, . . . , Cbk, E En be chosen so that (0 a typical codeword of C; can be
written as
bkj
~ XjCj, Xj E Z/pZ,
j-I

for i = I, ... , a, and (2) some rearrangement of C I, ••. , Cbk I forms the
rows of an upper triangular matrix.
Construction E. Let Lo = An and, for i = 1,2, ... , a, define

(104)

where the union is taken over all x I, . . . ,Xbk j E {O, 1, ... ,p - J}. By
abuse of notation we shall also use D and T to denote the maps
(D,D, ... ,D) and (T, T, ... , T) acting on Rmn. It is clear that Lo is a
lattice and that D and pT are endomorph isms of Lo.
Theorem 15 [Bos3]. For i = I, ... ,a,
(a) L; is a lattice in R mn , and in fact
L; = Z<L;_h T;Y(cI), ... , T;Y(Cbk) > ; (105)

(b) D maps L j into L; -I; and

(c) pT is an endomorphism of L j •
238 Chapter 8

Proof The proof is by induction on i, the results for i = 0 having already

been mentioned. (a) To show that Li is a lattice, we write any integer
combination of the elements on the right-hand side of (104) as
bk; bk;
/ +p ~ yjTiV(cj) + ~ zjTiV(Cj) , (106)
j-I j-I

where / E L i - I and 0 ~ Zj ~ P - 1. But

pTiV(cj) = pT' Ti-IV(C) E L i - I

by the induction hypothesis, so (106) becomes

bk;
/' + ~ zjTiV(c) ,
j-I

where /' E L i - b which by (104) is in L i . Thus Li is a lattice, and

therefore can be defined by the right-hand side of (105). Now (b) follows
from
DTiV(cj) = Ti-IV(c) ELi-I'

From (105) and (b) we have

r
pT(Li ) = pD-I(L) = ~ ajDj(L) ~ L i ,
j-O

which is (c).
Theorem 16 [Bos3]. For i = 0, I, ... ,a, the determinant, minimal norm
and center density of Li are given by
i
logp Li = n logp det A - 2b ~ kj' (107)
j-I

M = min {M, dj t 2j M for j=I, ... , i}, (108)

and
o= M mn / 2 / 2mn (det L i ) 1/2 (109)

respectively.
Proof Equation (107) is immediate from (104), and (108) follows from
the definition of Li and the fact that the minimal distance of Ci is d i .
Usually we are only interested in the finest lattice La. The construction
may now be applied to La' since it inherits D and T from A, and (99) still
holds. The values of p and t are unchanged, while b becomes nb. A
lattice obtained by applying Construction E to A is denoted by 7] (A) (in
Table 1.3 for example).

10. Examples of Construction E

In most of these examples T is a linear map from Rm to Rm that sends
minimal vectors of A onto deep holes for that lattice. It is helpful to think
Algebraic Constructions for Lattices 239

of T as a "screw" map. For example the map of R2 defined by (I13) is a

clockwise rotation through 45· followed by a contraction by a factor of .J2.
The minimal vectors (± 1,0) and (0, ± n are mapped onto the deep holes
2-1/2(± 1, ± n.
(a) Construction D. If we take A = 2Z and D ... 21, where I denotes the
identity map, then we find T == !hI, P = 2, b - 1, t,..!h, and
Construction E reduces to Construction D. Theorem 13 now follows from
Theorems 15 and 16.
(b) If D is a norm-doubler. In most of the following examples D is a
norm-doubler, so that t == 1/.J2, p = 2 and b = m12. We shall take the
codes C;, i ~ 1, to be maximal distance separable codes (§2.5 of Chap. 3)
over the field F 2b, with parameters [n, k; ... n - 2; + 1, d; - 2; 1. In general,
the largest n for which such codes are presently known to exist is 2b + 1
(see [Mac6, Chap. lID. When n = 2b + 1, C; for i ~ 1 can be taken to
be the cyclic code with generator polynomial
s
g(x) = II (x +e),
j--s

where ~ is a primitive nth root of unity in the field of order 22b and
s = 2;-1 - 1 (see the proof of Theorem 9, Chap. 11 of [Mac61(I».
Besides being additive, these codes are also closed under multiplication by
elements of F2b. However we make no use of this multiplicative structure
in our construction. Clearly Co ~ C 1 ~ C 2 ~ ... . For n less than
2b + 1 the codes are shortened by setting the appropriate number of
information symbols equal to zero. If n lies in the range 2 ~ n ~ 2b + 1,
we use the codes C h C 2 , . . . • Ca , where a is determined by
2 a ~ n < 2a+l. Then the value of };j_1 k j for use in (107) is
an - 2a +1 + a + 2. (110)

Since dj t 2j = 1, the minimal norm is unchanged (see (108». From (107),

(09), (110) and Eq. (17) of Chap. 1 we obtain:
Theorem 17. If an m -dimensional lattice A with center density 6 - 2"Y has
a norm-doubling map D satisfying the hypotheses stated at the beginning
of §9, then Construction E produces an N -dimensional lattice A' with
center density 6' = 2"Y', where N '" mn, 1 ~ n ~ 2m/2 + I,

'Y I == n'Y +; (an - 2a + 1 + a + 2) , (111)

and a = [IOg2n). The density ~ of A' satisfies

1
- log2 ~ =
JL - 1
- log2 -
2m
+ E, (12)
N m 2 'Ire

where 0 < E < (1/2n)Iog2 (4n).

(!)That theorem is incorrect if the field size q is odd (see [Mac6, p. xii])
but that does not concern us since here q is even.
240 Chapter 8

The examples that follow are descendants of the lattices Z2, K 12 , A24
and P 4Sq .
(c) Packings constructed from the lattice Z2. Let A be the two-
dimensional square lattice Z2 with minimal norm M = 1 (see Fig. 8.3).
Although this is not a particularly dense packing, it has surprisingly good
progeny. We let D map (1,0) to (I, 1), and (0, I) to (I,-I), or in matrix
notation, with D mapping v to vD,

D= [!_!], T=D- 1 =t[!_!], (I 13)

so that Eq. (99) reads 2D- 1 = D. Then A/DA ;;; Z/2Z, pb = 21, and we
may take K to be the I-dimensional lattice spanned by VI = (I, 0) (see
Fig. 8.3) .

0 v,

Figure 8.3 . The lattice A = Z2 (represented by the underlying grid of

points), and the sublattices D A (small circles) and K (solid circles).

We apply Construction E to Z2, with n = 2 and C 1 equal to the code

(00, 11} over F 2, and obtain the lattice D 4, with 'Y = - 3 (from (II J)) .
From now on we need not specify D, T or pb, since these are inherited
(see the remarks at the end of §9). For concreteness, however, we note
that if D4 is defined by Eq. (IS) of Chap. 1 or Eq. (86) of Chap. 4, then
T = D- 1 is given by (9I) of Chap. 4 and is essentially unique.
Applying the construction to D4 with n = I . ... . 5 (=2 b + I) we
obtain the laminated lattices
A4=D4,As=Es, A \2, A 16, A20

(see Chap. 6) for which 'Y is

-3, -4, -5, -4,-3

respectively.
E s (the first grandchild of Z2) also has an essentially unique norm-
doubler. If £s is defined by Eq. (99) of Chap. 4, then we may take
Algebraic Constructions for Lattices 241

h 0 0 0

D= ~~~~'h=[!_!]. (I 14)

o0 0 h

We apply Construction E to E 8, using MDS codes over F 16 of lengths

n "" 1,2, ... , 17. The lattices T/ (E 8) obtained in this '!.lly are sEown in
Tables 8.6 and 1.3. These include A 16 again, packings A32 and A40 with
the same density as the laminated lattices A32 and A4o, excellent packings
in dimensions 80 to 136, but indifferent packings in dimensions 24 and 48
to 72. (Nonlattice packings with these densities were already found by
Bos.) The children of E 8 extend only to dimension 136. Beyond this point
the lattices obtained by applying the construction to A\2 and A16 ~re
roughlY comparable, up to dimension 16(28 + O. The children of A2o, A32
and A40 are not quite as good as other known packings, although all of
these families contain some very dense members. Above dimension 140
these packings are inferior to Craig's lattices (§6) and the lattices
constructed in the following paragraphs.

Table 8.6. Center density 6 of N-dimensional lattices 1/ (E 8) obtained

from E8 (N - 8n).

n N log2 ~ n N log2 ~ n N log2 ~

1 8 -4 7 56 12 13 104 60
2 16 -4 8 64 20 14 112 68
3 24 -4 9 72 28 15 120 76
4 32 0 10 80 36 16 128 88
5 40 4 11 88 44 17 136 100
6 48 8 12 96 52

(d) Packings constructed from the Coxeter-Todd lattice K\2. We use the
complex 6-dimensional version of K \2 defined in Example lOa of the
previous chapter (the "2-base"). This has a norm-doubling map
D:(a,b,c,d,e,f) - (wa+wb,wa-wb, ... ,we+wf,we-wf),

for a, ... ,f E 8, satisfying D2 + D + 2 = 0, which we may regard as

acting on K \2. Construction E produces lattices T/ (K \2) in R12n for
1 ~ n ~ 65 with (from (I 11»
log2 ~ = -3n log23 + 6(an _2a + 1 +a + 2), (115)

where a = [log2n 1. In dimensions 228 to 780 these are denser than any
other lattices constructed in §4, but are inferior to Craig's lattices.
242 Chapter 8

(e) Lattices constructed from a norm-doubling map for A 24 . The Leech

lattice has (at least) two distinct norm doubling maps D. One is
D = 1- i, where i is the element of Aut(A 24) shown in Fig. 6.7, with
i 2 = -1 and T = D- 1 = 112 ([ + j). If A24 is defined by Eq. (134) of
Chap. 4 then the second norm-doubler D' is represented by the symmetric
matrix

3 -1 -1 -1 -1 -1 -1

-1 -3 -1
-1 -1 -3 (116)
4
-1 -1

-1 -1 -3 -1

where in the second row there is a - 3 in position 0, 1's in the positions

that are quadratic residues modulo 23, and -1 's elsewhere. Also
T = (D')-I = lh D'. Both D and D' map the minimal vectors of A24 onto
deep holes of type Ar
(cf. Chap. 23). MDS codes over F 4096 exist for
lengths n satisfying 1 ~ n ~ 4097. Then Construction E (using either of
the maps D) produces lattices 1/(A24) in dimension N = 24n,
1 ~ n ~ 4097, for which the center density satisfies (from (11 I)
log2Q = 12«m - I)n -2m +m + I) (117)

if 2m - 1 ~ n ~ 2m - 1. Some examples are shown in Tables 8.7, 1.3 and

1.4. Up to dimension 4096 these lattices run neck and neck with 1/ (A 12)
and 1/(A I6). For N = 24 n ~ 24· 4097 = 98328, 1/(A 24 ) is a lattice in
RN with density .l satisfying
1
N log 2 .l = -1.2454 ... + f (118)

from (112), where lEI < 12N- 1log 2(N/6). The first term on the right side
of (118) is -lh log2 (48/e1l").

Table 8.7. Center density 0 of N-dimensional lattices 1/ (A 24) obtained

from the Leech lattice (N - 24n).

n N log2 Q n N log2 Q n N log2 Q

1 24 0 10 240 228 341 8184 26712
2 48 12 21 504 696 342 8208 26808
3 72 24 43 1032 1896 683 16392 61608
4 96 48 85 2040 4680 1366 32784 139488
5 120 72 170 4080 11316 2731 65544 311496
6 144 96 171 4104 11400 4097 98328 491832
Algebraic Constructions for Lattices 243

(f) Lattices constructed from a norm-trebling map for A 24. Suppose the
MOG coordinates for a Leech lattice vector

WI W2 W3 W4 W5 W6
XI X2 X3 X4 X5 X6
(119)
YI Y2 Y3 Y4 Y5 Y6
ZI Z2 Z3 Z4 Z5 Z6

are represented by the quaternionic vector

(WI + xli +Yli + ZI k • ...• w6 + X6 i +Y6.i + Z6k) (120)

in 0 6. Then multiplication by the quaternion () = i + j + k is a norm-

trebling endomorphism D of A 24 . Using MDS codes over F312 and
Construction E we obtain lattices 1/3 (A 24) in dimension
N = 24n ~ 24(312+ 1) = 12754608 with
3a+l_l }
log2 fi = 12 (IOg23) { an - 2 +a + 1 , (121)

where a = [IOg3n), and

1 (122)
N log2 d = - 1.434 ... + Eo

These appear to be the densest packings known in dimensions 105 to lOS

(see Table 1.3).
(g) Pac kings constructed from P 4Sq. Our fourth starting point is P 4Sq
(Example 9 of the previous chapter), which has a norm-doubling map D
very similar to (116). If P 4Sq is defined by (69) of Chap. 7 then we may
take D to be the symmetric matrix
5 -1 -1 -1 -1 -1 -1

-1 -5 -1
-1 -1 -5
6
-1 -1
(123)

-1 -1 -5 -1

where in the second row there is a - 5 in position 0, 1's at the quadratic

residues modulo 47, and -1 's elsewhere. Also T = D- I "" VzD. This
produces lattices 1/ (p 4Sq) in dimensions N = 48n for n ~ 8 . lOS with
1
N log2 N = -1.4529 ... + Eo (124)

The 96-dimensional lattice is particularly good (see Table 1.3).

244 Chapter 8

(h) Higher dimensions. The construction may be applied repeatedly to

any of these packings, producing an infinite tree of lattices. We may
derive a lower bound on the density obtained in this way in high
dimensions by the following argument. Let A be any of the lattices
described in this section, with say dimension m = 2j.L, center density
0= 21', and norm-doubling map D. We apply our construction using codes
of length n = 2IJ., obtaining a lattice A' (=La) in dimension m' = 2// with
center density 0' = 21", where, from (111)
j.L' = j.L2IJ., /,' = 2IJ./, + j.L(j.L-2)2IJ. + j.L(j.L+2). (125)

It is simpler to work with the density ~ rather than with 0, so let us define
log2~
11 = ---.
m

(We know from Chap. 1 that for large dimensions m the best packings
satisfy O' 599 ~ 11 ~ I.) Using (18) of Chap. 1, (125) becomes

11' = 11 + 1- L log2 (271" j.L) +0 (1 / J.I..) • (126)

The solution of (125), (126) is 11 (m) = log;m , the smallest value of k for
which log210g2 ... log2 m (with k log2's) is less than 1. In other words
as the dimension N -+ 00, these lattices have density ~ satisfying

(127)

(127) holds for any choice of the initial lattice. The actual rate of growth
is quite slow. For example the 324-dimensional lattice 11 (K 12) has children
with
1
N log2 ~ = -2.006 ... +~ (J 28)

for N ~ 10 51 , the 65544-dimensional lattice 11 (~24) has children with

1
N log2 ~ = -2.2005 ... +~ (129)

for N ~ 10987 °, where ~ in (129) satisfies

I~ I < 32772 1
~ og2 16386'
N
(J 30)

and so on.
Acknowledgements. We thank E. S. Barnes, M. Craig, W. M. Kantor,
H.-G. Quebbemann and 1. G. Thompson for helpful correspondence and
discussions. §§8-11 are based on [Barl5], [Bos3], [Con37].
 9
Bounds for Codes and Sphere Packings
N. J. A. Sloane

This chapter surveys recent work on finding bounds for error-correcting

codes, constant weight codes, spherical codes, sphere packings, and other
packing problems in 2-point-homogeneous spaces. A simplified account is
given of the general machinery developed by Kabatiansky and Levenshtein
for setting up such problems as linear programs. Many other recent
bounds are also described.

1. Introduction
The introduction of the techniques of harmonic analysis and linear
programming into sphere packing and coding theory has resulted in a
considerable strengthening of the classical bounds. Our aim in this chapter
is to describe a derivation of the new bounds from the analytical theory,
which we quote without proof in §2.
The first step towards these bounds was taken by MacWilliams, who
proved the surprising result that if C is a linear code then the weight
distribution of the dual code C· is a certain linear transformation of the
weight distribution of C. (Eq. (50) of Chap. 3, [Mac2], [Mac6, Chap. 5,
Th. 1]. See §2 of Chap. 3 and [Mac6] for terminology from coding
theory,) Of course this implies in particular that this linear transformation
(it is actually a Krawtchouk transform, as we shall see in §3.n of the
weight distribution of C has nonnegative entries.
The next step was taken by Delsarte ([DeI8], [Mac6, Chap. 5, Th. 6])
who showed that the same, Krawtchouk, transform of the weight
distribution of any code, linear or nonlinear, has nonnegative entries (see
Theorem 1 below). This result has had far-reaching consequences. One of
the main problems in coding theory is to determine the function A (n, d),
which is the maximal number of binary vectors of length n that can be
found with the property that any two of the vectors differ in at least d
places. Using his result, Delsarte showed that the value of A (n,d) is
246 Chapter 9

bounded by the solution to a certain linear programming problem.

Although this reformulation has not led to a complete solution of the
problem, it has been very fruitful. One of its merits is its generality. In a
long paper in 1973 Delsarte [DeI9] showed that a number of other
combinatorial problems can also be bounded by linear programs, especially
problems which ask for the largest subset of an association scheme
satisfying certain constraints (see also [Banll], [Delli], [Gabl], [Goe4],
[Sei3], [SI05]). One such question, closely related to the coding problem
already described, is to determine the function A (n, d, w), which gives the
maximum number of vectors containing wi's and n - w O's, and having
the property that any two of the vectors differ in at least d places. Such a
collection of vectors is called a constant weight code. The linear
programming bound for A (n, d, w) is, as we shall see in §3, quite similar
to that for A (n,d).
The most spectacular application of the linear programming method in
coding theory came in 1977 when McEliece, Rodemich, Rumsey and
Welch ([McE4]; see also [McEI], [McE2], [Mac6, Chap. 171) obtained
upper bounds on A (n, d) and A (n, d, w) for large n that were a significant
improvement on the existing bounds, and are stilI the best bounds known.
In order to state their results it is appropriate to define the rate of the best
code to be
I
R (n,d) = - log2 A (n,d),
n

and to examine how this rate behaves as a function of the ratio din when
d and n both increase, for 0 ~ din ~ 1. In fact it is enough to consider
the range 0 ~ d In ~ lh, for it is not difficult to show (using for example
the Plotkin bound (45)) that, in the range V2 ~ d In ~ I, R (n, d) - 0 as
n - 00. The McEliece et al. (or lPL) bound on A (n,d) states that, for
o~ din ~ lh,

R (n,d) < min {J + h (u 2) - h (u 2 + 2du + 2d)} (J)

- 0 .. u .. 1-2d/n n n

as n - 00, where h(x) = H2 ((J - .JI-x )/2), and H 2(x) is the binary
entropy function defined in (72) of Chap. 7. In the range
0.273 < din ~ 0.5 the minimum in (I) is attained at u = I - 2dln, and
the bound simplifies to

R(n,d)~H2[~ _{:[I_:]}1I2], asn-oo. (2)

In the other direction Gilbert proved in 1952 ([Gill], [Mac6, Chap. 17,
Th. 30]) that, for 0 ~ 0 ~ lh, there exists an infinite sequence of codes
with din ~ 0 and rate

R ~I- H2 [ : ), as n - 00. (3)

Bounds for Codes and Sphere Packings 247

o.e
+ 06
:q
c
Ci" 04

0 .2

0.1 0 .2 0 .3 0 .4 0 .5
d/n-
Figure 9.1. Asymptotic bounds on the best codes, showing
R (n ,d) = n- I iog 2 A (n,d) as a function of din as n -- 00.

The bounds (1), (2) and (3) are plotted in Fig. 9.1. For large n the best
codes lie somewhere in the shaded region.
The linear programming method has proved equally effective in
studying codes of shorter length. [Bes3) gives a large number of bounds on
A (n, d) and A (n, d, w) that were obtained in this way for codes of length
n ~ 24. Some examples will be given in §3.3. Table 9.1 shows the
present state of knowledge about small values of A (n,d). Note that it is
enough to consider even values of d ~ 4, since A (n, 21 - 1) = A (n + 1,21),
and A (n, 2) = 2n- 1 [Mac6, Chap. 21. The unmarked entries in the table
are taken from [Bes3) or [Mac6). References to other tables will be found
in §2.I of Chap. 3. See also [Bes2J, [Roo1).
The next major advance was the paper by Kabatiansky and Levenshtein
[KabI), which succeeded in establishing bounds for sphere packings
analogous to the JPL bounds for codes. In order to obtain their bounds
Kabatiansky and Levenshtein extended the linear programming approach
still further to include a number of packing problems on Riemannian
manifolds. Their paper contains many new bounds, the most important
being an upper bound on the highest density .:In of any packing of equal
spheres in n-dimensional Euclidean space Rn. The classical results of
Minkowski and Blichfeldt (see §1.5 of Chap. 1, [Rog7]) state that
n n +2
-n <
-
< - -2 + log2 -2
log 2 .:l n _ -' for all n.

Earlier improvements to the upper bound had been obtained by Rankin

[Ran 1), Rogers (Eq. (39), (40) of Chap. 1, [Rog2), [Rog7]) Sidel'nikov
[Sid3), and Levenshtein (Eq. (42) of Chap. 1, [Lev4J, [Lev7]).
(Sidel'nikov [Sid 1J, [Sid2) and Levenshtein [Lev3) had also obtained new
upper bounds to A (n, d), but their results were superseded by the JPL
bound. See also [Lev2], [LevSJ, [Lev6l.) Kabatiansky and Levenshtein
showed that the upper bound in (4) can be replaced by
logz.:ln ~ - 0.599 n, as n -+ 00 . (5)
248 Chapter 9

Table 9.!. The best codes: bounds for A (n,d) (a: [Besl], b: [Hon3],
c: [Rom I], d: H. O. Hiimiiliiinen, personal communication).
n d=4 d=6 d=8 d=1O
6 4 2 I I

7 8 2 I 1
8 16 2 2 1
9 20 4 2 1
10 40° 6 2 2
11 72-79° 12 2 2
12 144-158° 24 4 2
13 256 32 4 2
14 512 64 8 2
15 1024 128 16 4
16 2048 256 32 4
17 2720 c -3276 256-340 36-37 6
18 5248 d -6552 512-680 64-74 10
19 10496d -13104 1024-1288 128-144 20
20 20480°-26208 2048-2372 256-279 40
21 36864-43690 2560-4096 512 40-48 b
22 73728-87380 4096-6942 1024 48-88 b
23 147456-173784 8192-13774 2048 64-150
24 294912-344636 16384-24106 4096 128-280

Their technique is to consider, instead of ~n itself, the related quantity

A (n, (J). We repeat the definition from §2.3 of Chap. 1. Let 0 n denote
the [(n - 1) -dimensionaI1 unit sphere in Rn :
On = {x = (Xb ... , xn) E R n : xf + ... + x; = I} .
A spherical code C of dimension n, size M and minimal angle (J is a set of
M points of 0 n with the property that
x . y ::;;; cos (J for x, y E C, x -;&. y .

A (n,(J) is the maximal size of such a code. Kabatiansky and Levenshtein

prove that, for 0 < (J < 7r /2, and large n,
Bounds for Codes and Sphere Packings 249

1 1 + sin fJ 1 + sin fJ
- log2 A (n ,fJ)
n
< 2 sin (J log2 2 sin (J
1 - sin (J 1 - sin (J
log2
2 sin (J 2 sin (J
and use this to derive (5). The main goal of the present chapter is to give
a simplified account of Kabatiansky and Levenshtein's general theory and
to sketch how their bounds are obtained.
The linear programming method has also proved successful in low
dimensions. As we shall see in Chap. 13, it can be used to solve the kissing
number problem in 8 and 24 dimensions.
The general theme of this chapter is the problem of finding nice
arrangements of points in various spaces. The four examples we are
primarily concerned with are the following.
Example El. The space is "Hamming space", i.e. the set F~ = {o,l}n of
binary vectors of length n, and a binary code is a subset of F~ (see §2.1 of
Chap. 3).
Example E2. The space is "Johnson space", i.e. the subset F~'w ~ F~
consisting of all vectors containing wI's and n - w O's. A subset C of
F~' w is a constant weight code, and A (n, d, w) is the greatest cardinality of
a constant weight code in which distinct codewords have Hamming
distance at least d apart. (The name Johnson space was chosen because
S. M. Johnson was the first to extensively study A (n, d, w) - see [Mac6J.)

°
Example E3. The space is the "spherical space" On, and a finite subset of
n is a spherical code. In this example we always assume n ~ 3, the
cases n ~ 2 being special (and trivia!).
Example E4. Finally there is Euclidean space Rn. A subset ceRn such
that N (x - y) ~ 4p2 for x, y E C, x ¢ y, is a sphere packing of radius
p.
In §2 we quote the machinery needed from harmonic analysis, the main
purpose being to construct certain functions, the zonal spherical functions,
associated with each of the packing problems to be considered. §3 shows
how the linear programming bounds (Theorems 2-4) are simply derived
from the results in §2. A short proof is also given of the "mean inequality"
for codes (Theorem 5). §§3.3-3.5 given many applications of these bounds
to Examples El, E2 and E3. Table 9.2 for instance gives bounds on the
number of spheres that can touch a pair of touching spheres of the same
size. The crucial link between spherical codes (Example E3) and sphere
packings (E4) is provided by Theorem 6. Finally §4 gives a brief account
of other recent bounds. As background references [DymI1, [Mac1],
[MiI3], [TerIl and [Vill] are particularly recommended.

2. Zonal spherical functions

In this section we prepare the way for the linear programming bounds by
constructing a set of functions, called zonal spherical junctions, for each of
the problems to be considered.
250 Chapter 9

Table 9.2. How many spheres can touch two spheres? Bounds on
A (n,cos-II/3), from [Assn The lower bound for n = 4 is taken from
[MacOl.

n A (n, cos-Il/3) n A (n, cos- l 1/3)

I 2 6 32-37
2 5 7 56
3 9 8 64-78
4 14-15 9 96-107
5 20-24 10 ~I46

2.1 The 2-point-homogeneous spaces. To start with we choose a group G,

whose properties will determine everything that follows. Although
Kabatiansky and Levenshtein's theory is somewhat more general, we shall
restrict ourselves here to two classes of compact groups: finite groups and
connected compact groups. We shall use Examples EI, E2 and E3 of the
previous section as illustrations (these are the settings for error-correcting
codes, constant weight codes and spherical codes respectively). The
appropriate groups G are as follows. In Example EI we take G to be the
automorphism group of the n -dimensional cube, called the hyper-
octahedral group, or [3n-2, 41 in Coxeter's notation (see [Cox28]). It has
order 2n·n!. (E2) G is the symmetric group Sn, of order n!. (E3) G is the
special orthogonal group SO(n), i.e. the group of isometries of fin with
determinant 1.
Next we need a 2-point-homogeneous G-space M [HeI6, p. 2891. This
implies three things. (i) M is a set on which G acts, so that there is an
element g (x) E M for every g E G and x E M, and furthermore
g2(gl(X» = (g2gl)(X) and I(x) = x for all x, where gj, g2 are arbitrary
elements of G and lEG is the identity element. (ii) M is a metric space
with a distance function T defined on it. (iii) T is strongly invariant under
G: for any x,x',y,y' E M, dx,y) = dx',y') if and only if there is an
element g E G such that g(x) = x' and g(y) = y'. We let T ~ C denote
the range of T.
Assumption (iii) certainly implies that G acts transitively on M. In
fact G is as close to being doubly transitive as it can be without changing
the distances T (x,y) between the points of M.
Let H be the subgroup of G that fixes a particular element Xo E M
(any x 0 will do, since G is transitive). Then M can be identified with the
space G/H of left cosets gH (with the point Xo corresponding to H itself).
Examples (continued). (EI) For M we take the set of vertices of the n-
dimensional cube, (_I)x: = «-lVI, ... , (-IV"» for x = (XI • . . . •
x n ) E Fq}, which of course we will often identify with Fq. We let
d (- I) x, (- I)Y) be the Hamming distance between x and y, so that
Bounds for Codes and Sphere Packings 251

T = {O, I, ... ,n}. A typical element of G has the form g == (J7r, and
consists of a permutation 7r E Sn followed by a sign change (J - diag
{(_I)a t , • • • , (_I)an), aj ,. 0 or 1. The subgroup H fixing the vertex
Xo = (1, I, ... , I) is Sn itself, and the natural map G - G /H .. M sends
g to (_I)a.
(E2) M = Johnson space Fi"w with w ~ n/2, dx,y) ., 'h Hamming
distance between x and y, T = {O,I, ... , w}, H" Sn-w x Sw' (E3)
M .. On' dx,y) = X'y, T "'" [-1,11, H '" SO(n-I). On this case
cos- 1 (x' y) is the metric on M, but it is simpler to work with x .y.)
The final assumption we shall make about G is that (a) if G is infinite,
then M is a connected Riemannian manifold and T is a constant times the
natural distance on the manifold, and (b) if G is finite, and
do = min dx,y) for x,y E M, x ¢ y, then M has the structure of a
graph in which x is adjacent to y if and only if T (x ,y) ... do, and
furthermore T is a constant times the natural distance in the graph.
These assumptions are quite restrictive. In fact if G is infinite then
Wang ([Wan2]; see also [HeI6, p. 5351, [Tit 11, [Woll, Th. 8.12.2]) has
proved that the following are the only possibilities for the space M:
(a) the sphere On,
(b) real projective space pn(R) == SO(n+I)/O(n),
(c) complex projective space pn(c) = SU(n+I)/U(n),
(d) quaternionic projective space pn(H) = Sp(n + I)/(Sp(n) x Sp(l),
(e) the Cayley projective plane p2(Cay).
These are the compact Riemannian symmetric spaces of rank one, and
(a,H) is an example of a Gelfand pair ([BouO], [Gell], [Kra2], [LeU],
[Let2]) .
The finite 2-point-homogeneous spaces M have not yet been completely
classified, although they have been the subject of an enormous amount of
research in connection with distance-transitive graphs, strongly regular
graphs, two-weight codes and association schemes (see [BanI I], [BanI21,
[Big2], [Big3a], [Big4], [Big5], [Buml], [Call], [Cam2], [Coin, [Coi2],
[Curl], [DeI9], [Dell 11, [Dell21, [Dell4], [Dun11-[Dun9], [Goe4], [Gor11,
[Hub11, [Kan51, [LeolO], [Leoll], [LinI21, [McKI], [Sei11-[Sei3], [SI05],
[Smi11-[Smi3], [Sta41, [Sta8]). Some of the most important examples are:
(f) Hamming space F~, or more generally the space F; of n-tuples from
Fq , with dx,y) = Hamming distance,
(g) Johnson space F~' w as in Example E2,
(h) the space whose elements x, y , ... are k -dimensional subspaces of F;,
where k ~ n/2, with dx,y) == rank(x n y),
(j) the space of maximal totally isotropic subs paces of an orthogonal,
unitary, or symplectic geometry over a finite field, with dx,y) ==
rank (x n y), and
252 Chapter 9

(j) the spaces of bilinear, alternating bilinear, or hermitian forms over a

finite field, with T (x,y) = rank (x - y),
but this list is by no means complete. It is remarkable that the bounds
obtained in §3 apply to subsets of anyone of these spaces.
2.2 Representations of G. Since G is compact, it has a unique normalized
measure p., Haar measure, which is invariant under G (see for example
[HaI6]). This induces a unique invariant measure on M, which will also be
denoted by p.. We assume that p. is normalized so that p. (M) = 1. In
Example (EO, p.(x) = rn for each x E F~, and in (E2) p.(x) = 1/(;)
for each x E F~'w. In (E3) p. is normalized Lebesgue measure on the
sphere, with p. (On) = 1.
Let L 2(G) denote the vector space of complex-valued functions u on G
satisfying
f
G
lu(g)1 2 dp.(g) < 00,

with the inner product

(u I> u 2) = f
G
u I (g) U2 (g) d p. (g).

Those u E L 2(G) that are constant on left cosets of H can be regarded as

belonging to L 2(M), which is defined similarly and has the inner product
(Ul> U2) = f
M
Ul(X) U2(X) dp.(x).

We shall study how G acts on the functions in L2(G) and L2(M).

The left regular representation of G on L 2(G) is given by
g --+R(g), g E G,

where

Then R (g) is a unitary representation with respect to our scalar product,

i.e. satisfies

for all UJ, U2 E L2(G) (see [Vill, p. 28]).

By the Peter-Weyl theorem ([Coil], [Coi2], [Grol1, [Vill], [War3])

the space L 2(G) decomposes into a countable direct sum of mutually
orthogonal subs paces {v(m,O, m = 0,1,2, ... ; i = 1, 2, ... ,d'm} with the
following properties. v(m,O is a vector space of continuous functions, of
dimension d'm, and affords an irreducible unitary representation of G. The
spaces v(m,]), v(m,2), ... , V(m,d',J are isomorphic and afford equivalent
Bounds for Codes and Sphere Packings 253

representations of G. Finally every irreducible unitary representation of G

is obtained in this way.
To obtain the decomposition of L 2(M) we must consider only those
irreducible representations of G which are of "class I with respect to H",
i.e. those representations p(g) on a space V(m,il which have the property
that there is a non zero vector a E V(m,il with p(h)a = a for all h E H
(see [Vin], pp.29, 52]). For the groups we are considering it is known
(cf. [Coi, Th. 3.51, [Coi2, Chap. 2, §41, [Ganl, Prop. 3.41, [Hig21, [Trai])
that if such an invariant vector a exists then it is unique apart from a scale
factor (a subgroup H with this property is called "massive").
Then (see [Vin, p. 54]) e(M) decomposes into a countable direct sum
of mutually orthogonal subspaces {V(k), k = O, 1, ... }. Each V(k) is one of
the spaces v(m,il mentioned above, and is a space of continuous functions
of dimension d k = d'm' V(k) affords an irreducible representation P<k)(g)
of G which is of class 1 with respect to H, and each such representation
arises just once in the decomposition, the V(k) being mutually
nonisomorphic. We take V(O) to be the space of constant functions, and
do = 1.
Let ref: i = 1, ... , dd be an orthonormal basis for V(k), chosen so that
ef is an invariant vector a. Let P<k) (g) be represented with respect to this
basis by the matrix (Pi)k) (g) ), where
dk
p(k)(g) ef = ~ pJk)(g)ej.
j-l

It is clear that pff> (h) = I for h E H. The functions p/f) (g) are constant
on left cosets of H, and can be written as p/p(x), x E M [Vill, p. 53].
Then
.p; Pl(f> (x), ... , .p; PJZl (x)
is an orthonormal basis for V(k) [Coi2, Th. 1.10]. Any element of L 2(M)
has a "Fourier expansion"
dk
u (x) = ~ ~ cN) p/p(x), x E M, (7)
k i-l

where the series converges in the norm of L 2 (M), and

cN) = dk f u(x) p/f)(x)dJJ,(x) (8)

M

(see [Viii, p. 54]).

2.3 Zonal spherical functions. To each space V(k) we associate the
function
dk __
h(x,y) = d k ~ p/f)(x) p/f)(y), x, Y E M, (9)
i-l
254 Chapter 9

which has a number of interesting properties.

(i) It is immediate from the definition that J k (x ,y) is positive definite, in
the sense that, for any u E L 2(M),

II
MM
h(x,y) u(x)iiTyTdjt(x) djt(y) ~ 0, (10)

or alternatively for any integer n, any points x I, . . . ,Xn E M, and any

complex numbers a I, . . . , an the relation
n n
~ ~ h(Xi, x) aJij ~ 0
i-I j-I

holds. (Conditions (10) and (11) are equivalent - see for example
Bochner [Bocl].)
(in Since the representation p(g) afforded by V(k) is unitary, it also
follows from (9) that J k (x ,y) is independent of the particular choice of
basis for V(k), and furthermore that

for all x ,y E M, g E G, i.e. J k (x ,y) is invariant under G, and so

depends only on T (x ,y ). Thus we can write

where
dk _ __
<l>k(r(X,y)) = ~ pH)(x) pH)(y). (14)
i-I

The function <I> k (r), which is a continuous function of t defined on the set
T (the range of T), is called the zonal spherical function associated with
V(k) (see [Coi2], [Dun?], [ErdJ], [Ganl], [Miill], [StalJ], [ViiI]). If we
let r(x, x) = TO, then

<l>k(TO) =.~ IpH) 1 2 = I, (15)

I-I

since {pH)L is the first column of a unitary matrix. Also <l>o(r) = I.

(iii) From (9), h(x,y) = Jk(y, x), and (13) and r(x,y) = T(y,X) then
imply that h (x ,y) and <I> k (t) are real.
(iv) The orthogonality of the spaces V(k) implies that the <l>k (t) are
orthogonal and satisfy

where jl is the measure on T given by

jl (A) = jt {x E M: r(xo,x) E A}.
Bounds for Codes and Sphere Packings 255

(v) An alternative expression for <l>k (t) may be obtained as follows ([Coi2,
p. 3S], [Yill, p. 30)). By definition,
p[fl(g) = (p(k) (g)a, a), (17)

where a is the unit vector in V(k) fixed by H. Furthermore it is easy to

see that ([Yill, p. 30))
(1S)

for any hI> h 2 E H, g E G. Thus p[P is constant on the double cosets

HgH. If gx and gy are any elements of G such that

gx(xo) = x, gy(xo) = y,

then
(I 9)

In the literature pa)(g), Fk(x,y) and <l>k(t} are all referred to as zonal
spherical functions (or kernels). The name comes from the identification
of points x E M with left cosets gH. Then Hx represents a "sphere" in M
centered at Xo and passing through x, and is identified with the double
coset HgH. From (1S) p{P(x) is constant on these spheres.
Example EI (see [Dunl], [DunS)). L 2 (M) consists of the complex-valued
functions on the vertices of the n-cube, and has a basis consisting of the
monomials c/>~l ... c/>;', where (al •...• an) E Fg and c/>i is the ith
coordinate function (with c/>l = c/>i)' G permutes and/or negates the c/>i
while H just permutes them. V(k) consists of the polynomials of degree k
in c/>l •...• c/>n, with dk = [Z], for k = 0,1, .... n. The fixed vector
a = ef in V(k) is ~ c/> ~l ... c/>;', the sum being over all (a 1, . . . • an) with
weight k (i.e. which contain k ones). Then <I> k (t) is a special case of a
Krawtchouk polynomial (see [Mac61, [Szel)):

<l>k(t} = Kk(t,n) = [Zr 1

.~
J-O
(-I)j [;] [Z =;), (20)

for k = 0, I ..... n.
Example E2 (see [De1l2], [Dun2)). L 2(M) consists of the complex-valued
functions on the w-subsets of an n-set, and decomposes into orthogonal
subspaces {V(k), 0 ~ k
~ w}. Vk has dimension d k = ~ [Z]- [k I]'
and is described explicitly in [De1l2] (where it is denoted by Harm (k».
Vk affords the irreducible representation of Sn corresponding to the two-
rowed Young tableau [n - k, k] (cf. [Mil3, §4.2)). The corresponding
zonal spherical function is

where
256 Chapter 9

k
Qk(t} = ~ (_1)i (22)
i-O

is a particular kind of Hahn polynomial [Kar2].

Example E3 (the most extensively studied case: see for example [Coi2],
[Dell 6], [Dun71, [Erdl], [Hobl], [Kool], [Miill], [Vill, Chap. IX)).
L 2(On) decomposes into an infinite direct sum of orthogonal subspaces
(V(k), k = 0,1, ... J. V(k) consists of all functions in L2(On) which are
represented by homogeneous polynomials f (x) • ...• x n ) of total degree k
in variables x) • ...• Xn that satisfy Laplace's equation

\1 2f = ft + ... + ft = 0.
aX[ ax;
V(k) is usually denoted by Harm(k) and its elements are called spherical
harmonics. It has dimension
d =
k
[n +k-
n-l
1) _ [n +nk- l- 3).
The corresponding zonal spherical function is a Gegenbauer or
ultraspherical polynomial, which is a special case of a Jacobi polynomial.
To avoid confusion of notation we write it as a Jacobi polynomial, using
the standard notation of [Abrl]:
pla ,f3l(t)
<I> k (t) = [k t a) , k = 0, I ..... (23)

where a = {3 = (n-3)/2. (The normalization is determined by (15),

remembering that TO = dx ,x) = X· x = U
Many of the other 2-point-homogeneous spaces that we listed in Section
2.1 have also been studied in detail - see [Coi2], [DeI9], [DellI], [DeIl2],
[DeIl4], [Dun2]-[Dun6], [Kool]-[Ko03], [Sta4]-[Stal11, [Vi111. For the
projective spaces pn (R), pn (C), pn (H) and p2(Cay) the zonal spherical
functions are Jacobi polynomials
pla ,fJ)
[k t a)
(2t 2 - 1)
<I> k (t) = ,
where a = (n - 2)/2 and {3 = - V2, 0, I and 3 respectively. For the finite
2-point-homogeneous spaces many (but not all) of the zonal spherical
functions belong to a general family found by Askey and Wilson (see
[Ask4], [LeoIO], [Leoll)).
2.4 Positive-definite degenerate kernels. We have seen that the zonal
spherical function h (x ,y) is a continuous, positive-definite, degenerate
(because of (9» kernel or p.d.k. on M. It is also invariant under G by
Bounds for Codes and Sphere Packings 257

(I 2). It is easy to see that the product of two p.d.k.'s is a p.d.k., and a
sum of p.d.k.'s with positive coefficients is a p.d.k. The p.d.k.'s that are
given by the zonal spherical functions are called elementary, and clearly
any sum
F(x, y) = ~ Ak It (x, y) (25)
k

with Ak ~ 0 and ~k Ak d k < 00 of elementary p.d.k.'s is also an invariant

p.d.k. It is surprising that the converse holds: any invariant p.d.k. can be
written in the form (25) as a sum of elementary p.d.k.'s with positive
coefficients. This was first proved for Example E3 by Schoenberg [Sch5],
and generalized to include all the groups considered here and many others
by Bochner [Bocl] and later writers [BinI], [Gel1), [KreI] (see also
[Kabl, Th. 2]).
It is a corollary that the product of any two zonal spherical functions
4>; (t) and 4>j (t) can be expanded as a sum
4>; (t}4>j (t) = ~ Cijk 4> k (t) (26)
k

with C;jk ~ 0 ([Koo3]). Of course this applies to those Krawtchouk, Hahn

and Jacobi polynomials given above. Such expansions with positive
coefficients have been extensively studied [Askl], [Ask2], [Gasll

3. The linear programming bounds

3.1 Codes and tbeir distance distributions. Having constructed the zonal
spherical functions <P k (t) we can now proceed to derive the linear
programming bounds. We shall study finite subsets C of any of the 2-
point-homogeneous spaces M; such subsets will be called codes, since they
generalize the notions of error-correcting codes and spherical codes. Of
course we are not interested in arbitrary subsets, but only those in which
the distances between vectors are restricted in some way. We assume that
a subset S ~ T has been specified, and that we wish to find codes C for
which r (x, y) E S for all x, y E C, x ¢y; such a code is called an S-
code, and the largest cardinality IC I of an S -code will be denoted by
A(M, S).
Example El. If we take S = {d, d + 1, ... ,n} then an S -code is an
ordinary error-correcting code of minimal distance d, and
A(M,S)=A(n,d). (E2)If S={o,o+l, ... ,n) an S-code is a
constant weight code with minimal distance 20, and
A (Fg· 2 , S) = A (n,e, w). (E3) If S = [-1, cosO), an S-code is a spherical
code with minimal angle 0, and A (fin, S) = A (n, 0).
The distance distribution {at} of any code C is defined by

at = I~I . (numberoforderedpairsx,y E Cwithr(x,y) =t).

It is immediate that
258 Chapter 9

a TO = 1, (27)

at ~ 0 for t E T, (28)

~ at = lei, (29)
t E T

but there are some additional inequalities.

Theorem 1. Let {3k = Iel- I . ~ at <l>k (t), k = 0, 1, ... , denote the
t
"trallsform" of the distance distribution by the zonal spherical functions.
Theil (3 k ~ 0 for all k.
Proof

~ 0 by (J]).

If G (and M) are finite this result is due to Delsarte [DeI8], [DeI9]; in

the infinite case it can be found (implicitly or explicitly) in [Del16l,
[KabIl, [Llol1, [OdIS]. See also [DunS]. In Example El {{3d is the
Krawtchouk transform of {atL If e is a linear code then {at} is its weight
distribution and by the MacWilliams identity (Eq. (50) of Chap. 3) {{3 t} is
the weight distribution of the dual code.
3.2 The linear programming bounds. Theorem 1 and (27)-(29) make it
possible to regard the problem of bounding A (M, S) as a linear
programming problem:
Primal problem: Choose a natural number s, a subset {r I, . . . , T s} of S,
and real numbers aT], ... , a Ts so as to

s
maximize ~ a Tj (30)
i-I

subject to
a Tj ~ 0, i = 1, ... , s,
s
~ aTj<l>k(Ti) ~ -1, k =0, 1, ....
i-I

This is a linear programming problem with perhaps infinitely many

unknowns at (t E S) and constraints (32). If e is an S -code then its
distance distribution {at} certainly satisfies the inequalities (3]), (32). So
if the maximal value of the sum (30) that can be attained (for any choice
of sand T I, . • . , T s> is A', then A (M, S) ~ 1 + A'. (The extra 1 arises
because the term a To = 1 does not occur in (30).)

The dual problem is as always more convenient, and reads as follows

(cf. [Dun], [Dun9], [SimO]).
Bounds for Codes and Sphere Packings 259

Dual problem. Choose a natural number N and real numbers f I, . . . ,fN

so as to
N
minimize ~ h (33)
k -I

subject to
h ~ 0, k = 1, ... ,N, (34)
N
~ h tfl k (t) ~ - 1, for t E S. (35)
k -I

Thus we have proved:

Theorem 2. If A * is the optimal solution to either of the primal or dual
problems then A (M, S) ~ l+A*.
The reason for preferring the dual problem is that by the duality theorem
any feasible solution to the dual problem is an upper bound to the optimal
solution of the primal problem, or in other words:
Theorem 3. If f I, . . . ,fN satisfy (34) and (35) then A (M, S)
~ 1 + fl + ... + fN.
This can be restated in an equivalent form which is sometimes easier to
use.
Theorem 4. Suppose the space M, the zonal spherical functions tfl k (t).
and a subset S ~ T. are given. and we wish to bound the cardinality
A (M, S) of the largest subset C of M such that T (x ,y) E S for all
x,y E C, x ¢ y. If we can find a linear combination
N
f(t} = ~ htflk (t), (36)
k-O

for some N. which satisfies f 0 > 0, h ~ 0 for k = I, ... , N, and

f (t} ~ 0 for t E S, then
A (M, S) ~ f (TO) /f o. (37)

Theorem 2, 3 and 4 are the general linear programming bounds we were

seeking.
As a consequence of Theorem 1 any code C must satisfy further
inequalities, the so-called mean inequalities. Let F (x ,y) be any invariant
p.d.k. on M. By §2.4 we can write F(x,y) = f(dx,y», where
f(t} = ~htflk(t}, h ~ 0,
k

and the sum converges for all t E T. The mean of F over C and over M
are defined by

F(C) = -I1-12 ~ F(x,y), (38)

C X,Y EC
260 Chapter 9

F(M) = ff F(x,y) d/l(x) d/l(Y) (39)

MM

= f
M
F(x,y) d/l(x), for any y E M,

since F (x,y) only depends on dx,y)' Thus F (M) = 10, using (! 6) and
<l>o{t) = l.
Theorem 5 (The mean inequality [Kahl1. [LevS], [Sid3]). For any
invariant p.d.k. F we have

F(C) ~ F(M) (40)

or equivalently

1
2 ~ I(dx,y» ~ 10.
C1 x,y E C

Proof From Theorem I, fh ~ Ok,O' Multiplying by Ik and summing on

k we obtain (41),
This gives an immediate proof of Theorem 4, avoiding the use of linear
programming.
Second prool 01 Theorem 4. Let C be any S -code. From Theorem S
1 .
10 = F(M) ~ I(C) = -I-1 2 ~ F(x,y)
C x,y E C

~
1
x TCf2
F(x, x) =
1
fc
1(70)' TCT
Note that the only properties of zonal spherical functions used to prove
Theorems 2-4 are that they are positive-definite kernels invariant under G,
The theorem quoted in §2.4, that the most general continuous functions
with these properties are positive sums of zonal spherical functions, (a)
justifies our restriction to functions of the form (36), and (b) is used to
derive the mean inequality and therefore in the second proof of Theorem 4.
In general the value of the linear programming bound 1 + A' (see
Theorem 2) is unknown. Nor is it known in general how close the bound
comes to the true value of A (M, S). For small problems it is usually very
close and occasionally does give the exact value of A (M, S)
3.3 Bounds for error-correcting codes. In Example E1 there are just n+1
distinct <P k 's, and so we may take N = n in (33) and (36) (and similarly
in all the cases where G and M are finite). Theorem 4 now states that if
n
I (t) = ~ !k Kk (t ; n) (42)
k - 0
Bounds for Codes and Sphere Pac kings 261

is a polynomial satisfying fo > 0, fk ~ {) (k = 1, ... ,n), and f(t) ~ 0

for t = d, d + 1, ... ,n, then A (n,d) ~ f(O) /fo. We shall give a
number of illustrations.
(j) For n = 8, d = 4 we can take
f(t) = l/2(t -4) (t -8) = Ko (t; 8) + 3K l (t; 8) + 7K 2(t; 8) (43)

and deduce that A (8, 4) ~ 16. On the other hand the Hamming code Jffs
(§2.4.2 of Chap. 3) shows that A (8,4) ~ 16. Thus A (8,4) = 16. The
code is self-dual and the nonzero codewords have weights 4 and 8, the
zeros of f(t).

(jj) The technique suggested by the first example also works in some other
cases. For example the Golay code C(j 24 implies that A (24, 8) ~ 4096.
The lowest degree polynomial that satisfies the constraints and vanishes at
the nonzero weights of the dual code is

f(t) = _1_ (t-8) (t-12)2 (t-16)2 (t-24) (44)

1728
= Ko(t·, 24) + ... + 168245
192
K (t· 24)
6, ,

and proves that A (24,8) = 4096.

(iii) We shall prove that the Nadler code is optimal [Bes3). This is a
nonlinear code containing 32 codewords of length 13 and minimal distance
6; thus A (13, 6) ~ 32. In this case there is no polynomial f (t) that we
can use in Theorem 4 to prove that A (13,6) = 32, so we shall apply
Theorem 2 instead. Suppose C is an optimal code, containing A (13,6)
codewords. There is no loss of generality in assuming that the distances
between the codewords of C are even (if not, delete one coordinate and add
an overall parity check). Let at(u) be the number of codewords in C at
distance t from a codeword u E C; then
1
at = -ici ~ at(u).
uEC

Also ao = 1 and the remaining at's are zero except perhaps for a6, ag, a 10
and a12' The primal problem is to maximize a6 + ag + alO + a12 subject
to ai ~ 0 and

a6 - 3ag - 7alO - 11 al2 ~ + [?),

- 6a 6 - 2ag + 18alO + 54al2 ~ - [Ii),
-6a6 + 14ag - 14a 10 - 154a12 ~ - (11),

15a6 - Sag - 25alO + 275a12 ~ - [11),

262 Chapter 9

15a6 - 25ag + 63alO - 297a12 ~ [1/),

-20a6 + 20ag - 36alO + 132a12 ~ - [Ii),

(the remammg inequalities (32) being redundant). Unfortunately the
solution only gives A (13, 6) ~ 40. However there is an additional
inequality that we can impose. We must certainly have
a12(u) ~ A (13,6, 12) = A (13,6, 1) = 1, for all u E C, and alO(u)
~ A (13,6,10) = A (13,6,3) = 4. Furthermore these can be combined.
For if a12(u) = 1 then alO(u) = 0, so alO(u) + 4 a12(u) ~ 4, and
averaging over u E C gives
alO+4a12 ~ 4.

Solving the primal problem by the simplex method with this additional
inequality gives A (13,6) ~ 32. Therefore A (13,6) = 32 and the Nadler
code is optimal. By analyzing the solution to this linear program in more
detail, Goethals [Goe31 has shown that the code is unique.
(iv) The best available bounds on A (n,d) for n ~ 24 and d ~ 10 are
given in Table 9.1; many of the upper bounds were obtained from Theorem
2 and the simplex algorithm, with the addition of extra inequalities (as in
the previous example) whenever possible.
(v) Returning to Theorem 4, the best choice for a linear polynomial is
f(t) = Ko(t;n) + n K 1(t;n)/(2d-n) = 2(d-T)/(2d-n), for 2d > n.
This satisfies the constraints of the theorem and proves that
2d
A (n,d) ~ -d--' for 2d > n.
2 -n

This is the Plotkin bound (see [Mac6, Chap. 2, Th. 1]). By a

combinatorial argument (see [Mac6, p. 41]) it may be improved to

A(n,d)~2[2d~nl (45)

for d even and 2d > n. Similarly one can show that

A (2d,d) ~ 4d, for d even. (46)

Provided Hadamard matrices of the appropriate orders exist, Levcnshtein

has shown that equality holds in (45) and (46) ([Levi], [Mac6, Chap. 2,
Th.8]). Thus A (n,d) is essentially known for n ~ 2d.
(vi) For n slightly greater than 2d McEliece (see [Mac6, Chap. 17, Th.
38]) has obtained bounds on A (n,d) using Theorem 4. Tietavainen [Tiell
has found considerably stronger bounds using analytic arguments and some
of the inequalities of Theorem 1. Other bounds on A (n, d) obtained by
linear programming can be found in [Besll-[Bes31.
Bounds for Codes and Sphere Packings 263

(vij) As we have already discussed in Section 1, for large nand d the best
upper bound on A (n, d) presently known is the JPL bound. The simpler
part, given in (2), is obtained from Theorem 4 by using a polynomial of
the form

f(r) = _1_ {Kk+l(t; n) Kk(a; n) - Kk(t; n) Kk+l(a; n)J2 (47)

a - t

for suitably chosen values of a and k. The details can be found in

[McE2], [McE4] or [Mac6, Chap. 171. Equation (26) plays a crucial role.
The more complicated part of their bound will be mentioned in the next
section. The JPL bounds have been generalized to tree codes in [Aall1.

3.4 Bounds for constant-weight codes. Example E2, continued. The

function A (n, d , w) is important in its own right, and because it can be
used to derive additional inequalities satisfied by the weight distribution of
a code, as we have seen in example (ij) of the previous section. Tables of
A (n,d,w) for n :::;; 24 and d :::;; 10 are given in [Bes3], and many of the
upper bounds there were obtained using Theorem 2 and the simplex
method (again with the addition of extra inequalities whenever possible).
The best tables of A (n, d, w) presently available may be found in [Bes3] or
[Gra2], but see also [Besl], [BroS], [Br09], [Coil], [Con411, [Hon2]-
[HonS], [Kibl1, [KlrPl1, [Zinl].
The more complicated part of the JPL bound, Eq. (1), is obtained by
using Theorem 4 to bound A (n, d , w) and converting this to a bound for
A (n, d) via Elias's inequality

;::;: 2n A(n, 20, w)

A(n, 2u") "'" .. 0 --- --- (4S)
, lor "'" w "'" n
(~ )
(see [BasI], [Bas2], [McE2], [Mac6, Chap. 17, Th. 33]). For the details
of the proof of (1) see [McE21.

3.5 Bounds for spherical codes and sphere packings. Example E3,
continued. Theorem 4 now takes the following form: if

f(t) = ~!k
N

k-O
p1a,a)(t), CI' = -=--,
n 3
2
(49)

is a polynomial satisfying fo > O,!k ~ 0 (k = 1, ... ,N), and f(t) :::;; 0

for -1 :::;; t :::;; cos 0, then

A(n,o):::;;f;~). (50)

(i) The kissing number problem. This result will be used in Chap. 13 to
obtain bounds on the kissing number problem, the case A (n, 7r /3), and in
Chap. 14 to prove that certain arrangements of spheres achieving these
bounds are unique.
264 Chapter 9

(ii) Tables of bounds on A (n, () for some other values of () and for certain
packing problems in the projective spaces pn (R), pn (C) and pn (H) have
been calculated in [Assll. For example Table 9.2 gives some bounds on
A (n, cos- 1 1/3), which is the maximum number of nonoverlapping spheres
that can touch a pair of touching spheres in Rn + 1 (Chap. 14, Th. 1). E.g.
five billiard balls can be arranged so that they are all in contact with two
touching billiard balls, implying A (2, cos- 1 1/3) = 5.
The values A (7, cos- 1 1/3) = 56 and A (23, cos- 1 1/3) = 4600
correspond to arrangements of spheres in the E 8 and Leech lattices, and
were independently found by Levenshtein [Lev7]. In Chap. 14 it is proved
that the corresponding spherical codes are unique.

(iii) Kabatiansky and Levenshtein [Kabll show that the best linear and
quadratic polynomials are f (t) = t - sand f (t) = (t - s) (t + I), where
s = cos (). The corresponding bounds are

A (n, () ~ [co~:s ~ 1 ] , if cos () < 0,

A (n, () ~ [2n (I - cos () ],

I-ncos()
if cos () < I/n. (52)

Actually the results in [Kab 1] apply to any of the examples in which M is

a continuous manifold (see §2. 1), but in this and the following subsection
we shall just state their bounds for the case when M is the unit sphere On.
Just as in the coding case (see §3.3v) the bound given by the linear
polynomials was already known. It is Rankin's bound (Eq. (59) of
Chap. 1, [RanI)). Several other bounds on A (n,() that can be obtained
by the linear programming approach have already been mentioned in §2.6
of Chap. 1 (including Astola's analogs «68), (69) of Chap. 1, [AstI)) of
Tietiiviiinen's coding bounds).

(iv) The principal achievement of [Kabll, as we have discussed in §I, lies

in the bounds obtained when the dimension n is large. As in the previous
subsection we state the results only in the case of the sphere. Working by
analogy with (47) the authors use the polynomial

f (t) = _1_ { ph.a? (t) p~a,a) (s) - p~a,a) (t) P~"+I) (s) J2 (53)
t-s

of degree 2k + 1, where ex = (n-3)/2, s = cos () and k = 1,2, .... This

implies

(54)

for cos () ~ d~L where the Jacobi polynomial p~a,a) (x) has k simple zeros
1 > t (al
l,k
> ... > t k,k
(a) .
Bounds for Codes and Sphere Packings 265

For large n Kabatiansky and Levenshtein show by a detailed investigation

of t f~l that (54) implies the bound (6). For 8 < 63· (6) implies the
simpler result that, as n - 00,

llog2 A (n, 8) < - 21 log2 (I-cos 8) - 0.099 (55)

n -

(Eq. (66) of Chap. 1), which for 8 = 71"/3 leads to the asymptotic bound on
the kissing number given in (49) of Chap. 1.
To deduce their bound (5) on the density of an n-dimensional sphere
packing, Kabatiansky and Levenshtein make use of the following analog of
Elias's inequality (48).
Theorem 6 [YagIl The highest density d n of any sphere packing in Rn
satisfies
d n ~ (sin V28)n A (n + 1,8), for 0 < 8 ~ 71". (56)

Proof Since the reference [VagI] cited by [KabI] is fairly inaccessible we

sketch a proof. Let S be a large sphere of radius p in Rn +1 and let II be a
hyperplane through the center of S. In II we construct an n-dimensional
packing of unit spheres with density dn. By shifting the packing slightly
we may assume that the portion of II inside S contains at least dnpn
centers. We project these centers, perpendicularly to II, "upwards" onto S.
The Euclidean distance between the new points is at least 2 and their
angular separation is at least 8, where sin 8/2 = I/p. The result is
therefore a spherical code of minimum angle 8, containing
d n (sin8/2)-2 ~ A(n+I,8) points.
The bound (5) follows from (55) and (56) by taking 8 = 1.0995.
For finite n a numerical bound on the density may be obtained by
combining (54), (56) with information about tf~ given in [Kabl, p. 12l
The bounds on log2 (j for n ~ 48 given in Table 1.3 were obtained in this
way.
In [Lev6], [Lev7] Levenshtein has found other polynomials f (t) that
lead to the bound (6) and give good bounds for small values of n (see also
[Sid4]) .

4. Other bounds
(i) Lovasz [LovI] has solved a coding problem of long standing by
determining the "capacity" (see [Sha4]) of the pentagon. His method,
which gives a new way to bound the capacity of any graph, turns out to be
closely related to Delsarte's linear programming bound for subsets of an
association scheme (see [McE2], [McE3], [Sch9]).
(iO An unusual method for obtaining lower bounds on A (n, d, w) was
given in [Gra2] (see also [GraI], [Gra3], [Hon2], [Hon3], [Klrj>I]). This
yields
266 Chapter 9

which implies
A(n,4,w) _nw-l/w! asn --+00;

and
A (n, 2 0, w) :::: n w - 0 + 1/w! as n --+ 00.

(iii) For codes in some of the other 2-point-homogeneous spaces mentioned

in this paper see [Bigll, [Big3l, [Cam3], [DeI15], [Ham3], [Ham4],
[Kan4], [Sta6], [Thall, [Tha2l.
(iv) Lloyd [LioIl has presented another method for deriving linear
programming bounds for spherical codes, by defining an association scheme
on On. Yet another way of approaching these problems has been given by
Neumaier [Neull. Urakawa [UraIl derived a bound on ~n that is
equivalent to Levenshtein's 1979 bound [Lev7l by considering the spectrum
of Laplace-Beltrami operators.
(v) Recently Tsfasman, Vladut and Zink [Tsf2l and others have shown
that codes over Fq , for q = p2, P ~ 7, constructed from algebraic curves
give an asymptotic improvement over the Gilbert bound (see §2.11 of
Chap. 3).
Acknowledgements. R. Askey, E. Bannai, A. R. Calderbank, P. Diaconis,
C. F. Dunkl, W. M. Kantor, T. Koornwinder, R. Lidl, A. M. Odlyzko, J. J.
Seidel and D. Stanton provided many helpful suggestions.
 10
Three Lectures on Exceptional Groups
J. H. Conway

The first lecture records certain exceptional properties of the groups

L 2 (P) and gives a description of the Mathieu group MI2 and some of its
subgroups, followed by a digression on the Janko group J I of order 175560.
With the exception of the Janko group material, all the structure described
appears within the Mathieu group M 24, which is the subject of the second
lecture, where M 24 is constructed and its subgroups described in some
detail. The information on M 24 is then found useful in the third lecture,
on the group Coo = ·0 and its subgroups. An appendix describes the
exceptional simple groups.
This chapter is based on lectures given in Oxford, England in 1970.
The groups of the title have been called the sporadic groups since [Con2],
but we have preferred not to update the language in this chapter.

1. First lecture
1.1 Some exceptional behavior of the groups Ln (q). The general linear
group GLn (q) is the group of all linear automorphisms of an n-dimensional
vector space over the field Fq , q being any prime power. The special linear
group is the normal subgroup consisting of the automorphisms of
determinant 1. The center of either of these groups consists of operations
of the form x - kx (for kEFq), and we obtain the corresponding projective
groups PGLn (q) and PSLn (q) by factoring out these centers. PSLn (q) is
a simple group (for n ~ 2) except in the two cases n = 2 and q = 2 or 3. It
was called by Dickson [Dicl1 the linear fractional group LF(n,q), but we
shall use Artin's abbreviation Ln (q) [Artll
When n = 2, we take a basis y,z for the space, so that the operations
of the group SLn(q) have the form y - ay + bz, z - ey + dz (with
ad - be = 1). The projective line PL (q) consists of the q + 1 values of
the formal ratio x = y/z (which are conveniently thought of as the q field
elements together with the formal ratio 00) and, on the projective line,
268 Chapter 10

L2 (q) becomes the group of all operations of the form

ax+b
x-- --
ex+d

with ad - be = 1, or equivalently with ad - be any non-zero square in

F q' We use the following names for subsets of PL (q ):
0= PL(q), 0' = Fq = 0\ too}, Q = {x 2 :x E Fq },

N = O\Q, Q' = Q\ to}, N' = N\{oo}.

In fact L 2 (q) is generated by three operations

a:x - x+ 1, (3:x - kx,

provided that Q' is the set of powers of the field element k; for (3b a Q takes
x to kbx + a, while (3b a Q-yac takes x to e - (kbx+a)-l, and plainly
every operation of L2 (q) can be expressed in one of these forms.
The necessary set of defining relations varies slightly with the structure
of q: when q is a prime congruent to 3 modulo 4 we have
L 2(q) = <a,(3,-y:aq = (3'h (q-Il =-y2=af3· a- k = «(3-y) 2 = (a-y) 3= 1 >,

since it is easy to see that these relations enable us to put every function of
a, (3, -y into one of the two forms above. (af3 denotes (3-1a(3,) In the cases
q = 3,5,7,11, we shall take k = 1,4,2,3; for p = 5, the relation
(a(3-y)S = 1 completes the above set.

It was proved by Galois (in a letter to Chevalier written on the eve of

his fatal duel) that L 2 (P) cannot have a non-trivial permutation
representation on fewer than p + 1 symbols if p > 11 [Hupl, p. 2141.
However, for p = 3,5,7,11 there exist transitive representations on exactly
p symbols, and by studying these we shall illuminate some of the
"unexpected" isomorphisms
L 2(3) ;;: A4, L 2(4) ;;: L 2(5) ;;: As,
L 2 (7) ;;: L 3 (2), L 2 (9);;: A6 , L 4 (2);;: As,

where An is the alternating group of degree n. These isomorphisms are

visible inside the Mathieu groups, and in the second lecture we shall find
the pieces fitting together. This study is also interesting as an example of a
common but puzzling phenomenon: the four cases have much in common
but in each particular case there is something peculiar to that case, so that
there is no completely general pattern.
It happens that thep objects permuted by L 2 (p) (P=3,5,7,1i) can in
each case be taken as p involutory permutations of the set O. For
p = 3,5,7,11, respectively, define 7r to be

(000)(12), (000)(14)(23), (000)(13)(26)(45),

(000)(16)(37)(9X)(58)(42) (where X denotes 10)
Three Lectures on Exceptional Groups 269

and define 71" i as a - i 71" a i . (Mnemonic: 71" interchanges 00 with 0, and takes
x to nx or x/n according as x E Q' or x EN', where n = 2,4,3,6 in the
respective cases.) It is convenient to define 71" to be the identity
00

permutation of n.
Theorem 1. The group L 2 (p) (p = 3, 5, 7,11) leaves invariant the set II
consisting of the p involutions 71" i (j En').

Proof II is obviously invariant under a and the mnemonic shows it to be

invariant under {3 also. So we need only check invariance under 'Y- Here
the miraculous enters - we find that (7I"i)'Y = 7I"io, where 0 is
(00)(0)(1)(2), (00)(0)(12)(34), (00)(0)(12)(36)(4)(5),
(00) (0) (1) (2X) (34) (59) (67) (8)

in the four cases.

We now discuss the cases separately.
1.2 The case p = 3. Here the 71" i are the elements of a Klein 4-group
which is contained in L 2 (3), and being invariant, is a normal subgroup of
L 2 (3), which is therefore not simple. In this case the representation on p
letters is not faithful, since we observe that 0 is the identity. Since L 2 (3) is
of order 12 and contains only even permutations of n, we have the
isomorphism L 2 (3) == A4 .
1.3 The case p = 5. The group L 2 (5) induces only even permutations of II,
and being of order 60, can only be the alternating group on II, and we
have the isomorphism L 2 (5) == As. Transforming II by the operations of
the symmetric group $6 on n, we obtain just 6 such sets of five involutions.
Each permutation of $6 defines a permutation of these six sets, and so S6
can be regarded as a permutation group on two distinct systems of six
objects. The S s fixing II is generated by the permutations 7I"}' 71" i 71" j and
contains the original L 2 (5), and so does not fix any element of n. It
follows that the two permutation representations of $6 on six objects are
essentially distinct, in the sense that they are related by an outer
automorphism of S6. It is known that only for n = 6 does $ n possess an
outer automorphism.
The details are as follows. Let IIi = 71" i' II 71" i (j E nL The sets IIi
are the six sets of five involutions, and the permutation 71" i of n induces the
permutation (II IIi) of these six sets, and symmetrically the permutation
00,

(00,0 induces the permutation taking IIx to lIar So there is an outer

automorphism 8 interchanging (00,0 with 7I"i, and hence all the 15
involutions of shape (ab)(c )(d)(e )(f) with those of shape (uv )(wx )(yz).
The former class corresponds to Sylvester's duads [Syll], the latter to his
synthemes, and the six sets IIi are the synthematic totals. We have
8 2 = 1, and (i,j)8 = 7I"i,j' where 7I"i.j = 7I")' 7I"i 7I"j = 7I"i' 7I"j 7I"i except
that 7I"oo,i = 7I"i,oo = 7I"i'
1.4 The case p = 7. In this case the IIi are the elements of an elementary
abelian group E of order 8. L 2 (7) acts as a subgroup of the automorphism
270 Chapter 10

group L 3(2) of E (L 3 (2) because we can regard E as a 3-dimensional

vector space over F 2), and since IL 3 (2)1=IL 2 (7)1 we have the isomorphism
L 2 (7) =L 3 (2).

The group E generates together with L 2 (7) a subgroup F of Ag of order

8·168, and so of index 15 in Ag. Since L2 (7) is generated by a, fj, ')' and is
transitive on IT, the group F is generated by a, fj, ,)" 7ro. But (a miracle!)
we observe that ,),o=7ro, so that F is equally generated by a, fj, 0, 7ro. The
group F therefore contains two subgroups L 3 (2) complementary to E,
namely the "original" L 2(7) = <a,fj, ,),>, and the "exceptional"
L 2(7) = L 3(2) = <a,fj,o >. These are not conjugate in F, for one of
them is transitive on n, while the other fixes 00. It is exceptional for the
holomorph of an elementary abelian group to exhibit this behavior.
We can describe this situation in another way by saying that F has an
outer automorphism 8 which fixes E pointwise and also fixes the quotient
group FIE pointwise, without of course fixing F. In fact we have 8 2 = 1,
aO = a, fjo = fj, ')'0 = 0, 0° = ')' and 7rf = 7r;. It is not hard to see that every
subgroup of F isomorphic to L 2(7) is conjugate to either <a,fj,,),> or
< a, fj, 0 >, so that 8 is essentially the only outer automorphism of F.
Another consequence of this situation is that F has two essentially distinct
faithful representations on 8 letters (compare the behavior of S6).
Since F has index 15 in Ag we obtain 15 sets of 7 involutions like IT on
transforming IT by the elements of Ag. These 15 sets we call the even sets
- if we transform instead by the elements of Sg\Ag we obtain 15 further
sets, the odd sets. Each of the 105 regular involutions of Ag (i.e. those of
shape (ab) (cd) (ef) (gh)) belongs to just one even set and just one odd set.
There is therefore a natural graph with vertices the 30 sets and edges the
105 regular involutions of As, with each involution joining the two sets
containing it. Each vertex of either set is joined to just 7 vertices of the
other set.
We can derive a similar graph from an elementary abelian group of
order 16 by taking as even and odd vertices its 15 involutions and 15
subgroups of order 8, joining each subgroup to the 7 involutions it contains.
To show that these two graphs are isomorphic we must turn our 15 even
sets IT; into an elementary abelian group of order 16. We do this by
defining the product IT;IT j of two distinct even sets to be the unique third
set joined in the graph to all the vertices joined to both IT; and IT j . There
is a simple combinatorial proof that this does indeed define a group, and
since each element of As acts non-trivially on this group, Ag is a subgroup
of its automorphism group Li2). Since 1L4 (2) 1 = V2· 8! we have the
isomorphism L 4 (2) = Ag. We do not go into further details here, since we
shall give another proof of this isomorphism in the second lecture.
In this sort of work it is useful to have a simple way of specifying at a
glance the structure of the groups that appear. We shall say that G is a
group of type A.B (or AB, when no confusion can arise) when we mean
that G has a normal subgroup A whose quotient is isomorphic to B. In
this notation, the cyclic group of order n is written simply as n, and the
Three Lectures on Exceptional Groups 271

elementary abelian group of order pn just as pn. Thus the group F we

have just discussed is a group of type 2 3 L 3 (2).
1.5 The case p = 11. In the cases p = 3,5,7 the group <a,{3,'Y,o> was
A4, S6, 2 3L 3 (2) respectively, and in the cases p = 5,7 this group had an
outer automorphism () fixing a and (3 and interchanging "1 with o. We
assert that for p = II the group <a,{3,'Y,o> is the Mathieu group on 12
letters, M 12, and that <a,{3,'Y> and <a,{3,o> are two subgroups of type
L 2 (ll) not interchanged by any automorphism of M 12.
We shall show here that <a,{3, "1,0> is a proper subgroup of A12, with
a 6-dimensional projective representation over F 3 • Take a 12-dimensional
space f!( over F 3, with basis vectors Xi(j En), and for S ~ n let Xs
denote ~Xi(j ES), with a similar notation in other cases. We consider the
space "If/ spanned by vectors wi(jEn), where Woo=X{}, and
Wi = XN-i - XQ-i(j En'), where for instance N - i means {n-i:n EN}.
The ternary Golay code ~12 (§2.S.5 of Chap. 3) is the set of all 12-tuples
(coo, co, ... ,cx) with ~CiXi E 11(

Theorem 2. ~12 is 6-dimensional, and ~CiWi = 0 if and only if

(Ci) E ~12'

Proof woo, WI> W3, W4 WS, W9 are linearly independent in "/I( so ~12 is at
least 6-dimensional. But WN = wQ = 0, so that W {} = 0 and
WN-i = WQ-i = 0 for i En'. Thus (Ci) E ~12 implies that ~CiWi = 0,
and the Wi satisfy at least six linearly independent relations, from which
the statements follow.
We define linear maps A, B, C, D on f!( by:
A :Xi -+ Xi+1> C :Xi -+ ± x-Iii, D :Xi -+ Xio,

where 0 = (oo)(0)(1)(2X)(34)(59)(67)(S), and the sign ± is + for i E Q,

- for i E N.
Theorem 3. A, B , C, D preserve 11(
Proof. We check the effects of these operations on the Wj, finding
miraculously that

where the sign + is the opposite of the previous sign ±.

It follows at once that a, {3, "1, 0 generate a proper subgroup M 12 of A12.
The following portmanteau theorem, which we shall not prove, gives us
considerable information about M 12. Most of it is easily proved by
mimicking the methods of the next lecture for M 24.
Theorem 4. M 12 is a quintuply transitive group of order 12· 11 . 10·9· S.
The group <A,B,C,D> is a non-splitting extension 2M l2, consisting
precisely of those automorphisms of f!( which preserve 11( It has an
outer automorphism () that satisfies
(}2 = I, A O= A-I, BO = B, CO = C- I = -C, DO = D
272 Chapter 10

and whose adjunction completes it to a group 2M\22.

The symmetry corresponding to 8 is illuminated by defining
f = [i£ / "H',' and vectors Vi to be the canonical images of the Xi in f. The
group 2M\22 is then a group of linear automorphisms of the space
fffi "HI; and as such permutes the 48 vectors ±Vi, ±Wi, the automorphism
8 interchanging each Vi with the corresponding Wi' The quotient group
M\22 permutes the 24 subgroups Vi = {O,Vi,-Vi} and Wi = {O,wi,-w;l of
fffi "H/.' We describe some of the subgroups of 2M\22 in these terms.
Fixing V we have a subgroup 2M II with orbits of sizes 2, 22, 24 on
00

the 48 vectors, namely {±voo } {±vi:i EO'} and {±wi:i E Ol. This
must be a direct product C 2 x M ll , (where C 2 is a cyclic group), since it
has a subgroup Mll of index 2 fixing Voo and -V oo separately, with orbits
of sizes 1, 1, 22, 12, 12, namely {v oo }, {-v oo }, {± vi:i EO'} and
{wi:i EO}, {-wi:i EO}. Note that the permutation representation of
2M lIon 22 objects here is not the direct product of permutation
representations of C 2 on 2 objects and M II on 11 objects, since the
subgroup MIl is still transitive on all 22 objects. There is a second
conjugacy class of subgroups M II in M \2, obtained by interchanging the
roles of the Vi and Wi'
There is a subgroup SL 2 (I1) = <A,B,C> with two orbits of size 24,
and a subgroup L 2 (I1) with orbits of sizes 1, 1, 11, 11, 1, 1, 11, 11,
namely {v oo }, {vi:i EO'}, {woo}, {wi:i EO'} and their negatives. The
group L 2 (I 1) is completed to a direct product 2 x L 2 (I1) by adjoining -I.
In the quotient group M\22 these both become L 2 (l1)'s but one is
maximal in M \2, and the other is not.
The subgroup 2 x M 10 fixing two subgroups V Va has orbits of sizes
00,

2, 2, 20, 24. It has a subgroup MIO fixing Voo and -V oo separately, and
this has a further subgroup M IO ' fixing also Va, -Va separately. M 10' has
orbits of sizes 1, 1, 1, 1, 20, 6, 6, 6, 6, the fixed points being obvious, and
the orbits of size 6 being {Wi: i EN}, {Wi: i E Q} and their negatives.
Coincidence of orders implies the isomorphism M 10' ;;: A6, but we can also
see M IO ';;: L 2 (9) by translating the 10 subgroups VJ, V 2 • • . . • Vx into the
10 points
0, 1, i, -i, 1 - i, -1 - i, i + 1,00, i - I , -1

of the projective line PL(9), when the permutations of M IO ' become linear
fractional transformations. We therefore have A6 ;;: M IO ' ;;: PSL 2 (9). The
three groups S6, M 10, PGL 2 (9) are all distinct, being the three subgroups
of index 2 in Aut (A 6) (Aut (A 6)/A 6 is a 4-group). S6 is completed to
Aut (A 6) by our outer automorphism 8, interchanging duads with
synthemes, MIa by an outer automorphism interchanging V with Va, and 00

PGL 2 (9) by its field automorphism, interchanging i with - i. The groups

S6, M IO , PGL 2 (9) may be distinguished as abstract groups by the numbers
of classes of elements of orders 3 and 5. S6 has two classes of order 3, one
of order 5, MIO has one of each, and PGL 2 (9) has one of order 3 and two
of order 5.
Three Lectures on Exceptional Groups 273

1.6 A presentation for M 12. Each of the triples a, {3, 'Y and a, (3, b satisfies
the defining relations for L 2 (l I). It is remarkable that the conjunction of
the relations so obtained with the miraculous relation hb)2 = {32, or
equivalently (b'Y{3)2 = 1, gives a presentation for M 12 , namely
<a,{3,'Y,b: a" = {35='Y2=b2~aP'a-3 =

(a'Y)3= (ab)3 = ({3'Y)2 = ({3b)2 = (b-y{3)2 = 1> .

This can be supplemented by the relations (J2 = 1, a8 = a-I, {38 = {3,

'Y 8 ='Y- I, b 8 =b to yield a presentation for M122. We can get a
presentation for 2MI22 by making products involving 'Y into -1 instead of
1, where - 1 is an involutory central element.
If in the presentation for M 12 we eliminate {3 by means of the
miraculous relation, we get the considerably simpler presentation
< a,'Y,t5:a" = 'Y2 = b2 = (a'Y)3 = (ab)3 - hb)IO=a'Y~'Y~'a2 = 1 >,

which we can further transform into

<a,'Y,1/:a"='Y 2 -h1/)2 = (a'Y)3 = (1/'Ya)3 = 1/10 = 1/-2 a 1/2· a 2 = 1 >

by replacing 'Yb by 1/, eliminating b. We get yet another by eliminating 'Y

instead. (These are quite essentially distinct, though in appearance very
similar, since we know that the subgroups <a,{3, 'Y> and <a,{3,b> are
thoroughly distinct - indeed one is maximal and the other is not.)
1.7 Janko's group of order 175560. The way in which the exceptional
representations of the small groups L 2 (P) can arise in other situations is
well illustrated by considering the permutation representation of Janko's
group J I of order 175560 on 266 letters ([Janll, [Jan2]; [Camll, [Evall,
[Gagll, [Marl], [Perl1, [Whi2]). The stabilizer of a point is an L2(l1),
with orbits of sizes 1, 11, 110, 132, 12. Knowing this much, it is easy to
construct J I and so verify its existence. The centralizer < i > x As of an
involu tion i of J I is such that the A5 is contained in an L 2 (1 I) . We take a
particular subgroup L 2 (l I) = L. Associated with L is a set S of 11
involutions centralizing As's in L, and permuted like our 11 permutations
'If" i of the projective line PL (1 I) - we let 'If" i be the permutation of PL (1 I)
corresponding to the involution i of S. The 12 cosets of L which form an
orbit under L are permuted like the points x of PL (1 I) - we let Lx be
the coset corresponding to x E PL (1 I). Any element of L can be expressed
as a product j(a,{3,'Y) of the generating permutations of L - the
permutation it induces on S is the corresponding product j (a, {3, b) .
Now the effect of an involution i of S on the 266 cosets must be
capable of being described in a manner invariant under L. This limits the
possibilities, and in fact forces the following unique situation. The 266
cosets are L, Li, Lij, Lxj, Lx, for x EPL(1I), i, j E S, i ¢ j.
Postmultiplication by the typical element of L permutes these by
permuting x by j(a,{3,'Y) and i and j by j(a,{3,t5); so we need only
describe the effect of postmultiplying by a typical k E S.
274 Chapter 10

Since k 2 is the identity and Liji = Lij, we need only consider

expressions Lijk and Lxjk in which i ,j, k are distinct. Now the
permutation 7r i7r'j'Tr k has two fixed points U and v in PL (l]) if it has any,
and U and v may be distinguished by the condition that the permutation
7rh which interchanges U and v also interchanges U7ri and V7rb but not
U7rk and V7ri. We define U = [i,j,kl, and then [k,j,il will be v. We
then have
Lijk = Lji if7ri7rj7rk has no fixed point,
Lijk = Lxh if [i,j,k] = x = [h,i,j],
Lxjk = Lyh if x = [h,k,j] and y = [j,k,hl,
Lxjk = Lhi if X7rj = [i,j,k] = [h,i,j].

These equations are quite easy to work with when we remember that the
equation [i,j,k] = U inverts to determine 7ri, 7rj' 7rk as interchanging the
pairs (U,U7rk7rj), (U7ri,U7rk), and (U,U7ri7r) respectively. (Any pair of
distinct points of PL (l]) is interchanged by just one of the permutations
7ri.)
That J, is a proper subgroup of A266 with the right order is also easy to
verify. If we join each of the 266 points to the orbit of size II in its
stabilizer we get a graph in which the joins are (L, Lh), (Li, Lhi),
(Lij,Lhij), (Lx,Lyh), (Lxj,Lyhj), where in each case h varies arbitrarily
and y = X7r h. It is not hard to verify that the operations defined above
preserve this graph. If instead we join each point to the corresponding set
of 12 points, we get the joins (in which y and h are arbitrary)
(L,L y ), (Li,Lyi), (Lij,Lyij),
(Lx,L), (Lx,Lxh), (Lxj,Lj), (Lxj,Lxhj).

2. Second lecture
2.1 The Mathieu group M 24 . We define M24 to be the group obtained by
adjoining the permutation o:x -x 3/9 (x E Q) or x -9x 3 (x EN) to the
group L 2 (23) acting on the projective line fl =PL (23). We list the
generators in full:
a = (00)(0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22),
{3 = (00)(15714510 201711222119) (0)(3612 12481691813),
"( = (00 0)(15 3)(7 13)(14 18)(5 9)(10 16)
(208)(174)(11 2)(22 ])(21 12)(196),
o = (00)( 14 17 11 19 22)( 15)(20 10 7 5 2])
(0)(18426 ])(3)(8 1613 9 12).
It is sometimes convenient to replace the pair ,,(,0 by the product "(0 2 ,
which generates the same group, since plainly "( and 0 commute. We have

"( = h0 2)5, 0 = h0 2)-2, and

"(0 2 = (00 0)(15 3)(1422241918 II 1 176)(2013 211658712109),
Three Lectures on Exceptional Groups 275

or, algebraically, /,0 2 interchanges 00 with 0, and otherwise takes x to

- (x /2)2 (if x E Q), or (2x)2 (if x EN). The operations /,,0 plainly
normalize the group <{3>, and in fact we have {3'Y={3-I, {30={33, and
h0 2)-1 (3 h0 2) = {32.
Theorem 5. M 24 is quintuply transitive on n.
Proof By randomly multiplying the generators, we find permutations of
cycle-shapes 123, 1211 2, 1373, 1454,2 12 , 1828 and 46 (for instance a, {3,
oa 2, 0, /" (ao)3 and (a I3 /,02)3). The shapes 1 23 and 212 show that M24
is transitive, and then that the stabilizer of any point must have a
permutation of shape 1 23, and so be transitive on the remaining 23 points.
In a similar way we see that the stabilizer of 2 points is transitive on the
remaining 22 (using shapes 12112 and 13 73), and the stabilizer of 3 points
is transitive on the remaining 21 (using 1373 and 1454), so that M 24 is
quadruply transitive. Now the subgroup fixing a set of 4 points as a whole
has permutations of shapes 1454 and 46 , so is transitive on the remaining
20 points, whence M 24 must act transitively on the set of all 5-element
subsets of n. But the subgroup fixing any 5-element set has permutations
of shapes 1454 and 18 28 which induce permutations of shapes 5 and 132 on
the 5-elements, and the quintuple transitivity follows from this, since any
two permutations of shapes 5 and 132 generate the full symmetric group on
5 letters.
We define M 24-k to be the pointwise stabilizer of a k -element subset of
n in M 24 (k ~ 5).
To show that M24 is a proper subgroup of the alternating group A24 , we
turn the set P (n) of all subsets of n into a 24-dimensional vector space
over F 2 (by defining the sum A + B of two sets to be their symmetric
difference (A\B) U (B\A» and show that M 24 preserves a 12-dimensional
subspace. The binary Golay code CC l2 (§2.8.2 of Chap. 3) is the space
spanned by the 24 sets N; (j En), where N = n and otherwise
00

N; = N - i = {n - i : n EN}. If S ~ n, we write N s for '2:.N; (j E S) .

Theorem 6. CC 24 is at most 12-dimensional.
Proof We can verify directly that N n = NN = 0, whence NN; = 0 for
each i En, and N c = 0 for each C E CC 24. If CC 24 is k -dimensional, we
therefore have at least k independent linear relations between its
generating sets N;, so that k ~ 24 - k. (The relations above are
consequences of well-known number-theoretical facts, but in any case it is
obvious a priori that each of N n, NN must have the form
aN + bQ + c{ oo} + d {OJ, so there is not much to check,)
A similar calculation gives the exact dimension of CC 24 , and a test for
membership in CC 24 .
Theorem 7. CC24 is exactly 12-dimensional. and we have C E CC 24 if and
only if Nc = O.
Proof We find N{-2.0,2,3} = {0,1,2,3,4,7,10,12}, a CC-set (i.e. member of
CC 24) with least element O. Adding i (j ~ 10) to the elements of this we
276 Chapter 10

get C{1 -sets with least element i (0 ~ i ~ 10). These obviously span an
II-dimensional space, and we get the extra dimension by adjoining any
C{1-set containing 00.
Theorem 8. M 24 preserves C{124'
Proof We have obviously N;a = N;-h N;f3 = N 2;, and we verify the
equations
NO'Y02 = Nco, Nco'Y 02 = No, Nl'Y02 = N(-I,O,I,3],
N_l'Y02 = N(3,12,2I),

from which we can deduce N; "10 2 E C{124 for all i, using

N 2; "10 2 = N;{3'Y02 2
= N; "10 {32. (There is really only one non-trivial
relation to be verified, since C{124 is spanned by the N; (for i E Q') together
with Nco = 0.)
Theorem 9. There exist 8 -element C{1 -sets, called (special) octads, and
each non-empty C{1 -set is the symmetric difference of a strictly smaller
C{1 -set with an octad. Each 5-element set is contained in just one octad.
Proof We have already found one octad, {0,1,2,3,4,7,1O,12}, and so
each 5-element set is contained in at least one octad, by Theorem 5. Any
C{1-set with at least 5 points is therefore the symmetric difference of an
octad containing those 5 points and a strictly smaller C{1-set. If some non-
empty C{1-set had fewer than 5 elements, every set of the same cardinal
would be a C{1-set, by Theorem 5, and by taking symmetric differences we
should obtain every two-element set, and so every set of even cardinal, as a
C{1-set, which cannot be. It follows that the octads are the smallest non-
empty C{1-sets, for from any smaller non-empty one we could obtain a still
smaller one, and so on. No 5-element set can be contained in two distinct
octads, for then their symmetric difference would be a C{1-set with at most
6 members.
The last sentence of Theorem 9 asserts that the octads form a Steiner
system S (5, 8, 24) (§3.1 of Chap. 3). We can deduce from it that there
are just

permuted transitively by M 24 . Witt [Wit2], [Wit3] proved that the system

S (5, 8, 24) - likewise the systems S (4, 7, 23), S (3, 6, 22), S (5, 6, 12),
S (4, 5, 11) which are involved with the other Mathieu groups - is
essentially unique, and defined M 24 to be its automorphism group.

2.2 The stabilizer of an octad. M 24 contains permutations of shapes

135 15 and 12 248 2, for example oa ll and oa 5• The octad containing the
5-cycle in the first case (being fixed) can only be the union of the 5- and
3-cycles, and in the second case the octad containing the 4-cycle and a
I-cycle must be the union-of the 4-, 2-, and I-cycles. The fourth power of
the second operation is therefore a permutation of shape 18 28 whose fixed
Three Lectures on Exceptional Groups 277

points form an octad. We can therefore suppose that M24 contains the
permutation X = (abcde Hfgh)(i) (jkl ... x) and a permutation J.L of shape
I s2s interchanging i and j and fixing a,b,c,d,e,f,g,h. The permutations
J.L\ J.L,,2, ••• have the same fixed points but now interchange i with k, i with
I, etc., and so the pointwise stabilizer of {a, b, c ,d, e,f, g, h} has at least 15
involutions and order at least 16. The pointwise stabilizer of {a,b,c,d,e}
has at least three times this order, since it contains also the permutation xs.
of order 3. It follows that M 24 has order at least
24·23·22·21·20·16·3 = 244823040.
Theorem 10. The subgroup of M24 fixing an octad setwise is a group
24As• an extension of an elementary abelian group of order 16 (fixing the
octad pointwise) by the alternating group on 8 letters. which is isomorphic
to L 4 (2). M24 has order exactly 244823040. and contains all
permutations of n that preserve the Golay code rc 24.
Proof We consider the subgroup H of index 16 in the stabilizer of an
octad that in addition to fixing the octad {a,b.c,d,e,f,g,h} as a whole,
fixes the point i. Since {a, b, c ,d ,e,f ,g, h, i} contains just one non-empty
rc-set, the rc-sets disjoint from it form a space of codimension 8, and so
dimension 4, and it is easy to see that every non-trivial element of H acts
non-trivially on this space, so that H is a subgroup of its automorphism
group L 4 (2), whose order is
(16- 1) (16-2) (16-4) (16-8) = 20160.

The group of permutations of a,b,c,d,e,f,g,h induced by H is transitive

and contains the 3-cycle (hgf) (induced by X5), and so must be the
alternating group As of order Yl·8!=20160; the coincidence of orders
establishes the isomorphism H == As == L 4 (2). Since 20160 =
244823040/759·16, the order of M24 is exactly 244823040, and the
subgroup fixing a, b, c, d ,e,f, g, h individually has order exactly 16, being
transitive on the remaining 16 letters. It consists of the identity and the
15 involutions we found earlier, and so is an elementary abelian group. (It
is the dual of the above 4-dimensional space of rc-sets, and is also
permuted by the group H.) The upper bound by this argument for the
order of M 24 applies also to the group of all permutations of n preserving
rc 24. and establishes the identity of the two groups.
Theorem 11. Let {a" a2 • .... as} be an octad. Then the number of
octads intersecting {al .... • aj} in (al •... • a) (exactly) is the (j+l)th
entry in the (; + l)th line of Table 10.1.
Proof Since M 24 is transitive on those sets of any given cardinal that are
contained in octads, the number of octads containing {a I •...• aj} is
exactly

and we have the rightmost entries. The others are deduced from these by
repeated use of the rule that the sum of two adjacent entries in any line is
278 Chapter 10

Table 10.1 How many octads?

759
506 253

330 176 77

210 120 56 21

130 80 40 16 5

78 52 28 12 4

46 32 20 8 4 0
30 16 16 4 4 0 0

30 0 16 0 4 0 0 0

Table 10.2 How many dodecads?

2576

1288 1288
616 672 616

280 336 336 280

120 160 176 160 120

48 72 88 88 72 48
16 32 40 48 40 32 16

0 16 16 24 24 16 16 0

0 0 16 0 24 0 16 0 0

the entry just above them in the previous line. Table 10.2 gives the
number of umbral dodecads meeting (a I • . . . • ai} in (al . .. '. a) (see
below).
2.3 The structure of the Golay code ~ 24
Theorem 12. ~24 has weight distribution 0 1 8759 12 2576 16 759 24 1.
Proof Since by Theorem 11 two distinct oct ads intersect in 0 or 2 or 4
points their symmetric difference has cardinal 16 or 12 or 8. The theorem
follows using Theorem 9 and the fact that ~-sets come in complementary
pairs.
2.4 The structure of P( n) / ~24
Theorem 13. Every subset of n is congruent modulo ~24 either to a
unique set of cardinal at most 3 or to each of 6 distinct sets of cardinal 4.
Proof If a subset S has 5 or more members we obtain a congruent set
with fewer members by taking the symmetric difference with an octad
containing 5 members of S. If S has 4 members we obtain 5 more
Three Lectures on Exceptional Groups 279

4-element sets congruent to S by taking the symmetric differences of S

with the 5 octads that contain it (Theorem 11). If two distinct sets of 4 or
fewer elements are congruent modulo rt'24, their symmetric difference is an
octad, and so they must be disjoint and have 4 elements each, so the
theorem is best possible.

Thus 4096 = 1 + 24 + (224) + (234) + !

(244). In other words rt' 24 is a
triple-error-correcting code with covering radius 4.
2.5 The maximal subgroups of M 24. The complete list of maximal
subgroups of M 24 is known. 1. A. Todd [Tod3) observed eight distinct
types in his paper on M 24, and a further type, the octern group of order
168 described below, has since been discovered. Chang Choi [Choll and
Curtis [Cur2)-[Cur5) have given proofs that these nine groups are indeed
the only conjugacy classes of maximal subgroups of M 24. With two
exceptions these groups are easily described in terms of the Golay code
rt' 24, and we describe them in some detail. We shall give a further
description of these groups in terms of the MOG in Chap. 11.
A 12-element rt'-set is called an (umbra/) dodecad, and a
complementary pair of umbral dodecads is a duum. A triplet of mutually
disjoint octads is a trio, and a system of 6 tetrads with the property that
the union of any two is an octad is a sextet. (Mnemonics:
duum = 2 umbrals, trio = 3 octads, sextet = 6 tetrads') In general, we use
n-ad for an n-element subset of n, except that the terms octad and
dodecad usually presuppose the corresponding adjectives special and
umbral.
The maximal subgroups M 24 can now be described as the stabilizers of
the concepts monad, duad, triad, octad, sextet, trio, duum, along with two
further groups L 2 (23), L 2 (7) (compare Table 11.1 of the following
chapter). The L 2 (23) is the group we started from, and the L 2 (7) is the
last maximal subgroup. We give theorems asserting the exact amount of
transitivity of each of these groups on various configurations in n. We say
that a group is a + b transitive on two sets A and B if it contains elements
taking any a elements of A, and simultaneously any b elements of B, to
preassigned positions.
Theorem 14. The group 24AS fixing an octad is 6 + 1, 3 + 2, and 1 + 3
transitive on the octad and its complement.
Proof The stabilizer of an octad and one point in its complement is the
group H = As, sextuply transitive on the octad, and, in its role as L4 (2),
doubly transitive on the remaining 15 points of the complement, which can
be regarded as the involutions of a group 24 acted upon by L4 (2). The
stabilizer of two points of the complement is therefore a group 23 L3 (2),
which can only be the group we met in the first lecture, triply transitive on
the octad. The stabilizer of one point in the octad together with two of the
complement is L3 (2), and is easily seen to be transitive on the remaining
14 points of the complement.
280 Chapter 10

Theorem 15. M 24 is transitive on dodecads. The group fixing a dodecad

set wise is the Mathieu group M 12, which is 5 + 0, 3 + I, I + 3, 0+5
transitive on the dodecad and its complement. This group M 12 has index
2 in a group M 12 2, the stabilizer of a duum.
Proof The group stabilizing the sets {a,b} and {c,d,e,f,g,h} is plainly a
subgroup 24 S6 inside our group 24 A8 , and it is easy to see that its subgroup
24 is transitive on the 16 octads which by Theorem 11 intersect
{a,b,c,d,e,f,g,h} in {a,b). If one of them is {a,b,i,j,k,l,m,n}, it
follows tha t the stabilizer of the two sets {C, d, e ,f, g ,h) and
{i,j,k,l,m,n} is a group S6, being by symmetry sextuply transitive on
each of the two sets. (The permutation representations on these sets
cannot be permutation identical, for then - the notation being suitably
chosen - there would be an element of order 3 fixing each of
a,b,c,d,e,i,j,k, and our knowledge of the pointwise stabilizer of
{a, b, c ,d, e} contradicts this.)
Since every dodecad is expressible in this way as the symmetric
difference of two octads with two points in common (by Theorem 9), it
follows that M 24 is transitive on dodecads. Again, any 5 points of a
dodecad belong to one of a pair of octads whose symmetric difference is the
dodecad, and we deduce that the stabilizer of a dodecad is quintuply
transitive on the dodecad (and by symmetry on the complementary
dodecad). (The permutation representations are not permutation identical,
for the subgroup S6 has two orbits of size 6 in one dodecad, but an orbit of
size 2 in the other.)

When we have identified this group with M 12 we can say that the
stabilizer of a point in one of the dodecads is a group M I), triply transitive
on the other. In fact, identification with our previous M 12 is unnecessary,
since we could easily define M 12 to be the stabilizer of a dodecad in M 24.
But it is quite easy - the groups V =, Yo, ... , Vx , W =, W o, ... , Wx of
the first lecture become the points
00,15,7,14,5,10,20,17,11,22,21,19,0,3,13, 18,9, 16,8,4,2, 1, 12,6

of n, and we easily check that the permutations a, fl, 'Y, (j of that lecture
do indeed correspond to permutations of M 24 as defined here. (a and fl
become fl and (j rather plainly, so we need only check the effects of 'Y, (j on
the N j , using the test of Theorem 7 for CC-sets. Only two of the
calculations are really required.) The outer automorphism (J of the first
lecture translates into the operation 'Y of this lecture, interchanging the two
dodecads of the duum, here taken as Nand Q.
Theorem 16. M 24 is transitive on monads, dyads, and triads, whose
stabilizers are respectively groups M 23, M 22 2, and M 21 S3. These groups
are a + b transitive on the two appropriate sets if a + b ~ 5 and
a ~ 1, 2, 3 respectively.
Proof These statements follow at once from the quintuple transitivity of
M 24 ·
Three Lectures on Exceptional Groups 281

If we give the points 2, 3 •...• 22 the coordinates

100,010,001, wwO, 111, wll, lwl, lwl, Oww, llw,
101, wwO, wlw, wOw, 110, wlw, 011, Oww, wll, wOw

then the 21 octads containing 00,0,1 yield the 21 lines of the the projective
plane over F4 = {O,I,w,w}. Using standard techniques we can then
identify M 21 with L3 (4) and deduce the simplicity of M 21> M 22, M 23, M 24.
Theorem 17. M24 is transitive on sextets. The stabilizer oj a sextet is a
group 26 .3. S6. and the subgroup 26 .3 fixing the tetrads individually is
2 + 1 + 1 + 0 + 0 + 0 and 3 + 1 + 0 + 0 + 0 + 0 transitive on the
tetrads (in any order).
Proof The transitivity follows at once from the quadruple transitivity of
M 24 and the fact that a sextet is determined by anyone of its tetrads (as,
say, the family of all tetrads congruent to the given one modulo ~24).
Now in the stabilizer of the sextet containing {a,b,c,d} and {e,j,g,h},
consider the subgroup fixing the four points a, b, c , i . This is contained
within the group H = A8 considered earlier, and so we can see what it is -
a group S3 permuting J,g,h in the obvious way. It permutes the three
tetrads disjoint from {a,b,e,i} in the same way, and so the group fixing
the tetrads individually as well as the 4 points a, b, e, i has order 1. Since
this group has index at most 4·3·4·4 = 192 in the group of all
permutations fixing the tetrads individually, the latter has order at most
192. But the number of sextets is

and so the stabilizer of a sextet has order 244823040/1771 == 192·6!. We

conclude that there are 192 permutations fixing the tetrads individually,
and that all 6! permutations of the tetrads are induced by permutations
fixing the sextet. The transitivity 2 + 1 + 1 + 0 + 0 + 0 is also
established. We leave the transitivity 3 + 1 + 0 + 0 + 0 + 0 to the
reader, and also the exact discussion of the group of order 192 - Todd
remarks that as well as the identity it contains 45 permutations of shape
18 28 ,18 of shape 2 12 , and 128 of shape 16 36 .
Theorem 18. M 24 is transitive on trios. The stabilizer oj a trio is a group
26 (S 3 x L 3 (2». and the subgroup 26 L 3 (2) fixing the octads separately is
2 + 1 + 1 and 3 + 1 + 0 transitive on the three octads (in any order).
Proof The element (abcde )(Jgh)(O (jkl ... x) of shape 1 3 5 15 permutes
the 30 octads disjoint from {a,b,c,d,e,j,g,h} (Theorem 11) in two orbits
of size 15, consisting of those containing, and those not containing, the
point i. Each octad therefore belongs to 15 trios, and M 24 is transitive on
the trios, of which there must be 759·15/3=3795. Since the tetrads of a
sextet can be grouped to form 15 distinct trios, and since
1771·15 = 3795·7, each trio can be refined into 7 distinct sextets. Now
each sextet corresponds naturally to an element of P(f1)/~24
282 Chapter 10

(Theorem 13), and the 7 sextets refining a trio must therefore, together
with the empty set, form a 3-dimensional vector space over F2 (a subspace
of p(n)/C{j24). The stabilizer of a trio has therefore two interesting
normal subgroups - one fixing the octads individually, and another fixing
each of the 7 sextets refining the trio. The quotient by the first is S3, since
all permutations of the three octads are induced, and the quotient by the
second is L 3 (2) or a subgroup thereof, since the permutations of the 7
sextets must preserve the vector space structure. The index of their
intersection therefore divides 6'168, and so the intersection, fixing both the
octads and the sextets individually, has order of form 64n (since
64·6'168'3795 = 244823040). But the intersection is a subgroup of the
group described in the previous theorem, and since an element 16 36 cannot
fix 7 sextets, it has order at most 64. This gives the structure of the group.
We make only the following remarks about the transitivity. If we fix
the three octads individually, and then also a point of one of them, we
obtain a subgroup 23 L 3 (2) of index 8 in 26 L 3 (2), which is represented on 7
points of the first octad (with kernel 2 3) and in two distinct ways on the
8 points of the other two oct ads (faithfully in each case, related by the
outer automorphism of the first lecture). Fixing a point of the second
octad in addition, we come down to a group L3 (2), represented on 7 points
in the first and second octads, on 8 points in the third, and doubly
transitively in each case.
Theorem 19. M24 contains subgroups of type L 2(23), doubly transitive on
n and transitive on the set of759 octads.
Proof Such is the subgroup <a,fl,"!>. The double transitivity on n is
obvious, and the transitivity on octads is proved as follows. An easy if
rather tedious calculation shows that at most 8 linear fractional
transformations preserve some given octad. This octad therefore has at
least! L 2 (23)!/8 = 759 images under L 2 (23).
Theorem 20. M24 contains subgroups L 2(7), transitive on n but with
imprimitivity sets of size 3, not contained in any of the above subgroups
of M 24 .
Proof Consider the permutation II =

(]2 13 14)(21 7 8)(17 I 20)(21915)(63 11)(00 5 10)(1609)(4228),

not in M 24. Direct calculation reveals that it commutes with the

permutation"! and also with the permutation
(12)(21 172600 164)(13)(7 I 1935022)(14)(182015 II 1098)

of M 24 . These generate a subgroup of type L 2(7) whose imprimitivity

blocks are the eight 3-cycIes of II. It is also easy to show that this L 2 (7)
preserves no monad, dyad, triad, octad, sextet, trio, or duum, and
consideration of orders shows that it cannot be contained in the group
L 2 (23).

Thus the new maximal subgroup of M 24 is easily described as the

Three Lectures on Exceptional Groups 283

centralizer in M 24 of the permutation IT, itself not in M 24. So it is the

octern stabilizer, the cycles of IT being ordered triads, or terns.
In Chang Choi's analysis, subgroups of M 24 are divided into three
classes - intransitive, transitive but imprimitive, and primitive - and his
arguments for the second and third classes are very short. Here we present
a quick way to deal with the intransitive groups.
Theorem 21. Any intransitive subgroup of M 24 fixes a monad, dyad,
triad, octad, dodecad, or sextet, and is so contained in one of the groups
of Todd's list.
Proof Let S be a non-empty proper subset of n invariant under the
group. If S E C€24, then S or its complement is an octad or dodecad fixed
by the group. Otherwise S is congruent modulo C€24 to a unique monad,
dyad or triad, or to the six tetrads of a sextet, fixed by the group, as we see
from Theorem 13.
Similar ideas enable us to deal with some of the imprimitive groups, as
we shall see in a moment. They also enable us to give a complete
classification of the subsets of n under the action of M 24.
2.6 The structure of p(n)
Theorem 22. The subsets of n fall into 49 orbits under M 24, related as in
Fig. 10.1. Each node corresponds to one orbit, and the nodes are joined
by lines indicating the number of ways a set of one type can be converted
to one of another type by the addition or removal of a single point. Thus
in an umbral heptad (U7 ), there is one point whose removal leaves a
special hexad (S6), while the removal of any of the six others leaves an
umbral hexad (U 6 ). In the complement of an umbral heptad there are
two points whose addition gives a transverse octad (T s), and 15 that yield
umbraloctads (Us).
Proof We discuss 8-element sets as an example. Each 8-element set S is
congruent modulo C€24 to one of
(j) the empty set,
(ii) a unique 2-element set T,
(iii) each of the tetrads To, ... , Ts of a sextet.
In case (j), S is a special octad S s. M 24 is transitive on these.
In case (ij), S is obtained from the special octad S + T by adding one
point and subtracting another. Since the stabilizer of a special octad is
1 + 1 transitive on it and its complement, M 24 is transitive on sets of this
type, which we call transverse octads T s.
In case (iii) S + T; is a special octad if T; contains two points of S
and two points of its complement, and an umbra I dodecad if T; is disjoint
from S. Counting points of S shows that there are four tetrads of the first
kind and two of the second, so that S can be obtained (in two ways) by
removing four points from an umbral dodecad. Since the stabilizer of an
umbral dodecad is quadruply transitive on the dodecad, M 24 is transitive
on sets like S (which we call umbral octads Us).
284 Chapter 10

In general, a set of cardinal n < 12 is called special (Sn) if it contains

or is contained in a special octad, otherwise umbral (Un) if it is contained
in an umbral dodecad, and transverse (Tn) if not. A non-umbral dodecad
is extraspecial (St2) if it contains three special octads, special (S12) if it
contains just one, penumbral (Ui2) if it contains all but one of the points
of an umbral dodecad, and transverse (T 12 ) in all other cases. Sets of
Three Lectures on Exceptional Groups 285

more than 12 points are described by the same adjectives as their

complements.
Figure 10.1 enables us to read off many interesting relationships at a
glance. Thus on removing any point from a U g we find a U 7 , in which
there is a unique further point whose removal leaves an S 6. The 8 points
of a U g therefore split into 4 sets of 2 in a natural manner, so that the
stabilizer of a Ug acts imprimitively on the Ug. We can also see from the
diagram that the stabilizer of any 8-element set also stabilizes a special
octad. For S g, this is the S g itself; for T g, the unique S g sharing 7 points
with the Tg; and for Ug, the union of the two tetrads whose addition
separately would convert the U g to a U\2.
Theorem 23. If a subgroup of M 24 has two imprimitivity sets of size 12,
or three of size 8, it is contained in one of the groups of Todd's list.
Proof If the 12-ads are umbral, this is obvious. If not, they are congruent
modulo ct' 24 to (the same) one of the sets of size 1, 2, or 3, or to the six
tetrads of a sextet. In the other case, we observe that the group stabilizes
the three corresponding special octads. If the union of these is 0., they
form a fixed trio, and if not, the group is intransitive.
In a similar way, with more complicated arguments, one can handle the
case of four sets of size 6, or six sets of size 4, but other ideas are needed
beyond this point.
All maximal subgroups of the Mathieu groups M 23, M 22, M \2, M IJ are
also known. Table 10.3 gives descriptions. The left hand column gives the
action on the set 0., with semicolons separating the orbits of the Mathieu
group under discussion. The symbol [1; 1; 7, 15] for example indicates a
subgroup of M 22 (fixing the two 1's) that has orbits of size 7 and 15 on the
remaining 22 points. In this case there are two conjugacy classes
according to which of the two fixed points completes the orbit of size 7 to
an octad. Again, the symbol [8, 16] indicates a group with two orbits of
sizes 8 and 16, while [4 6 ] means that the group is transitive, with six

Table 10.3 Maximal subgroups of the Mathieu groups.

Mil

[241 L,(23) [I; 231 23.11 [I; I; 1,211 M21 [12; 121 L,(lI) [I; II; I, III L,(lI)

[1,231 M2J [I; 1.221 M" [I; I; 2, 4'1 2'-S, [I, II; 121 Mil [I; I, 10;6'1 M lo
[2,221 M".2 [I; 2, 211 M21.2 [I; I; 6,161 24.A6 [12; I, III Mil [I; 2. 9; 3'1 M9.2
[3,211 M 21 .S] [I; 3,4'1 2'(3xS,) [I; I; 7.151 A7 [2.10; 6'1 MlO.2 [I; 3. 8; 4. 81 M,.S,
[4 6 1 26 3S 6 [I; 7,161 2'.A 7 [I; I; 7,151 A7 [6'; 2, 101 MlO.2 [I; 5, 6; 2, 101 S,
[8'1 26 (L,(2)xS]) [I; 8, 151 A, [I; I; 8,141 2'.L](2) [3,9; 3'1 M 9·S,
[12'1 M".2 [I; II, 121 Mil [I; I; II, III L,(I I) [3'; 3, 91 M 9·S]
[3'1 L,(7) [I; I; 10.6'1 MlO [4,8; 4, 81 M,.S,
[8, 161 2'.A, [6x2; 6x21 2xS,
[4x3; 4x31 A,xS,
[4'; 4]1 4 2 DJ2
286 Chapter 10

imprimitivity sets of size 4. (Thus [4 6 ] is not the same as [4,4,4,4,4,41.)

The symbol 6 x 2 denotes a set of 12 points with at the same time six
imprimitivity sets of size 2 and two of size 6, so forming a 6 x 2 table in a
natural way. It is very easy to recognize the groups from this information,
so we refrain from further comment.

3. Third lecture
3.1 The group Coo = . a and some of its subgroups. This group is the
group of automorphisms of the Leech lattice in 24-dimensional space R24.
It contains as subquotients 12 exceptional simple groups, and it seems fair
to say that these groups are easiest studied inside' O.
3.2 The geometry of the Leech lattice. Let R24 be spanned by the
orthonormal basis Vi (i E 0 = PL (23», and define Vs to be ~ Vi (i E S)
whenever S ~ O. We use ~24 for the binary Golay code, ~(8) for the
set of special octads, ~(I2) for the set of umbra I dodecads, and finally
o (n) for the set of all n-ads. Let Ao be the lattice spanned by the vectors
2ve for C E ~(8).
Theorem 24. Ao contains all vectors 4VT(T E 0(4», and
4Vi - 4vj (i,j E 0). A vector belongs to Ao if and only if its coordinate-
sum is a multiple of 16 and the coordinates not divisible by 4 fall in the
places of a ~ -set, the coordinates being all even.
Proof If T, V, V are three tetrads of a sextet, Ao contains

We illustrate this by the addition sum

2222 2222 0000 ...
+ 2222 0000 2222 ...
0000 2222 2222 ...
= 4444 0000 0000 ...
and the sum
4 4 4 4 0...
a 4 4 4 4 ...
4 a a a -4 ...

proves the second statement similarly. The final statement then follows
since ~(8) spans ~24 and the set of vectors 4VT and 4Vi - 4vj spans the
lattice of all points with coordinate-sum a multiple of 16 and each
coordinate a multiple of 4.
The Leech lattice A (called A24 in the rest of the book) is the lattice
spanned by Vo - 4voo together with the vectors of Ao. This is the
definition given in §ll (ji) of Chap. 4. The addition sum
-3 1 I 1...
4 -4 a 0 ...
-3 l...
Three Lectures on Exceptional Groups 287

shows that A contains each vector Va - 4Vi'

Theorem 25. The vector (xoo,xo, ... ,X22) is in A if and only if
(i) the coordinates Xi are all congruent modulo 2, to m, say;
(ij) the set of i for which Xi takes any given value modulo 4 is a f{{ -set;
(iii) the coordinate-sum is congruent to 4m modulo 8.
For X, yEA the scalar product X· Y is a multiple of 8, while X' X is a
multiple of 16.
Proof (j), (ij), (iii) and the statement about scalar products are satisfied
by the generating vectors of A, and so for all vectors of A, by linearity. If
x satisfies (i), (ij), Gii), we can subtract a suitable multiple of v n - 4v00

to make all coordinates even and their sum a multiple of 16, and then
x E A by Theorem 24.
We can use these conditions to enumerate the vectors in A of any given
small length. We write A(n) for the set of x E A with X·X = 16n, and we
find, for instance, that A(I) is empty, while A(2) contains 196560 vectors,
namely 27 '759 vectors of shape «±2)8016) (the non-zero coordinates
having positive product and being in the places of an octad), 2 12 '24 vectors
of shape (+3(± 1)23) (the lower sign taken on a f{{-set), and all the
2·24·23 possible vectors of the shape «±4)2022), in an obvious notation.
The corresponding decompositions of A(3) and A(4) are also included in
Table 4.13 of Chap. 4, these sets being orbits under the group N below,
and the signs of the coordinates being suppressed.
3.3 The group ·0, and its subgroup N. We define the group Coo or ·0
(pronounced "dotto") to be the group of all Euclidean congruences of R24
that fix the origin and preserve the Leech lattice A as a whole. Since each
vector 8Vi is in A, the elements of . 0 have rational orthogonal matrices in
which all the denominators divide 8. If 7r is a permutation of fl, we
extend 7r to a congruence of R24 by defining Vi7r=Vi ... , and if S ~ fl, we
define a congruence f s by Vif S = Vi (j ~ S) or -Vi (j E S).
Theorem 26. The following conditions on an element A of . 0 are
equivalent:
(j) ViA = ± Vj for some i,j E fl, and some sign ±;
(ii) A = 7r f C for some 7r E M 24 and some C E f{{ 24.
These operations form a subgroup N = 2 12 M 24 .
Proof Plainly (ii) implies (j). We show (j) implies (ii). Now (j) implies
that the ith row of the matrix of A contains a single non-zero entry ± I in
its jth place. Since A is orthogonal, the jth place of every other row must
be zero. Now the image of 4Vi + 4Vk under A is 4 times the sum of the ith
and kth rows, and so has a coordinate ±4. Since this vector must be in
A(2), it has shape «±4)2022), and the kth row has also a single non-zero
coordinate ± I. It follows that A = 7rfs for some permutation 7r and some
set S. But if C E f{{24, the non-zero coordinates of 2VcA are in the places
of C 7r, so 7r preserves f{{ 24, and is in M 24. Again, the coordinates
congruent to 3 modulo 4 in (v n -4V oo )A are in the places of S, so that
S E f{{24, and A satisfies Gi). The final statement is obvious.
288 Chapter 10

Suspecting that N is a proper subgroup of . 0 we seek an additional

operation. If· 0 is transitive on A(4), it will contain an element A taking
8v to some vector 4VT, and a row in the matrix of A will then have '/2 in
00

every position of T and zeros elsewhere. But 4 times the sum or difference
of this row and any other must be a vector of A (2), and this limits the
possibilities - if this vector of A(2) has shape ((±2)80'6), the other row
has four entries ± '/2 in the places of some tetrad from the same sextet as
T, and if the shape is (( ± 4) 20 22 ), we get entries ± '/2 in the places of T.
This suggests that we try a matrix which is the direct sum of six 4 x 4
matrices of ± V2's in the places of a sextet. The signs must be chosen
carefully - if 2 is a sextet we let 7/ = 7/:=; be the map taking Vi to
Vi - '/2 VT (j E T E 2), and then define ~ = h to be 7/ ~ T, 2 being the sextet
of T.
Now since two octads intersect in 0, 2, 4, or 8 points, any octad is
either the union of two tetrads of 2 or contains two points from each of
four tetrads, or has three points from one tetrad and one from each of the
others. Supposing the coordinates properly ordered, we apply 7/:
x = 2222 2222 0000 0000 0000 0000
X7/ = 2222 2222 0000 0000 0000 0000
x = 2220 2000 2000 2000 2000 2000
x7/=1113 1111 1111 1111 1111 1111
x = 2200 2200 2200 2200 0000 0000
X7/ = 0022 0022 0022 0022 0000 0000
-
x=3111 1111 1111 1111 1111 1111
-
x7/=3IlI 1111 1111 1111 1111 1111

(ii denotes - n). Since the vectors x 7/ are not all in A, 7/ is not in . 0, but
on changing the sign of any tetrad we get a vector of A, and so ~ = 7/ ~ T is
in ,0.
In [Con3] we proved that N is a maximal subgroup of . 0, and at the
same time computed the order of . O. The following method involves
explicit calculations with the element h, but is perhaps as elegant.
Theorem 27. ·0 is transitive on each of the three sets A(2), A(3), A(4),
and N has index IA(4) I /48 in . 0, which is generated by N together with
any element h.
Proof We have already shown that h takes certain elements of A(2)2
(see Table 4.13) to elements of A(2)3, and in a similar way we see that if i
and j are in distinct tetrads of 2, h takes 4v; +4vj to a member of A(2)2,
so that the subgroup <N,h> of ·0 is transitive on A(2). Similar
calculations (they are all easy) establish the transitivity on A (3) and A (4).
Now the particular vector 8v has just 48 images under N (the vectors
00

± 8v), and every operation of . 0 taking 8v to one of these is in N. From

00

these remarks the rest of the theorem follows. (If A E . 0 we can find
J.LE <N,h> with 8vooJ.L=8vooA, so AENJ.L.)
Three Lectures on Exceptional Groups 289

Thus Coo = ·0 has order

222 . 39 . 54 . 7 2 . 11 . 13 . 23 = 8315553613086720000.
The permutation representation of ·0 on A(4) is imprimitive, since the
vectors of A(4) come in naturally defined coordinate-frames, each
consisting of 24 mutually orthogonal pairs of opposite vectors. We can
distinguish these frames geometrically - the vectors of the frame
containing x are all those vectors of A(4) that are congruent to x modulo
2A. To prove this, it suffices by transitivity to consider the particular case
x = 8v ...
Theorem 28. Every vector of A is congruent modulo 2A to one of
(j) the zero vector;
(jj) each vector of a unique pair x, -x (x EA(2»;
(jij) each vector of a unique pair x, -x (x E A(3»;
(jv) each of the 48 vectors of a coordinate-frame in A(4).
Proof Let x,y be two vectors of A(O) U A(2) U A(3) U A(4) congruent
modulo 2A, with y;z! ± x. Then since x ± y E 2A we have
(x ±y) . (x ±y) ~ 128, whence x·y = 0 and x·y = y.y = 64, since we
know that x . x and y . yare both at most 64. It follows that x and yare
members of A(4) both in the same coordinate-frame. We therefore have at
least IA(O) I + IA(2) 1/2 + IA(3) 1/2 + IA(4) 1/48 distinct classes of A/2A,
and since this number comes to exactly 224 we have accounted for every
class.
Let us say that a vector x E A(n) (of norm 2n) has type n, and that x
has type nab if it is also the sum of two vectors of types a and b.
Theorem 29. Every vector x of type n has some type nab in which
a + b = ~(n +k), where k = 0,2,3 or 4 (corresponding to the cases of
Theorem 28), and these possibilities are exclusive. ·0 is transitive on
vectors of each of the types
2, 3, 4, 5, 6 22, 6 32, 7, 8 22 , 832 , 842 ,9 33 , 942 , 10 33 , 1042 , 10 52, 11 43 , 1152

(which include all vectors of type n < 12).

Proof For the first part, we suppose x congruent to
y E A(O) U A(2) U A(3) U A(4) modulo 2A, and let ~ (x +y) and ~ (x -y)
have types a and b. For the second part, we give only a few typical
samples.
Type 6 32 • Each such vector is x-y, where x EA(3), y EA(2), x-y=-8,
and so x +y E A(4). (This follows from the first part by a sign-change.)
To within transformation by some element of ·0, we can take x +y as
(8,0,0, .. .), from which we get x - (5, 1,1, .. .) E C, Y = (3,1,1, .. .) E C for some
C E CC24, 00 ~ C (this is the only way to express (8,0,0, .. .) as x + y,
x E A(2), y E A(3». To within a further transformation we can suppose C
empty, so that x - y is equivalent under· 0 to (2,2,2, .. .).
Type 5. In this case the typical vector is x - y, where x,y E A(2),
x·y = -8, so that x + y E A(3), and can be taken as (5,1,1, .. '). From
290 Chapter 10

this, without loss of generality, we get x = (3,(-1)7,1 16), Y = (2 8,0 16) or

vice versa, or x = (4 2,0 22), Y = (I, - 3,122) or vice versa, so that x - y is
one of ±(I,(-3)7,116) or ±(-3,-7,122), the coordinates -3 in the first
case being in a special heptad S7. Applying a suitable h we see that
1333 3333 1111 1111 1111 1111,
-
3111 7111 1111 1111 1111 1111
become
5111 3333 1111 1111 1111 1111,
-
3111 5333 1111 1111 1111 1111,

which are plainly equivalent under M 24.

These proofs get easier as we go further, since we then have more
transitivities already known. We can avoid explicit calculations with h by
using instead the counting methods of [Con3].
3.4 Subgroups of . O. Many subgroups of . 0 are easiest considered in
relation to the infinite group Co"" or . 00 of all euclidean congruences of A,
including translations. Any finite subgroup G of . 00 fixes a point (not
necessarily a lattice point), and so there is a translation t of R24 (not
necessarily in . 00) such that G 1 ~ ,0. If S is the name of a simplex, we
use'S for the subgroup of those elements of . 00 which fix every vertex of
S, *S for the subgroup of elements fixing S as a whole, and !S for the
subgroup fixing the centroid of S. Plainly' S ~ *S ~ !S . A simplex is
then named by the types of its edges. Thus the stabilizer of two points
whose difference is a vector of type n is written 'n, the stabilizer of the
vertices of a triangle whose sides have types a, b, c, is written 'abc, and so
on - see Fig. 10.2. The particular groups . 1, . 2, . 3 are also called Co I,
Co 2 , Co 3 , respectively.

D a

Figure 10.2.
Three Lectures on Exceptional Groups 291

These groups are identified in Table 10.4. Many of the identifications

are easy. Thus· 632 can be taken as the stabilizer in ·0 of the vector
(2,2,2, .. J. The argument of Theorem 29 shows that this is expressible in
just 24 ways as a sum x + y (x E A(3),y E A(2», and we can see all of
them: x = (5,1 23 ), y = (-3,1 23 ) and their images under coordinate
permutations. Since the vectors x - yare the 8v;, we have .632 !: N, and
so .6 32 = M 24. At the same time, we get· 632 ., M 23, stabilizing the same
two points together with (5, }23). The groups· 632 and· 432 are identical,
since a parallelogram with sides .J2 and .J3 and one diagonal .J6 has J4
for the other diagonal. Since· 432 = ·632 = M 23, the group M 23 must be
contained in each of the groups· 2, ·3, ·4, .6 32 - we can see this also by
taking suitable coordinates for the vectors involved, namely (-3, }23),
(5, }23), (8,0 23 ), and (224).

Table 10.4 Various groups associated with the Leech lattice. HS, MeL,
p., pI +2. denote respectively the Higman-Sims group, the McLaughlin
group, an elementary abelian group of order p., and an extraspecial
group of order p2.+1 respectively.

Name Order Structure Name Order Structure

·0 2223 9547211.13.23 2.Co l ·222 2 15 36 5.7.11 psu6 (2)

·1 22139547211.13.23 COl ·322 2 736 537.11 McL
·2 2 18 36 537.11.23 CO 2 ·332 29 32 537.11 HS
·3 2 10 375 37.11.23 C0 3 ·333 24 37 5.11 35 .M 11
·4 2 18 3 25.7.11.23 211 M 23 ·422 2 17 32 5.7.11 21O .M22
·5 2 836 5 37.11 McL.2 ·432 2 73 25.7.11.23 M 23
.6 22 2 16 36 5.7.11 PSU6 (2).2 ·433 2103 2 5.7 24.A8
.6 32 2103 3 5.7.11.23 M24 ·442 2 12 32 5.7 2 1+8.A 7
·7 29 32 537.11 HS ·443 27 325.7 M21·2
.8 22 2 18 36 537.11.23 CO 2 ·522 27 365 37.11 McL
.8 32 2 7 36 537.11 McL ·532 28 36 5.7 psu4(3).2
.8 42 2 15 3 25.7 2 1+8 .A8 ·533 2432537 PSU J (5)
.9 33 2 5 375.1 I 3 5.M II .2 ·542 27 3 25.7.11 M22
.9 42 27 32 5.7.11.23 M 23 ·633 26 335.11 M12
.10 33 2103 2537.11 HS.2 *2 =!2 2 19 36 5 37.11.23 (·2) x 2
.10 42 2 17 32 5.7.11 2 1O .M22 *3 -!3 2 11 375 37.11.23 (-3) x 2
.11 43 2103 25.7 24.A8 *4 2 19 325.7.11.23 (·4) x 2
. 1152 28 36 5.7 psu4(3).2 !4 2223 35.7.11.23 212.M24
!333 27 39 5.11 36 .2.M 12
!442 2 15 34 5.7 2 1+8.A 9
292 Chapter 10

3.5 The Higman-Sims and McLaughlin groups. In a similar way we see

that M 22 is contained in each of the groups' 222, . 322, . 332. Since' 222 is
the group PSU6 (2) it follows that M 22 has a 6-dimensional projective
representation over F4 (and from this it follows fairly easily that the
multiplier of M 22 has order divisible by 3).
D. G. Higman and C. C. Sims [Hig3] have described a simple group of
order 44352000 = 100lMd as the group of even permutations of a certain
graph on 100 vertices, and 1. McLaughlin [McLl] has discovered a simple
group of order 898128000 which is a group of automorphisms of a graph
on 275 vertices. We here identify the Higman-Sims group HS with our
. 332, and sketch the identification of McLaughlin's group MeL with' 322.
Let X = 4v; + Va, Y = 4vj + Va, Z = 0, where i and j are distinct
monads, so that XYZ is a triangle of type 332. Then there are exactly 100
points T for which XYZT has type 332222, namely the point
P = 4v; + 4vj, 22 points Qk = Va - 4Vk (k E n \ {i, j}), and 77 points
RK = 2VK ({i, j} ~ K E rt'(8». If we say that two of these points are
incident when their difference has type 3, then the incidences are (p, Qk),
(Qk. R K ) (k E K), and (R K , R K,) (K n K' = {i, j}), and the incidence
graph is visibly identical with the Higman-Sims graph.
Now X - Y + P - Z = 8vj, so that the stabilizer in' oo of X,Y,Z,P
is a subgroup of N, and in fact this stabilizer is the group M 22 of
permutations of n that fix i and j. We identify . 332 with the Higman-
Sims group, and at the same time provide an easy proof of the latter's
existence, by showing that· 00 has operations fixing X, Y, Z but disturbing
P. For let X be an operation of . 00 such that XX = 2vc, YX = 2VD,
ZX = 0, where C E rt'(I2), D E rt'(I2), C + D E rt'(8). (Such X exist
by the transitivity of . 00 on triangles of type' 332.) Then the subgroup H
of those operations of N that fix each of XX, YX, ZX fixes no one of the
100 points TX that differ by type 2 vectors from each of XX, YX, ZX, and
so the group XHX- 1 has operations that fix each of X,Y,Z, but not P.
(The argument has proved transitivity of . 00 on tetrahedra of type 332222
given transitivity on triangles of type 332.)
Sims [SimI] has shown that a doubly transitive group on 176 letters
described by G. Higman [Hig4] is isomorphic to the Higman-Sims group.
(See also [Smi8]-[SmilOJ.) G. Higman's group is the automorphism group
of a "geometry" of 176 "points" and 176 "quadrics," there being 50 points
on each quadric and 50 quadrics through each point. Now there are 352
points G such that GX has type 3 while GY and GZ have type 2, and
these naturally form 176 pairs PK = {AK,B K} (K E rt'(8),
K n {i, j} = (i}), AK = 2VK, and BK = X - A K . There are similarly 176
pairs QK = {CK , D K } obtained by interchanging the roles of i and j. If we
regard "points" as the pairs PK , and "quadrics" as pairs QK, and say that
PK is on QK' just if IK n K'I = 2, then we obtain G. Higman's geometry
and another proof of the identity of his group with the Higman-Sims
group.
If we consider instead a triangle of type 322, we find that there are just
275 points which complete it to a tetrahedron of type 322222, and that the
Three Lectures on Exceptional Groups 293

incidence graph (where x and yare incident if and only if x - y has type
3) is now McLaughlin's graph. The details of the identification are rather
complicated, but the result is a fairly simple definition of McLaughlin's
graph. If we take the triangle XYZ to be X - 0, Y = 4Vi + V 0,
Z = -4vj + Vo (i ¢ j), then the 275 points fall naturally into the
following three sets: 22 points Uk (k E n \ (;, j}), 77 points
VK ( {i, j} ~ K E rt'(S) ), and 176 points GK , (K' E rt'(S) ,
{; , j} n K' = {;} ), and the incidences can be simply described by
combinatorial conditions on k, K, K'.

3.6 The group C0 3 = . 3. We shall also show that C0 3 = . 3 has a doubly

transitive permutation representation on 276 letters, the stabilizer of a
point being a group McL.2. To this end we consider the stabilizer of the
vector x = (5, 123 ). There are 276 unordered pairs {y ,z} with x = y + z,
where y,z E A(2), namely 23 with y (say) of shape (4 2,0 22 ) and 253 with
y of shape (2 8 ,0 16), the coordinate y .. being non-zero in each case.

Letting {Yo,zo} be one of these pairs, we have for each other pair {y,z}
that either z - Yo or Y - Yo is of type 2. Assuming transitivity of . 0 on
triangles 322 and tetrahedra 322222 then suffices for the result: . 3 is
doubly transitive on the 276 unordered pairs {y ,z}. The stabilizer in . 3 of
the point Yo has index 2 in the stabilizer of the pair {Yo,zo}, and is visibly
. 322. If we take Yo = 4v .. + 4vo the situation is invariant under the group
M 22 fixing 00 and 0, and the 276 pairs fall into orbits of sizes 1, 22, 77,
176 under this group, which is contained in three distinct subgroups of . 3
- an M 23 with orbits 1 + 22 and 77 + 176, the McLaughlin group ·332
with orbits 1 and 22 + 77 + 176, and a Higman-Sims group . 332 with
orbits 1 + 22 + 77 and 176, the latter neatly displaying both the 100 letter
representation of Higman and Sims and the doubly transitive 176 letter
representation of Higman.

A pretty series of subgroups of . 0 arises in the following way [Th07].

The centralizer of a certain element x of order 3 in . 0 has the form
<x> x 2A 9, the 2A9 being the Schur double cover of A9, and containing a
natural sequence of subgroups 2An (2 ~ n ~ 9). The centralizers Bn of 2An
are, for n = 2,3 •...• 9, groups '0, 6Suz, 2G 2 (4), 2HJ, 2U3 (3), 2L 3 (2),
2A 4, C 6 , where HJ (also called h) is the Hall-lanko simple group [Ha14]
and Suz is Suzuki's sporadic simple group [Suzl1. It follows that HJ has
a multiplier of order divisible by 2, and S a multiplier of order divisible by
6, and also that HJ has a 6-dimensional projective representation that can
be written over Q(F-3, ~), while Suz has a 12-dimensional projective
representation over Q(F-3). The latter can be obtained as follows. Take
an element w of order 3 with no fixed point, and so satisfying (as a matrix)
the equation w2 + w + 1 = O. Then, in the ring of 24 x 24 matrices, w
generates a copy of the complex numbers in which it is identified with
e 21<;/3. If we define, for x E A, x(a+be 21ri / 3) == ax + b(xw), the Leech
lattice becomes the complex Leech lattice Ac (see Chap. 7, Example 12),
a 12-dimensional lattice (or module) over the ring Z[e 21ri / 3 ] of Eisenstein
integers, and the automorphism group of Ac is the group 6 Suz. The
complex Leech lattice has a natural coordinate system that displays a
294 Chapter 10

remarkable analogy between it and the real Leech lattice, with 2

everywhere becoming f} = R = w - w2 (see Table 7.3).
The details are as follows. Define vectors Xi, Yi, Zi (; E PL (I 1)) by

Let w be the operation of ·0 that takes Xi - Yi - Zi - Xi for each i. In

terms of the coordinates Xi, and the ternary Golay code rc 12 , Ac is now
spanned (up to a scale factor) by the vectors f}xc (C E rc\2), 3Xi - 3wS xj'
Xn+3Xi (!l=PL(I1)). These are obtained from 2vc, 4Vi+4vj,
v n - 4Vi in the real case by replacing 2 by f}, 4 by f}2 = -3, and ± 1 by ws •
It is interesting that the order of the binary Golay code is 212 , and that of
the ternary Golay code is f}\2 = 36, and that 6 Suz is generated by its
monomial part together with a matrix of 3 x 3 blocks whose entries have
the form ±wi /f}. (Compare the real case, when we have 4 x 4 blocks with
entries ± 1 /2.)
3.7 Involutions in ·0. Each involution in ·0 is conjugate to an involution
~ c (C E rc24). If C is an octad, the centralizer preserves the corresponding
8-dimensional subspace, which intersects A in a copy of the 8-dimensional
lattice Es. The subgroup of the centralizer that fixes every point of this
space is therefore in N (since the space contains vectors Vi), and is easily
seen to be an extraspecial group 21+S of order 29. The quotient by this
extraspecial group is a subgroup of the Weyl group of E s, and in fact it is
the derived group W(E s)', so that the whole centralizer is a group
2I+S(W(E s))'. If C is a 16-ad, we get the same centralizer.
If C is a dodecad, the fixed space can contain no vector of shape
«±2)SOI6) or «+3)(±1)23), and so contains in A(2) only the 22.66
vectors of shape «±4)2022) whose non-zero coordinates are in the fixed
space. We call two such vectors skew unless they are equal, opposite, or
orthogonal. Then the only vectors of the set which are skew to all the
vectors skew to X = 4Vi + 4vj say, are just the four vectors ±4Vi ± 4vj
having the same non-zero coordinates as x. It follows (since
(4Vi + 4vj) + (4Vi -4vj) = 8v) that the centralizer of ~ c is in N, and so is
the group 212M\2 centralizing ~ c in N. (The centralizer of this type of
involution ~ c in . 3 is < ~ c> x M \2, as is seen by taking the fixed vector
of ·3 to be (2 12 0 12) - compare the situation in lanko's group J I.)
The only remaining involution in ·0 is ~ n = -1, which is central in the
whole group.

3.8 Congruences for theta series. Let

E>A(Z) =}; N(m)q2m = 1 + 196560q4 +

be the theta series of A (Eq. (I38)-(I41) of Chap. 4). From §6 of Chap. 7

we know that E> A(z) is a modular form of weight 12 for SL 2 (Z). Since the
Eisenstein series 1 + C}; 0"1l(m)q2m (c=65520/691) and the cusp form
Three Lectures on Exceptional Groups 295

Mz) = '1;T(m)q2m are modular forms of weight 12, it follows from

Theorem 17 of Chap. 7 that
65520
N(m) = 691 (0"11 (m) -T(m)),

where the T(m) are the Ramanujan numbers (Eq. (37)-(39) of Chap. 4).
Ramanujan's remarkable congruence T(m) == 0"11 (m) (mod 691) is
particularly evident, and indeed we can use the formula to find congruences
for T(m) modulo any prime power dividing g/c = 218 37 537.11.23.691
(where g is the order of ·0). Thus to modulus 23 we have
N (m) == N (m) (a), where N (m) (a) is the number of vectors of A(m) that
are fixed by the element a of order 23. But the fixed vectors of a are those
of the shape av + bv n\
00 and since this has type m if and only if
00,

16m = a 2 + 23b 2 we obtain T(m) == O"l1(m) == 0 (mod 23) whenever m is

a non-residue modulo 23.
3.9 A connection between . 0 and Fischer's group Fi 24. ·0 can be regarded
as a permutation group on the 196560 vectors of A (2). It has a subgroup
N = 212M24 that permutes these in three orbits of sizes 27 .759,2 12 .24,
and 22(224). The elementary part 212 of N is permuted by M24 like the
binary Golay code ct'24. The extension 212M 24 is split.

The group Fi 24 of B. Fischer [Fisl1 is a permutation group on 306936

objects (the involutions in a certain conjugacy class). It has a subgroup
N* = 212M 24 that permutes these in three orbits of sizes 25.759, 2°· 24,
and 2 10 (224). The elementary part 212 of N* is permuted by M24 like the
quotient P(O,)/ct'24. The extension 212M24 is non-split.
These facts suggest that there is a relation between the two groups, and
this is borne out by the fact that the subgroups of M 24 corresponding to
the three orbits are identical, while the elementary subgroups are dual.
The appropriate name for this relationship - whatever it is - is twinning,
as we see from the following parable. "Once upon a time there was an egg
(the elementary group p(n) of order 224), which after fertilization (by the
action of M 24 ) split into two (the group ct'24 and its quotient p(n)/ct'24),
which grew into two healthy twins (·0 and Fi 24 )."
That the parable is meaningful is suggested by the existence of similar
structure relating· 0 to itself via its subgroup 6 Suz, with elements of order
3 like w taking the place of Fischer's involutions. The elementary 212 in
the subgroup N* of Fi24 has 24 of Fischer's 3-transpositions (the special
involutions of his group), permuted in the natural way by M 24, the
centralizer of anyone being a group 2Fi 23. In a similar way ·0 has a
group 36 .2M 12 in which the 36 has 12 special subgroups of order 3
(isomorphic to the groups Vi of the first lecture) permuted naturally by
M 12, the centralizer of anyone being the group 6 Suz.
Acknowledgements. I should like to thank D. Livingstone and his
colleagues for some remarks which I have made use of in the second
296 Chapter 10

lecture, and in particular, for the material of Table 10.3, and J. G.

Thompson for his sustained interest in the subject of the third lecture.

Appendix on the exceptional simple groups

We discuss under six headings the 26 non-abelian simple groups not
occurring among the infinite families of Chevalley and twisted Chevalley
groups and alternating groups. With each group we give its order,
preceded by its Schur multiplier and followed by its outer automorphism
group (see [ConI6], [Gril], [Gri3], [Gri9]). Asterisks indicate groups that
belong to one of the infinite families but are included here by analogy.
The Mathieu groups

M24 1(2 10 33 5.7.11 ·23) 1 M12 2(2633 5.11)2

M 23 1(2732 5.7.11.23)1 Mil 1(24 32 5. 11) 1
M22 12(2732 5.7.11)2 *A 6 6(2 3325)2 x 2
*M 21 4 x 4 x 3 (2 632 5.7) 2 x S 3 *A 7 6(2 332 5.7)2

We have already discussed the isomorphisms M21 == L 3(4) and MIO' == A6

== L 2 (9), but the unusual multipliers show that the usual definitions of
these two groups are unrevealing. For alternating groups other than A6, A7
our symbol reads 2(n!/2)2. The groups under the next two headings are in
some sense the Mathieu groups "writ large".
The Fischer groups

Here Fin has a conjugacy class of involutions anyone of which has

centralizer a group 2Fin- 1 and any two of which have product of order at
most 3. A maximal commuting set of such involutions has exactly n of
them, generating a group 2n - 12 whose normalizer is a group 2n - 12M n • We
have the isomorphism Fi21 == U6 (2). (See also [Con71, [Enrl], [Enr2],
[Hunl], [Par3].)
Lattice stabilizers in . 0
·1 = COl 2(22139547211.13.23)1 *·222
·2 = CO 2 1(2 18 36537.11.23) 1 ·322 == MeL 3(27 36 537.11)2
·3 = C0 3 1(2 10 37 537.11.23)1 ·332 == HS 2(2 932537.11)2

The first group contains M 24, the next two contain M 23, and the remainder
contain M 22 • We have the isomorphism ·222 == U6 (2). (See also [Fin2].)
Three Lectures on Exceptional Groups 297

The Suzuki chain

Suz 6(2 13 37 527'11'13)2 2(2 733527)2

2(2 12 33527'13)2 1(2 5 337) 2

These are the central quotients of the groups Bn described in §3.6. We

have the isomorphism u 3 (3) ;;;; G 2 (2)/.
It is perhaps worthy of note that the group u4 (3) has an unusual
multiplier (3 x 3 x 4). The group 32 U 4 (3) is the centralizer of two
commuting elements of order 3 in . 0, and U 4 (3) is closely connected with a
number of the groups above.
Centralizers in the Monster
M = F1 1(24632°597611213317'19'23'29' 31·41·47· 59'71)1
B =F2+ 2(241313567211'13'17'19'23'31'47)1
COl = F 2- 2(22139547211'13'23)1
Fi24 = F3+ 3(221316527311'13'17'23'29)2
Suz = F 3- 6(2 13 37 527'11'13)2 He = F7+ 1 (2 1°3 3527317) 2
6(2 3325' 7) 2

HJ = F 5-

The Monster group M, which was discovered a few years after these
lectures were given, led to the discovery of several additional simple groups
and casts further light on some of those already known. These groups
usually arise from studying the structure of centralizers of suitable
elements of the Monster. Here we consider elements of prime order, and
write Fp+, Fp_ etc. for the nonabelian composition factor in the centralizer
of an element in the conjugacy class of M called p+, p- etc. in [Con161,
[ConI71. Some of the groups that are obtained in this way are shown
above.
M = FJ, the Monster, or Fischer-Griess group, or Friendly Giant, was
discovered independently by Fischer and Griess in 1973, and constructed
by Griess in 1980 [Gri41-[Gri8l. A simplified construction is described in
Chap. 29. The baby Monster B = F2+ was discovered by Fischer slightly
earlier and was constructed by Leon and Sims [Le091, [Sim31.
The existence of the Thompson group Th = F313 (constructed by 1. G.
Thompson and P. E. Smith [Th02l) and of the Harada-Norton group
HN = F 5+ (constructed by Norton [HarOl, [Harll, [Nor21, [Nor8]) was
suggested by that of the Monster, although the entirely analogous groups
of Fischer (Fi/ 24 = F3+) and Held (He = F7+) were already known [Helll,
[Hel2l. All of these groups are best understood in terms of their
relationship with the Monster. We remark that COl = F 2-, Suz = F 3-,
HJ = J 2 = F 5-, M12 = F I1 +.
298 Chapter 10

The remaining groups

J, 1(2 3 3'5'7'11'19)1 Ru 2(2'43 3 '5 3 '7'13'29)1

J 3 3(27 355'17'19)2 ON 3(2 9335'7 3 '11'19'31)2

Most of these were discovered via their centralizers of involutions.

J 2 = HJ (the Hall-lanko group) and J 3 (the Higman-lanko-McKay group
([Hig5), [Jan4), [Con45), [Gor6), [Lin4), [Lin5), [Smi4), [Tit4), [Wall),
[Wei3)-[Wei5D, have involutions with the same centralizer, as do He,
M 24, and L 5 (2). The centralizers of suitable elements of order 3 are 3A 7 in
the Held group He, and 3McL in the Lyons group Ly ([Lyoll, [Sim2D,
exhibiting the 3-parts of the multipliers of these groups. J, is a subgroup
of G 2 (I 1), and G 2 (5) is a subgroup of Lyons' group.
The three remaining groups are lanko's group J 4 [Jan6), [Conl2J,
[Nor3), Rudvalis's group Ru [ConIO), [Con44), [Rud3), and Q'Nan's
group ON [And2), [Q'Nall. It seems that the simplest way of computing
with all six groups is to use matrices over various finite fields. The groups
J, 3.J 3 J4 Ru 30'N Ly

have representations of degrees

7 9 112 28 45 111

over the fields of orders

11 4 2 2 7 5

respectively (cf. [ConI6), [Evall, [Mey3), [RyblD.

Postscript. The maximal subgroups of . 1 == Co, have been completely
enumerated by Wilson [WiI4), following earlier work of Curtis [Cur2],
[Cur51. Maximal subgroups of other sporadic simple groups have been
classified in [Butl), [Fin2)-[Fin4), [Klel), [Leml J, [Nor8), [Will)-
[Wi1l7J, [WolO), [Yos21. Character tables and much other information
about sporadic groups will be found in [Con161. General references for
simple groups are [Car2), [Car3), [Cox28), [Gor2)-[Gor5), [Tit3)-[Tit9).
See also [SoiIl, [Soi31.
 11
The Golay Codes and The Mathieu Groups
1. H. Conway

This chapter contains a detailed description of the binary Golay code of

length 24, the Steiner system S(5,8,24), and the Mathieu group M 24 • The
MOG (Miracle Octad Generator) and the hexacode are computational
tools that make it easy to perform calculations with these objects. The
MINIMOG and the tetracode perform similar services for the ternary
Golay code of length 12, the Steiner system S (5,6, 12), and the Mathieu
group M 12 .

1. Introduction
The Golay code ~ 24, the Steiner system S (5,8,24), and the Mathieu
group M 24 are beautiful combinatorial objects with a great wealth of
structure and applications. The MOG, and its companion-at-arms the
hexacode, are computational tools that enable one to perform mental
calculations on these objects with great ease. In particular, one can check
at sight that a word is in ~24, complete an octad of S (5,8,24) from any
five of its points, or write down many permutations of M 24 in a few
moments. There is also a MINIMOG, with companion the tetra code,
which together perform similar services for M \2. Some discoveries about
the "lexicography" of M 24 and M \2 indicate that these devices may have
some theoretical as well as practical value.
Many calculations in the larger sporadic simple groups reduce to
calculations inside M 24, which has consequently been used in a number of
constructions. In particular, we mention the construction of the Conway
groups via the Leech lattice (see Chap. 10), the Fischer groups Fi 24, Fi 23,
Fi22 (§3.9 of Chap. 10), the Janko group J 4 [ConI2], [Nor3], and most
recently Griess's construction of the Monster group (see [Gri4]-[Gri8] and
Chap. 29).
The MOG array was invented by R. T. Curtis [Cur2], [Cur3] in the
course of his work on the Leech lattice, and he and others have made
300 Chapter 11

valuable use of it for many other investigations involving M 24. The special
virtues of the approach via the hexacode emerged much later, during the
construction by Norton, Parker, Benson, Conway and Thackray of the
largest lanko group J 4 [ConI2], [Nor3]. Although that company included
several practiced MOG users, it was found that the hexacode effected still
greater simplifications in the enormous number of M 24 -related calculations
that were required in the course of the construction. More recently, it has
performed a similar role for the many calculations with the Leech lattice
that were carried out in the determination of its covering radius and the
enumeration of its deep holes (see Chap. 23).
There are almost as many different constructions of M 24 as there have
been mathematicians interested in that most remarkable of all finite
groups. Usually, when one constructs a large group from some "simpler"
object, the construction itself will have a smaller group of symmetries, and
one finds that elements inside this "visible subgroup" are "easy" to
understand, and the others "hard." It is a remarkable fact that the MOG
construction manages to have several "visible groups," each of them a
maximal subgroup of M24! We shall construct it here in a way that is
specially related to the sextet group 26: 3'S6, since that offers maximal
opportunities for the use of the hexacode, which is the main topic of this
chapter. However, we feel that the reader should be aware that the MOG
was first used by Curtis in connection with the octad group 24:A g, that
when he wrote it up it seemed easiest to use the trio group 26: (S3XL2(7»,
and that still later the triad group prL 3 (4) was found to be just as simply
related to the MOG. In fact much of its power stems from the ability to
switch groups while seeking the element one desires.
In the first part of the chapter we shall describe the construction of the
hexacode and MOG, and show how to use them for the basic calculations
that arise in every investigation with M 24. The later part has a section for
each maximal subgroup, described where possible in close relation to the
hexacode or MOG.

2. Definitions of the hexacode

As usual F4 = {O, I,w,w}, with the relations
I + w = w, 1 + w = w, w + w = ww = I , w 2 = w, w 2 = w, w 3 = 1 .

The hexacode CC6 (§2.S.2 of Chap. 3) is a 3-dimensional code of length 6

over F 4. In order to display certain symmetries, we usually group the 6
digits of a hexacodeword into 3 couples. The code may be defined in
several different ways:
Definition 1. CC6 is generated by the words

,
- ww- ww,
ww - ww ww ww, ww ww ww, ww ww ww .

No No Yes Yes No Yes •

The Golay Codes and The Mathieu Groups 301

Definition 2. Cff6 has a word W(ljJ) = ab cd el for each quadratic

function ljJ (x) = ax 2 + bx + c defined over F4. The first three digits
(a,b,c) specify the function ljJ, and the last four (c,d,e,f) give its values
at 0, 1,w,w:

Definition 3. Each word ab cd el of Cff6 has a slope, s. Then ab cd el is

a word of Cff6 with slope s if and only if it satisfies the following relations:
The I-rule: a + b = c + d = e + 1 = s,
Thew-rule: a +c +e =a +d +1 =b +c +1 =b +d +e -ws,
The w-rule: b + d + 1 = b + c + e = a + d + e = a + c + 1 = ws.
(Thus the digits of each couple add to s, while three digits from distinct
couples add to ws or ws according as it is the left or right possibility that
is chosen an odd number of times'>
The code under any of these definitions is 3-dimensional; in the first
case because the four given words add to zero, but obey no further linear
relation; in the second case obviously; and in the third case since the
relations
a+b=c+d=e+l=s, a+c+e=ws

between the seven quantities a, b, c ,d, e,f, s imply all the rest. (The first
three of these imply that the replacement of any digit by its mate will alter
the sum by s.>
Given the 3-dimensionality, the identity of the three codes follows
readily by checking that some 3 independent words can be found in all
three codes. We suggest that the reader checks this for the generators of
the second version:
W(x 2 ) = 1001 ww, W(x) = 01 01 ww, w(1) = 00 11 11 .

From either the first or the third definition, it is obvious that Cff6 has the
following symmetries:
1) scalar multiplication by any power of w,
2) flipping of the two digits in any two of the couples,
3) bodily permutation of the three couples in any way.
A little calculation now reveals that under such symmetries every
hexacodeword is an image of one of the following five:
0101 ww ww ww ww 00 11 11 11 ww ww 000000
(36 images) (12 images) (9 images) (6 images) (1 image)

Since its images amount to more than half the code, the reader is advised
to memorize the first of these - the others are easily reconstructed from
this one and the known symmetries.
302 Chapter 11

3. Justification of a hexacodeword
It is important to be able to recognize at sight whether a word is or is not a
hexacodeword. Here's how to do this.

First check that it obeys the shape rule: up to permutations and flips
of its three couples the word should have one of the forms
Oa Oa bc, bc bc bc, 00 aa aa, aa bb cc, 000000

(where a,b,c are I,w,w in some order), and then check that it obeys the
sign rule: if we say that a couple of the form
ox or y wy has sign +
x 0 or y w y has sign -
00 or y y has sign 0

with x,y ;z!' 0, then the three couples have signs +++, +--, -+-, --+
or 000 (i.e. are either all 0 or have product +).
Thus Ow Iw wO obeys the shape rule, with a = W, b = I, c. = w, but the
signs of the three couples are ++-, so that it disobeys the sign rule, and is
not a hexacodeword. It would become a hexacodeword on flipping any
couple, for example Ow wi wO, with signs +--.
We refer to the process of checking the shape and sign rules as the
justification of a hexacodeword. Which of the words at the foot of this
page are hexacodewords ? (Answers are on the page behind.)

4. Completing a hexacodeword
In the application, one is often faced with the problem of determining a
hexacodeword from some partial information about its digits. In many
cases this reduces to one of two standard problems:
The 3-problem: determine a hexacodeword from any three of its digits.
The 5-problem: determine a hexacodeword from any five of its digits, one
of which may be mistaken.
In either case, the answer is unique, and so we earnestly recommend
that you use
The Best Method: guess the correct answer, and then justify it.
This is both easier and quicker than the methods that follow. But if
you're stuck, you can always solve such problems like this:
The 3-problem: If the given digits are from distinct couples (such as
a, c, e), use the w- or w-rule to find s, and then the I-rule for the
remaining digits (here b, d ,f).

?
OOllww 101010 wOwOwl wwwwOO Iwlwlw Ilwwww •
The Golay Codes and The Mathieu Groups 303

Example. w? I? O? • wS = w+1+0 = w, so s = w, W = wO 1w Ow, with

signs - - +. (Although the answer must be right if we've calculated
correctly, the reader is recommended always to check the answers by the
shape and sign rules')
If instead the three given digits include a couple (as do a, b, C), use the
I-rule to find s and a fourth digit (d), and the shape and sign rules (or the
w- and w-rules) for the other two (e,J).

Example. wI O? ?? • s = w+1 = w, so d = w, w = wI Ow wO, with

signs -+-. (Here the shape rule gives {e,J} = {O,w}, and the sign rule
then gives their order, wO')
The 5-problem. Any 5-problem reduces to several 3-problems, as follows.
The incorrect digit, if any, can affect at most one couple. But on deleting
the information for any couple, we still have at least 3 given digits, enough
to determine a hexacodeword. The required word must be one of these
three. Notice that in one case we actually have four digits - the
calculation can be speeded up slightly by noticing that these will be part of
a hexacodeword if and only if they are consistent, i.e. add to 0 in F 4 .
Example. wI Ow I? The three 3-problems are:
wI Ow?? , inconsistent, since w+ 1+O+w ;z!: 0 ,
wI ?? I? , so s = w, W = wI wI Iw (signs --+, but not an answer) .
?? Ow I?, so s = w, W = wO Ow 1w (signs -+-, which is) .
Another example. wI Ow I? This time the two couples are consistent, and
so we need only try wI Ow?? ,which gives wI Ow wO (with signs -+-).
The reader, who is naturally keen to acquire the ability to make
lightning calculations involving the Golay code, should now practice by
computing the hexacodewords defined by the 3- and 5-problems at the foot
of the page. Answers will again be found on the page behind. We
strongly advise that as many of these problems as possible be solved by the
Best Method, rather than by the slower methods outlined above.

5. The Golay code fC 24 and the MOG

The Golay code ~24 has already been mentioned several times (see in
particular §2.S.2 of Chap. 3 and §2 of Chap. lO). In this chapter we shall
often use blank and non-blank symbols instead of the digits 0 and 1.
Robert Curtis [Cur2] found that the easiest way to compute with words of
~ 24 is to arrange their digits in a 6 x 4 array, called by him the Miracle
Octad Generator, or MaG.
Words of ~24 in this notation are readily obtained from hexacodewords
by

?w O? w? 11 ?O ?? wI?w wI wO ?w 11
?
?w ?I w? wO wO O? ?I?w ?w O? wI ?? •
304 Chapter 11

either j) replacing each digit by an odd interpretation, in any

way such that the top row becomes odd;
or ii) replacing each digit by an even interpretation, in any
way such that the top row becomes even.

A row or column is called odd or even according as it contains an odd or

even number of non-zero digits. The odd and even interpretations of the
digits are shown in Fig. 11.1. Note that the odd interpretations have either
just one non-zero digit, or just one zero digit, in each column, while the
even interpretations differ from the odd ones only in their treatment of the
top row.

DaB 0 o a 8 0
o w w o w w

(0 ) (b)
Figure Il.l. Interpreting hexacode digits.
(a) Odd interpretations. (b) Even interpretations.

Justifying rtf -sets. The following procedure can therefore be used to verify
that a given set is a C{j -set (i.e. the set of places where a codeword of C{j24
is non-zero).
1) For each column, and for the top row, compute its count (the number of
non-zero digits it contains).
2) For each column, compute also its score, obtained by adding O,l,w,w
for any non-zero digit in the first, second, third, fourth row, respectively.
(Alternatively, a column has a given score digit if and only if it is one of
the four interpretations of that digit).
Then the set is a C{j -set if and only if the counts have all the same parity,
while the scores form a hexacodeword. We shall call this process
justifying the C(j -set.
In Fig. 11.2 we have given several interpretations of the standard
hexacodeword 01 01 ww. We recommend to the beginner our practice of
writing counts above, and scores below, the MOG diagram. Of course in
Fig. 11.2 the score words are all 01 01 ww.

EB3 E5E
020222 1 1 1 31 1 420222 3 1 3 11 3
***4 *** 3 * * 2 *
* * * * * ** * *
* ** ** *** * * **
*' * * ** * * *
0101ww 01 01 ww 01 01 ww 01 01 ww
Figure 11.2. Four Golay codewords from one hexacodeword.
The Golay Codes and The Mathieu Groups 305

~
tttJ
? 1 1 ? 1?

IF 000, UTI
~
YIELDS
li~I:+ ~I'
001111 (FAILS)
?01?1?

IF EVEN,
~
UTI 1:~I:rI2
** *
1770?0 YIELDS 1w wOw 0 ( WORKS)
Figure 1l.3. An example.

Which of the figures at the foot of the page correspond to Golay

codewords? You should know where to find the answers.

6. Completing octads from 5 of their points

An octad is (the C€ -set corresponding to) a Golay codeword of weight 8.
The octads of C€24 constitute a Steiner system S(5,8,24) (§3.I of Chap. 3,
§2.1 of Chap. 10); which is to say that any 5 points of the 24 belong to just
one octad. It is of considerable value to be able to locate the octad
containing 5 given points. This is easily reduced to 3-problems and 5-
problems when we note that to change the score of a given column we must
change at least one point in that column, and at least two points if we are
to preserve the parity of that column.
If the given points are located so as to make three columns odd and
three even, then we must correct the three columns having whichever is the
wrong parity, and so can afford to make no change to the others. We
therefore suppose first that the right parity is odd, second that it is even,
and then in either case we know three correct digits of the corresponding

EE8** ** tEE *
*
*
?
•

EEB tm
* * *
*
*
*

,
** * **

ww 00 ww 11 00 11 wI Iw wI ww ww 11
Iw wI wI wO wO wI wI Ow Ow Ow wI wO •
306 Chapter 11

hexacodeword, from which we can complete that word. Only one of the
two choices will yield an octad compatible with the given five points.
Figure 11.3 shows an example.
If instead the given five points are distributed so as to make five of the
columns have the same parity, then that parity must be the correct one,
and we can find the desired hexacodeword (whence the octad) by solving
the corresponding S-problem (see the example in Fig. 11.4).

102002 202022

ITTI aJU2
UTI
? OwOOw
YIELDS ~
wOwOlw
+
Figure 11.4. Another example.

(0) (b) (c) (d)

Figure 1l.5. The first column is (a) rather than (b); the fifth IS (c)
rather than (d).

It is perhaps worthwhile to spell out in more detail the way in which

the octad is reconstructed from the hexacodeword in this example. Which
of the two even interpretations should we choose for the two digits that
have to be altered? For the first column there is no problem. We already
have a star in that column, and so are forced to choose Fig. II.Sa rather
than Fig. II.Sb. But now we have determined all of the top row except its
fifth entry, and so must choose Fig. II.Sc rather than Fig. II.Sd for the
fifth column, so as to make the top row even.
Experience shows that when we have three odd and three even columns,
and are in doubt about the correct parity, it is usually best to consider the
odd case first (partly because odd interpretations are easier to see, and
partly because for octads the parity condition on the top row is
automatically satisfied in the odd case). However, the correct parity can
often be determined directly from the given information in some other way.
Thus oct ads must necessarily split 4 2 0 4 or 24 0 2 or 31 5 across the six
columns, and so from the distribution of the five given points across the
columns we can deduce the following information:
410 4 or 320 4 yield 4 20 4 , and so even parity (and trivia)),
31 20 3 or 150 yield 31 5 , and so odd parity,
yields 24 0 2 , and so even parity,
may yield 31 5 (odd), or 24 0 2 (even).
The Golay Codes and The Mathieu Groups 307

This time the foot of the page contains a number of sets of five points
for you to complete to octads.
Locating other l'€ -sets. Any set consisting of an odd number of the 24
points can be converted to arc-set by changing the status (member/non-
member) of 1 or 3 points. Our discussion of the octad-completion problem
generalizes almost trivially to this one - once again we must solve a 5-
problem or one of two 3-problems according as the given points are located
so as to make the columns be five of one parity and one of the other, or
three of each.
Sets consisting of an even number of points are either convertible to
rc -sets in a unique way by changing the status of 0 or 2 points, or in six
distinct ways by changing the status of 4 points. In the first case, the
columns all have the same parity, or will be four of one parity and two of
the other, and we solve a 5-problem (but with 6 given digits!) or two 3-
problems (with 4), as before. In the second case, we can change the status
of an arbitrary point, and will then be able to change three more to achieve
a rc
-set.

7. The maximal subgroups of M 24

As we have already discussed in §2.5 of Chap. 10, there are just nine
conjugacy classes of maximal subgroups of M 24. With the exception of
L2 (23), they can be characterized by the non-trivial partition of the 24
elements they fix, as shown in Table 11.1.
Several of these have been used as starting points for constructions of
M notably L 2 (23) by Carmichael, prL 3 (4) by Witt and Tits,
24 ,
26:(S3XL2(7» by Turyn, 24 :A g by Curtis, and of course 26 :3'S 6 for our
own construction.
In the rest of the chapter we shall devote a section to each maximal
subgroup in turn, relating it to the MOG as far as is possible. The sections
are in a rather haphazard order - first the projective group, since it

a;go
NO YES YES NO
040404 002222 I 333 I 1 202220

~I
,
3
'* '* '* *
* *
***"* *
001 1 1 1 Iw Iw Iw (;)0 Iw;:;;O

*
...
Efjj
I 1 1 31 1

... ...
3
~
202022
... * 2
•
* *
* ... * **
1 Oww 0 I wOwOwl
YES YES

Em ffij EEB
*
*
** '*
'*
*
* *
*** ffij
* '* '*
* ?
•
308 Chapte r 11

Table 11.1. Maximal subgroups of M 24.

Group structure Partition Name

M23 1, 23 monad group

M22:2 2,22 duad group
PrL 3 (4) 3,21 triad group
26 :3'S 6 46 sextet group
24 :A g 8, 16 octad group
M12:2 122 dodecad group
2 6 :S 3 X L 2(7) 83 trio group
L 2(23) 24 projective group
L 2(7) 38 octern group

provides the most usual labeling for the 24 points, then the four groups
most closely related to the MOG (the sextet, octad, triad and trio groups),
a few short sections on the monad, duad and octern groups, and then a
specially long section on MI2:2, which deserves an even longer one!

8. The projective subgroup L 2 (23)

One of the maximal subgroups of M24 is the group PSL 2 (23), or L 2 (23)
for short, which acts on the 24 points, when labeled 00 , 0, 1,2, ... ,21,22, by
linear fractional transformations z -'> (az + b) /(ez + d), with ad - be a
non-zero square modulo 23. The Golay code <fl = <fl24 can then be
defined as the code spanned by the images of the set Q consisting of 0 and
the quadratic residues modulo 23:
Q = {O, 1,2,3,4,6, 8,9,12, 13, 16, 18} .

(This is the definition of <fl24 as an extended quadratic residue code - see

§2.8 of Chap. 3.)
It is a remarkable fact that a suitable labeling for the MOG can be
constructed as follows. For the 12-element <fl -set (dodeead) corresponding
to Q we shall take that indicated in Fig. 11.6a. Then write the numbers

* * * 0 I2
* * * 3 4 -5
* * * 6 -7 8
* * * 9 -10 -II

(0) (b) (c)

Figure 11.6. Constructing the MOG labeling.
The Golay Codes and The Mathieu Groups 309

0 CD I II 2 22
19 3 20 4 10 18
15 6 14 16 17 8
5 9 21 13 7 12
Figure 11.7. The standard MOG labeling.

-(0)--- ---(b)--- -(c)---

Figure 11.8. Interpretations giving elements of :z6: 3.

0, 1,2,3,4,5,6,7,8,9,10,11 into these places in the natural order, and affix

signs where necessary so as to produce elements of Q (Fig. 11.6b). The
remaining numbers (00 and the quadratic non-residues modulo 23) are
found using the action of the permutation 'Y: z ~ -lIz shown in
Fig. 11.6c. The resulting labeling, displayed in Fig. 11.7, is called the
standard (modulo 23) MOG labeling. Any element of L 2 (23) now
becomes a permutation of M 24.

9. The sextet group 26 :3'S 6

A sextet (mnemonic: six tetrads) is a collection of six 4-element sets of
which the union of any two is an octad. It is known that M 24 is transitive
on sextets, and that any tetrad belongs to a uniquely determined sextet.
The standard sextet is the one whose six tetrads are the columns of the
MOG array.
The stabilizer of a sextet is a group 26:3'S 6, in which the subgroup 26:3
consists of the permutations that preserve every tetrad, while the quotient
group S6 describes the action on the tetrads. We shall describe this group
for the standard sextet in some detail.
The elementary abelian subgroup 26 is isomorphic to the additive group
of the hexacode, and its elements can be found from hexacodewords by
interpreting their digits as in Fig. 11.8a. This group has two further cosets
in the subgroup 26: 3, and elements of these cosets can be found by
interpreting digits in the alternative ways shown in Figs. 11.8b and 11.8c.
So a knowledge of the hexacodewords gives us the ability to recognize all

I:ll rffij1 Ern

222002

+ +
31 1 1 1 1

+*:*
"
+
+ +
"
"
"
1 11 31 1

" +
1
111113

~1
" " +" :
,
•
w 1 wO Ow Ow w 1 WO w 11 w 1 w wwwwOO
310 Chapter 11

192 elements of the group 26 : 3 at sight! Figure 11.9 shows the three
elements obtained like this from the word 01 01 ww.
But this is not all! By suitably interpreting the digits of suitable
hexacodewords, we can also obtain all the involutions in the subgroup 3'S 6
that preserves the top row of the MOG array. (Any permutation of the
top row extends to just three elements of this group, but this is a non-split
extension, and contains no subgroup S6.)

ftim ° ° 0
o
o
8im0
0

Figure 11.9. Three elements from one hexacodeword.

t 01° °1
PAIRING' 4 ...... - ~

1°
o I ww 10
Figure 11.10. A word, a pairing, and blobs.

The blob trick. Here's how the trick is done. We take any hexacodeword
W with two 0 digits, and any pairing of the digits in this word for which
the 0 digits form a pair. For each non-zero digit of W we place a blob in
the corresponding row and column of the MOG, but in the columns
corresponding to 0 digits we place blobs instead in the row corresponding
to the non-zero digit that appears at least twice in W. Figure 11.10 shows
our pairing and the blobs for the word 01 ww 10.
Now, using the terms defined in Fig. 11.11, and according to the cycle-
shape 142 or 1222 or 23 of our permutation, we
(I 42) replace the O-pair by an odd flipper, and other digits by odd fixers;

(I 222) replace the 0 digits by even fixers, and other pairs by even flippers;

(2 3) replace the O-pair and the other pairs by odd flippers.

These fixers or flippers must always take blobs to blobs. Figure 11.12
shows the results for our example.

~(a)- ---(b)--- (c) ---(d)---

Figure 11.11. The fixers and flippers: (a) odd fixers, (b) odd flippers,
(c) even fixer, and (d) even flippers.
The Golay Codes a nd The M a thie u Groups 3 11

R° wwI 10
~.. ..
°
.

ww I
~

Olww l O
10
Figure 11 .12. Three more permutations from a hexacodeword.

ttE
3 1 1 1 1 1 2 2 20 20
a a a be f
a b b bd c
c e f d c f
a oa
c c
c 3
~4
d fee e d c
wOOwwl
~
I ww Ow 0

Figure 11.13. A sextet and two octads.

f"i"G}l
~

Figure 11.14. A labeling, and some more interpretations.

.. ......-...... ~ :1 r:
1I I I X :J :J 1.;
° I ° I Ww °IOlww °IOlww

Figure 11.15. Three new interpretations of 01 01 ww.

Other sextets. As we remarked, any tetrad belongs to a unique sextet,

which can be found by completing octads from various sets of five points
including that tetrad. An example is shown in Fig. 11.13, with some of the
octads that justify it. (This sextet was in fact completed mentally at the
typewriter from its tetrad aaaa. It is instructive to check that the octads
formed by any two of its tetrads really are all octads.)
Some other sextets are particularly elegant, and the results for our
standard sextet can be transferred to these with little trouble. For instance
the six 2 x 2 blocks into which we regularly subdivide our MOG, and the
elements of its group 26 : 3'S6 can be obtained from the labeling and
interpretations shown in Fig. 1l.l4. Figure Il.lS shows some of these
elements.

10. The octad group 24: A8

It is known that M 24 is transitive On octads, and that the stabilizer of an
octad is a group 24 :A8. It is significant that the alternating group A8 is
isomorphic to the linear group GL 4 (2). In fact the group is best
understood by saying that the complement of the fixed octad acquires the
structure of an affine 4-space over the field F 2, or a vector space if we fix
also a particular point of it called the origin.
312 Chapter 11

83j
*
*
*
*
*
'* *
I~ ! ** **
'* * * *
*" * * *"
0
0
c
a+
o+c o+d c+d
d c+d

(01 (bl b b+c b+d c+

b d
o+b o+b o+b
o+b +c +d+c+d
(c )
Figure 11.16. (a) The standard octad and (b) its complementary square,
with origin marked O. (c) The entries in the square as vectors.

Figure 11.16 shows (a) our standard octad and (b) its complement, the
standard square. and the standard origin in that square. Figure 1l.16c
illustrates the vector space structure in the standard square.
The octad and its complementary square are linked by the fact that
there is a I-I correspondence between the partitions of the octad into two
tetrads, and of the square into four parallel 2-spaces, given by the
requirement that the resulting family of six tetrads should be a sextet. In
Fig. 11.17 we show all these 35 sextets. It is useful to note that the tetrads
(that is, 2-spaces) appearing in the square are precisely those that meet its
four columns in one parity, and its four rows in another (possibly the
same). This makes it easy to complete such a tetrad from any three of its
points.
(This remark is the basis of Curtis's original way of finding octads in
the MOG [Cur2], which uses the fact that every octad meeting the
standard one in four points is visible in one of the 35 sextets of Fig. 11.17.

Figure 11.17. The 35 standard sextets of the MOG.

The Golay Codes and The Mathieu Groups 313

t!E -trn
o x

1010ww
x
xx
x
0

x
xx
x

1010ww
x
x ~-c:PJ1
~~
wwwwOO wwwwww
(a) (b) (c) (d)
Figure 11.18. Transposing two octads.

:r
.....
H.
OO

:
(c)
ROO
. ..
:-t
~

(d)
Figure 11.19. Some permutations of 24 :A s.

EE8 EE8'
111113 022022

EEfl]
1001 ww
3

0110 ww
?
Figure 11.20. A problem, some calculations, and its solution.

Indeed, every octad except the three of the standard trio meets at least one
octad of that trio in four points, and so the figure can actually be used to
find all octads'>
The group 24 : As. The normal subgroup 24 consists of all elements that fix
the standard octad pointwise. In affine language, these are just the
translations of the complementary square.
The general element of 24: As acts as an element of the alternating
group As on the standard octad, and as an affine symmetry of the 4-space
on the complementary square. Every affine transformation of the square
extends to a unique element of M 24 by adjoining the appropriate even
permutation of the octad. Conversely, every even permutation of the octad
extends to 16 elements of M24 - we can force uniqueness by specifying a
destination for anyone point outside the octad. By requiring that the
origin be fixed, we restrict to a subgroup A8 complementing the group 24.
We illustrate some of these ideas by an easy calculation. The operation
of transposing the square is obviously an affine symmetry - let us find the
element of As that extends it to a permutation in M 24 . We note that the
two octads of Figs. 11.18a and c each have six points (x) in the square,
and only two (0) in the octad. We obtain the octads of Figs. 11.18b and d
by transposing the parts lying in the square, and then completing to octads.
Now the desired permutation is easily completed (Fig. 11.19a), using the
fact that it is an involution, and must be an even permutation. Some other
elements of M24 obtained in this way are also shown in Fig. 11.19.
The involutions of M 24 lie in two conjugacy classes, having cycle shapes
18 28 and 212. There is an easy way to compute an arbitrary permutation of
314 Chapter 11

·. X x x ··..
···... X
X :-t :1 :1
X X X ··.. :1 :1 :1

:1:I
:I :I

.. -
Figure 11.21. Elements in 24 :A g and other groups.

type 1828 from its fixed points (which must form an octad) and one of its
transpositions (which may be freely chosen). We just use the fact that if
an octad contains four of the fixed points and the two points of a
transposition, then its remaining two points must also be a transposition of
the permutation. Figure 11.20 illustrates the calculation.
The process of extending permutations by completing octads is one that
soon palls, however, and it is handy to know that it can often be avoided.
In particular, we can find many permutations in our group 24:A8 by using
other groups for which the calculation is trivial. If the desired element
preserves the sextet made from the six columns, or that made from the six
2x2 subsquares, we can use the sextet group. If it fixes the standard trio,
we can use the group 26: (S3xL2(7». Finally, if it fixes the standard triad,
or equivalently, the standard way of regarding the square as a 2-space over
F4 (rather than merely a 4-space over F 2) we can use the triad group
prL 3 (4), explained in §11. A miscellany of elements obtainable in these
ways is displayed in Fig. 11.21.

11. The triad group and the projective plane of order 4

When 3 points of the 24 are distinguished, the remaining 21 points acquire
the structure of a projective plane over F4. The points of this plane can be
coordinatized with triples (x,y, z) of elements of F4 not all zero, with the
usual understanding that (x,y, z) = (kx, ky, kz). The five points with
z = 0 are specified by the ratio y Ix and form the line at infinity, while the
other 16 points can be given normalized coordinates (x,y, 1), which we
abbreviate to (x,y), and they form an affine plane over F4.
In the MOG, we take the three distinguished points to be the lower
three points in the first column, the line at infinity to be the remainder of
the standard octad, and the affine plane to be the standard square, the
coordinatization being as in Fig. 11.22.
The Golay Codes and The Mathieu Groups 315

y/x

000
-
o
X
I·ww
0
I;'-~":"~f--+---t~ ~Y
...

Figure 11.22. The MOG as a projective plane.

y=x
~
[fu
COORDINATES 0,1
r'I; T'I
X=O , y=O
Figure 11.23. Four octads seen from a projective viewpoint.

We follow Witt [Wit2], [Wit3] in calling the three special points I, II,
III, and we shall refer to them as the Romans. The oct ads of the Steiner
system have the following four forms:
(3+5) All three Romans together with a line of the plane.
(2+6) Two Romans with a six-point oval (hyperconic) in the plane.
(]+7) One Roman, together with a subplane defined over F 2 •
(0+8) No Romans, and the sum of two lines of the plane.

(The sum is modulo 2, so that the intersection of the two lines does not
appear.) A hyperconic consists of the 5 points of a conic together with the
unique point on no chord of that conic. Figure 11.23 shows examples of all
four types of octad. We shall call this the Witt-Tits construction, because
Witt constructed M24 from PSL 3 (4), while Tits [Tit2] explored the implied
geometry in some detail.

SEMI LINEAR
(b)

Figure 11.24. Appropriate ways to permute the Romans.

Any automorphism of the plane now extends uniquely to a permutation

of M 24 by adjoining the appropriate permutation of the Romans, as
follows. If it results from a linear (over F 4) transformation of the
coordinates, the appropriate permutation depends on the determinant in the
way shown in Fig. 11.24a. If instead it results from a semilinear map, we
get one of the permutations shown in Fig. 11.24b, and the first of those
permutations if this map is in PfL 3 (4). In summary
316 Chapter 11

PSL 3 (4) fixes all three Romans (and is M 21 ),

PGL 3 (4) permutes them cyclically,
P~L3(4) fixes I, and permutes {II, III}.

You can now write down a permutation of M24 for any element of
prL 3 (4). Figure 11.25 shows some examples.

. . ..--
....
(0)
E (b) (c)
Figure 11.25. Four elements of prL 3 (4):
(d)
IIIIX

(a) x' = y, y' = x, z' = z; (b) x' = wx, y' = wy, z' = z;
(c) x' = z, y' = y, z' = x; (d) x' = x,
y' = y, z' = z.

M 24 is transitive on trios (mnemonic: three octads) consisting of three

disjoint octads. The subgroup fixing a trio is a group 26: (S3XL2(7)), it
being significant that L 2 (7) is isomorphic to L 3 (2). This subgroup is best
understood by saying that the three octads acquire the structure of three
parallel affine 3-spaces over F 2 .

8
0 c
3 2 o o+c
5 I b b+c
o+b
6 4 o+b +c
(0 ) (b)
Figure 11.26. Two labelings for an octad.

(0)
BB88B8B8
0 A B C D E F G

(b)
B88BBEH38
0 0 b c d e f
Figure 11.27. The two codes of Turyn's construction. (a) These sets and
g

their complements form the line-code. (b) These sets and their
complements form the point-code.
The Golay Codes and The Mathieu Groups 317

We take the standard trio to be that formed by the left, middle, and
right couples of columns of the MOG, and label each of these octads as
shown in Fig. 11.26a. The affine structure on any of these octads is that
corresponding to Fig. 11.26b.
According to Turyn's construction [Mac6, Chap. 18, Th. 12], the
Golay code C€24 consists of the words
(X+t, Y+t, Z+t)

and their complements, where X + Y + Z ... 0, and X, Y,Z belong to the

line-code of Fig. 11.27a, while t belongs to the point-code of Fig. 11.27b.
The tetrads of the line-code are the 2-spaces of the affine 3-space and yield
octads when placed in any two of the three standard octads. The tetrads of
the point code are precisely those tetrads that yield dodecads of C€24 when
repeated in all three octads of the standard trio. (One of these dodecads is
the one labeled with elements of Q in the standard labeling of the MOG
- see Fig. 11.7.)
The group 26: (S3XL2(7» contains the following subgroups:
1) The elementary abelian normal subgroup 26, whose elements are
obtained by combining any three translations adding to O.
2) The subgroup S3 consisting of the bodily permutations of the three
octads of the standard trio.
3) The group L 2 (7), whose elements are precisely those permutations of
M 24 that act in the same way on all three octads.
Figure 11.28 contains some examples. The first picture is made of three
translations adding to 0 (namely, those by vectors b,c, and b+c), the
second is the cyclic permutation of the three standard octads, and the
remaining three pictures are the elements
a:z ~ z+l, (3:z ~ 4z, -y:z ~ -lIz

of L 2 (7) under the labeling of Fig. 11.26a.

We remark that the illustrated tetrads of either of the line- and point-
codes are images of each other under the element a of Fig. 11.28. If we
call two tetrads incident when they meet in two points, Fig. 11.29a shows
how the two codes get their names.

-
..... ..... .....
[HmJEj X X X - .....

a ~ y
Figure 11.28. Five elements of the trio group.
318 Chapter 11

Figure 11.29. Incidences between points and lines .

. . . . ---
... . ---
I I I 1X
a' (3' y'
Figure 11.30. Generators for another L 2 (7).

If we consider the group 26 :L 2 (7) that preserves all three octads, we

find that it acts on anyone of the octads as a group 23 :L 2 (7), in which the
normal subgroup 23 consists of our affine translations, while a
complementary L 2 (7) is the one determined by the labeling of Fig. 11.26a.
However, another complement to the normal 23 is the L 3 (2) that fixes the
point 00 and acts as automorphisms of the projective plane whose seven
points are the other numbers 0, 1,2,3,4,5,6 and whose seven lines are
obtained from the tetrads of the line-code by deleting 00 (see Fig. 11.29b).
So to obtain the isomorphism from L 2 (7) onto L 3 (2) we simply multiply
any given permutation of L 2 (7) by the affine translation that restores 00 to
its rightful place.
Figure 11.30 shows three generators for the L 2 (7) (= L 3 (2» that
preserves all three octads, and also fixes a point from each of the first two.
These were obtained from the permutations a, {J, 'Y of Fig. 11.28 by
applying this process to the first two octads.

13. The octern group

This, the smallest maximal subgroup of M 24 , is the stabilizer of a
particular partition of the 24 points into eight cyclically ordered triads
(mnemonic: eight terns).
There are two distinct ways to select a 4-dimensional subspace of C(i24
consisting of the empty and full sets together with 7 pairs of
complementary dodecads. In each case, we obtain a partition of the 24
points into eight triads by considering the intersections of triples of
dodecads. In the first case, one of the dodecads may be taken to be the set
Q of Fig. 11.6a, and the others its images under powers of the element a
we defined in connection with the trio group (§12). The resulting set of
triads consists of the triads of points similarly numbered in Fig. 11.26a,
and the stabilizer is the subgroup S3XL2(7) of the trio group.
The Golay Codes a nd The Mathie u Groups 319

24 21 41 42 12 14

54 51 31 32 62 64

.,.
44 61 II 52 22 34

°1 "1 °2 "2 0 .
( 0)

1I
..........
(e) (d)

Figure 11.31. (a) Numbering for the octern group. (b) - (d) The elements
a· , fJ· , 'Y • .

In the second case the triads are (n],n2,n4} for n = 00,0,1, .. . ,6

(Fig. 11.31a). The stabilizing group (the octern group) is the L 2(7)
generated by the elements a *, {3*, 'Y * shown in Fig. 11.31 b-1.31d, with
effects
a* n/ ---? (n + 1)/,
{3* n/ ---? (2n)2/,

'Y * n/ ---?
[- ~ Ln 2 or/2'

where nand t are to be read modulo 7, and the second alternative in the
last line is used only for n = 00 and n = O. One of the dodecads consists
of all the points Or. 1/, 2/ or 4/ in Fig. 11.31 a. This group can be defined
as the centralizer in M 24 of the octern element (not in M 24):

14. The Mathieu group M 23

By fixing anyone point of the 24 (say the point 00), we obtain the
maximal subgroup M 23, which could alternatively be defined as the
stabilizer of the Steiner system S (4,7,23) whose 253 heptads are obtained
by deleting 00 from all the oct ads of S (5, 8, 24) that contain it. Although
M 23 is a smaller group than M 24, it cannot be too strongly stressed that all
properties of M 23 are best understood by embedding it in M 24, and we
shall therefore say no more about it.

15. The group M 22: 2

By fixing two points of the 24 (say 00 and 0), we obtain the Mathieu group
M 22, which of course is not maximal, since we can add to it the
320 Chapter 11

permutations which interchange those points, to get M 22: 2. This is the

stabilizer of the Steiner system S (3,6,22) obtained by deleting 00 and 0
from all octads of S (5,8,24) that contain them both. It is of course
equally true to say that these groups are also best studied by embedding
them into M 24. However, there are some remarks about the hexads which
may be found helpful.
Anyone of the 77 hexads is disjoint from 16 others, and meets the
remaining 60 in two points each. (These relationships turn the family of
77 hexads into a rank 3 graph.) One of the hexads for the M 22 fixing 00
and 0 is the rest of the standard octad, and the 16 hexads disjoint from this
are any row plus any column in the standard square. The other 60 hexads
come in 15 sets of 4, and can be read from Fig. 11.17. There are 15
sextets of that figure for which one tetrad includes {00,0J, and we obtain
the 60 hexads by taking the remaining two points of that tetrad with one of
the four tetrads corresponding to it in the standard square.

16. The group M 12, the tetracode and the MINIMOG

The Mathieu group M 12 is both a subgroup of M 24 and an analog of M 24.
We treat it in the latter role first.
M 12 may be defined as the stabilizer of a Steiner system S (5,6, 12), but
it is most easily understood in connection with the ternary Golay code ct'12
(§2.8.5 of Chap. 3, §1.5 of Chap. 10), which we shall define here in terms
of the tetracode ct'4 (§2.5.J of Chap. 3) and the MINIMOG 4x3 array.
The tetracode ct'4 consists of just 9 words over F3:
0000, 0 +++, 0 ---,
+ 0+-, + +-0, + -0+,
- 0-+, - +0-, - -+0

(We usually prefer to use the symbols 0, +, - rather than 0,1,2') In the
typical word a bed, the first two digits (a,b) specify a linear function
rj)(x) = ax + b, while the last three give the values rj)(0), rj)(I), rj)(2):
a b=rj)(o) e=rj)(+) d=rj)(-).

In other words the leading digit defines the slope s of the word, and the
other three digits cyclically increase by s. This makes it very easy to solve
the two types of problem:

I 221 3 300 2022 2202

tE:J [] 8SJ [tJ

-
+
+ + -
+

0
+
+
-
-
? ? ??
+
-
+-
-- +
0(0) 0 0
- ~
- 0(-)+
0 +

,
I 221 2220 I 122 0222 •
[]
+ -
- 0 - +
IT]
+ -
-
o - -(-I
+
0 ad l[J
- - + 0 (0) 00 0
The Golay Codes and The Mathieu Groups 321

The 2-problem: complete a tetra codeword from any 2 of its digits.

The 4-problem: correct a tetracodeword given all 4 of its digits, one of
which may be mistaken.
The MINIMOG is a 4x3 array whose rows are labeled 0,+,-, and
whose columns correspond to the four digits of the tetracode. We shall use
the word column (abbreviated "col") also for the length 12 word with +
digits in the entries of one column (and 0 digits elsewhere), and the word
tetrad ("tet") similarly for the weight 4 word with + digits in the places
corresponding to a tetra codeword and 0 digits elsewhere (see Fig. 11.32).

o
[[]
6 309
+ 5 2 7 10
- 4 1 8 11
+ 0 + -
Figure 11.32. The MINIMOG array with shuffle numbering, a "col" and
a "tet."

Then the ternary Golay code CC 12 is defined by the observation that

modulo CC l2• any column is congruent to the negative of any tetrad.

(Compare CC 24 , modulo which any column is congruent to (the negative

of) any hexad.} In particular, this tells us that the combinations
~-~, ~+M, M-M, ~+~-M

all yield CC l2 -words, and in fact they all yield signed hexads (CC l2 -words
of weight 6) - see the examples in Fig. 11.33. We have also entered in
the figure the column-distribution of the 6 points, and the odd-men-out.
(If a column contains just one non-zero digit, or alternatively just one zero
digit, then the odd-man-out is the name of the corresponding row.
Otherwise there is no odd-man-out for that column, and we write ? under
it.)
If we ignore signs, then from these signed hexads we get the 132
hexads of the Steiner system S (5,6, 12). These turn out to be all possible
hexads for which
1) the odd-men-out form part of a tetra codeword, and
2) the column-distribution is not 32 10.
The problem of completing such a hexad from any 5 of its points is easily
solved in terms of 2-problems or 4-problems, and then the signs can be
added, if desired, by using the expression in terms of columns and tetrads.

[D
0~o:
+
+
? ? ??
-
-
[t-l
~
+ ? + -
[j
- ? - +
EEJ
I;: I
- + 0
++-
+ 0 + -
Figure 11.33. Four signed hexads and their odd-men-out.
322 Chapter 11

[J
**
+ + 0
(0 )
?
to
* *
**0 -
- +
(b)
D ++0

(C)

u:: I
- +0
(d)
?
U::J
- + 0 -
(e)
ffiI]o+
- ++ -
- + 0 -
(f )
Figure 11.34. From pentads to signed hexads.

For example, the 5 points of Fig. 11.34a have column-distribution 2210

and so must complete to a hexad with distribution 230 or 2212. In either
case the middle two digits of the odd-man-out sequence are correct, and so
we solve the 2-problem ? +o? *- +0-. This shows that it is the first
column that must be amended, and so the answer is Fig. 11.34b. The
answer to the problem in Fig. 11.34d is given in Fig. 11.34e.
Now to put in the signs! For the second of these two problems, this
presents no difficulty, since the column-distribution 2212 tells us that the
hexad has the form col + col - tet, and therefore the signs are as shown in
Fig. 11.34f (we've inserted those 0 digits to show you where the subtracted
tetrad was). But for the original problem, the column-distribution implies
a form tet - tet, and we need two tetrads, neither of which is the odd-
men-out tetrad - +0-, since in this case we can see that those odd men
really are out! The answer is that we take the two tetrads 0 --- and
+ 0+- that agree with that tetrad in the digit corresponding to the empty
column (where we might have written? rather than -), and so derive the
signs as shown in Fig. 11.34c ([ 0 ---] - [+ 0+-] ). This is one of
many cases when the process is easier performed than described.
At the foot of the page we have set up a number of 5-element sets. On
the page behind we give their completions to hexads of CC 12 , together with
a proper choice of signs. Note that there are two correct ways to sign any
hexad of S (5,6,12) - the one you first thought of, and its negative.
The relation between the MOG and the MINIMOG. The MINIMOG can
be inserted into the MOG as shown in Fig. 1l.35. With this inclusion, the
hexads of S (5,6, 12) are just those 6-element subsets of the right region
(marked by letters r in the figure) which can be completed by two points

[[J U
**
[J
* "" to
""
*
?
•
II:: "I brJ
"
"" 11*: :1 [D
"*"
The Golay Codes and The Mathieu Groups 323

r r r
r r r
r r r CONTAINS :CLOSES TO
r r r

Figure 11.35. How the MOG contains the MINIMOG.

~BECOMES
~
m*
?OOwl?
"*
"***
,EXTENDS TO m*
* *
***
*
wOOwlw

Figure 11.36. How a hexad becomes an octad.

of the left region to form an octad of S (S, 8, 24). Figure 11.36 shows an
example.

17. Playing cards and other games

The numbering of Fig. 11.32 has many remarkable properties. In the
course of an investigation into the mathematics of shuffling playing cards,
Diaconis et al. [Dia2] discovered a remarkable new way to generate M 12.
More generally, for any divisor n of 12, let us take n cards, and number
them 0,1,2, ... ,n -1. Let us consider the group Mn of permutations of
these cards generated by the operations
rn t-?n-l-t,
Sn t - ? min { 2t, 2n - 1 - 2t } .

In playing-card language, rn is the operation of reversing the order of the

cards, while Sn is a special permutation called the Mongean shuffle. Then
for n = 1,2,3,4,6, 12 we find these groups are respectively
S], S2, S3, A4, PGL/S), M 12 ,

where the PGL 2 (S) acts on the cards in the way it permutes 0,00,1,2,3,4,
while M 12 acts upon them as on the numbers in Fig. 1l.32.
Many remarkable facts about M 12 appear when we regard it as acting
on a set ordered in this particular way. We consider the M12 as embedded
in M 24 acting on two such populations, as in Fig. 11.37. This has a simple
relation to the projective numbering of Fig. 11.7, given by
n+ whichever of n, -n is in Q ,
n whichever of 8/n, -8/n is in ll\Q.

The duality of this M 12 is the permutation of M 24 that interchanges u+

with u-. It is a particularly interesting outer automorphism of M 12 •
Whenever ab = 12, there is a standard partition of the 12 cards into a
sets of b, namely one of the partitions indicated by the bars in the
324 Chapter 11

0+ 0- 1+ r 2+ 8-
2- 3+ 5- 4+ 10- 5+
1- 6+ 6- 7+ 9- 8+
3- 9+ 4- lat 11- 11+
Figure 11.37. The shuffle numbering for M 24 .

10 I 2 3 4 5 6 789 10 III FOR a = I, b 12

10 I 234516789 10 III FOR 0= 2, b 6

10 I 23145 6718 9 10 III FOR 0= 3, b = 4
10 I 21345167 819 10 III FOR a = 4, b = 3
101123145167189110 III FOR a = 6, b = 2
101112131415161718191101111 FOR a = 12, b =

Figure 11.38. The standard partitions.

appropriate line of Fig. 11.38. Then the stabilizer of such a partition turns
out to be a group Ma x M b , where Ma acts on the a rows and Mb on the b
columns of the appropriate array of Fig. 11.39. Moreover, the standard
duality takes the stabilizer of the standard partition into a sets of b into
that of the standard partition into b sets of a, with the map Sa on the rows
dualizing to sa-Ion the columns! The map ra on the rows dualizes to a
map which reverses the order of the columns, although in the case a = 6
this is not ra alone, but its product with the r2 that interchanges the two
rows.
o
2
01 3
01 2 32 4
0123
012345 345 45 5
o 1 234567891011 7654
678 76 6
67891011
891011
(0) 91011 89 7
(b)
(c) 1110 8
(d)
9
(e)
10
11

(f)
Figure 11.39. Six ways to lay down the cards.

The group generated by the turnaround permutations

T}. T 2, T 3 , T 4, T 6 , Ts, T 12 , T24
shown in Fig. 11.40 is 2 x M 12, the central group of order 2 being
generated by T 24. This gives a strikingly simple definition of M 12.
The Golay Codes and The Mathieu Groups 325

Identifying points with the same number in Fig. 11.40 produces the same
M 12 that we have just defined.

o 2 3 4 5 6 7 8 9 10 11 11 10 9 8 7 6 5 4 3 2 1 0
o 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 T1

o 0 a 0 ~ 0--0 ~ ~ 0--0 ~}T2

0-----<) 0--------0 0------0 0-----0 0------0 0----0 .:...-----0 0----0 T3

~----<o 0 0--0 0 C 0--0 o}T4

0--0
} T6
0

0--0

}Te
0

} T'2
0

j'M
0

Figure 11.40. The turnaround permutations Th T 2 • . . . • T24 that

generate 2 x M 12.

Some more surprises happen when we plot the hexads of S (5,6,12) in

terms of the sums of their elements, in this shuffle numbering. The
resulting histogram (Fig. 11.41) suggests that there should be something
very special about the 11 hexads with sum 21, and indeed there is! We
shall call these the light hexads, and their complements, which are the 11
hexads with sum 45, the heavy ones.

NU~ER 1, ,-,
,
HEXADS I- l-
I- 1--1- l-
I- l-
I- l-
I- l-
f- l-
I-
I- -
I- -
I- -
20 22 24 26 28 30 32 34 36 38 40 42 44 46
21 23 25 27 29 31 33 35 37 39 41 43 45
SUM
Figure 11.41. Pyramid between obelisks.
326 Chapter 11

Table 11.2. Duality between hexads and duads.

123456 is dual to 0
o 1 237 8 is dual to 2
o 1 245 9 is dual to 2 3
o I 346 7 is dual to 3 4
0123510 is dual to 4 5
o 1 246 8 is dual to 5 6
023457 is dual to 6 7
o12 36 9 is dual to 7 8
o 1 345 8 is dual to 8 9
o1 2 5 6 7 is dual to 9 10
0123411 is dual to 10 II

It turns out that the 11 light hexads are precisely all the hexads of
non-negative integers whose sum is 21, and that when we join two hexads
if they have just 3 points in common, then these 11 hexads naturally form
a chain, in which each joins just those adjacent to it. Moreover, under our
duality, these hexads correspond to the 11 duads {0,1}, {1,2}, ... , {10,11}
as in Table 11.2. [The hexad {a, b, c, d, e,f} is dual to the duad {x,y} if
and only if {a+,b+,c+,d+,e+,f+,x-,y-} is an octad of S(5,8,24).] A last
remark is that the number of heavy hexads that contain (respectively)
0,1,2,3,4,5,6,7,8,9,10,11 is 1,1,2,3,4,5,6,7,8,9,10,10.
The Steiner system S (5, 6, 12) is obtained from the 11 light hexads by
complementation in {O, 1,2,3,4,5,6,7,8,9,10, II} and by addition of hexads
whenever the sum is again a hexad. The Golay code re
24 is spanned by
the corresponding 11 octads together with the universal set.
It will not be easy to explain all the above observations. They are
certainly connected with hyperbolic geometry and with the "hole" structure
of the Leech lattice. For example, using the terminology of Chap. 24, if
we take coordinates for the Leech lattice centered at a hole of type A [4,
then the vertices of that hole are the 48 f -points (±4,023), and the next
nearest points are the 4096 g -points (± 124), the signs being determined by
a re-set. In these coordinates the center of a certain D 24 -hole is

C = 23
1 ( 11 12., -11,-9,-7,-5,-3,-1,1,3,5,7,9,11 )

when the coordinate positions are labeled in the order

0-,1-,2-, ... ,11-; 0+,1+,2+, ... ,11+ .
The Golay Codes and The Mathieu Groups 327

The vertices of this hole are the f -points that have inner product 44/23
with e, and the g -points whose coordinates -1 are against the 11 octads
mentioned above.
The lexicography of the hexacode and the Golay code reM' If we define a
list of numbers in f e scale of 4 by the requirement that each should be the
smallest such numbe, differing in at least 4 digits from all previous ones,
then it turns out that the first 64 of them:
000000, 001111, 002222, 003333, 010123, ... , 333300

are just the hexacodewords (with 0,1,2,3 for O,l,w,W). This observation
turns out to be equivalent to the "ax 2+bx+e" definition of the hexacode,
and R. A. Wilson [Wil6al has generalized it in various ways. By using the
scale of 2, and demanding that the numbers differ from previous ones in at
least 8 digits, M. 1. T. Guy found the Golay code CC24 , with the order
obtained by reading down the MOG columns, from left to right. Such
"lexicographic codes" are discussed in greater detail in [Con411.

18. Further constructions for M 12

The arithmetic progression loop. There is a loop multiplication, ., on the

12-e1ement set L = {00,0,1,2,3,4,5,6,7,8,9,X = 10} defined by the
following two assertions.
1) 00 is an identity, and is the square of every element (e.g.
00.5 = 5*00 = 5, 5.5 = 00).
2) If a, b, e are three distinct numbers in arithmetic progression, and b-a
is a quadratic residue modulo 11, then a * b = b * a = e. The quadratic
residues are 1,-2,3,4,5, so for example 7*8 = 9, 6.8 = 4.
This loop yields two constructions for M 12.
m M 12 is
generated by the permutations 7ra (a E L), where
7ra:X - 7 x*a. E.g. 7rQ = (000)(129748365X), and the other 7ra are
obtained by translating the numbers in this by a, modulo 11.
(II) We consider the set of all permutations 7r of L for which there exists a
7r' such that

Then M 12 is generated by these 7r and 7r'. In fact the permutations 7r

constitute the standard L 2 (I1) of maps of the form
x - 7 (ax+b)/Ccx+d) with ad-be a quadratic residue, and the 7r' form
another subgroup L 2(I1) of M 12 . This yields the generators for M12 used
in Chap. 10, thus:
a = a' = (00)(0123456789X) (t - 7 t+1) ,
{3 = {3' = (00)(0)(13954)(267X8) (t - 7 3t) ,
'Y (00 0)(1 X)(2 5)(3 7)(4 8)(6 9) (t - 7 -1ft) ,
'Y' (00)(0)(1)(2X)(34)(59)(67)(8) .
328 Chapter 11

The permutation "(' was called {) in Chap. 10.

Here is a dictionary between this modulo 11 numbering and the shuffle
numbering:
shuffle: o 2 3 4 5 6 7 8 9 10 11
modulo 11: 0 9 3 4 5 00 8 6 2 X 7
mnemonic: 0 +1 -2 +3 +4 +5 00 -3 +6 -9 -12 -15
The Rubicon or Rubik-icosahedron construction of Mil. If we call
permutations a and fJ of twelve letters congruent when afJ- I E M 12, there
are 12!/1 MI21 = 7! = 5040 congruence classes, as shown in Table 11.3.

Table 11.3. Classes of permutations of S 12 modulo M 12.

Even permutations:

(a) 1 class represented by the identity permutation

(b) 440 classes represented by the 3-cycles
(c) 495 classes each with 3 representatives of shape 22
(d) 1584 classes each with 12 representative of shape 5
2520 total
Odd permutations:

(e) 66 classes represented by the 2-cycles

(f) 990 classes each with 3 representatives of shape 4
(g) 1320 classes each with 12 representatives of shape 2 13 1
(h) 144 classes each with 110 representatives of shape 6
2520 total
Examples of the classes containing more than one shortest
representative are (in the shuffle numbering):
Class c:
(09)(36) == (12)(45) == (78)(10 II).

These are the squares of:

Class f:
(0396) == (1425) == (7 11 8 10) .

Class g:
(12)(963) - (02)(1074) - (01) (1 1 8 5)

- (45)(069) - (35) (1 7 10) - (34)(281 I)

- (78)(930) - (68)(104 I) - (67)(11 52)

- (1011)(036) - (911)(147) - (9 10)(2 5 8).

The Golay Codes and The Mathieu Groups 329

(To understand these, arrange the numbers in the standard 4x3 array as in
Fig. 11.39d.)
For classes d and h we change to the modulo II numbering
{oo,O,I, .. . ,xl.

Class d:
(19435) - (2X687) - (005789)

- (0753X) - (004X25) - (0096X3)

- (003271) - (OX 49 8) - (08152)

- (06917) - (001864) - (02346) .

Class h: The map

(00)(0)(IX)(29)(38)(47)(56) (t ---?- -t)

belongs to a class of type h. This class is invariant under the group

L = L 2 (11) of maps t - (at +b)/(et +d) where ad-be is a quadratic
residue modulo II. The 110 shortest representatives of this class are the
conjugates of (00 0 2 3 46) by elements of L . (Since the permutation
(00 0 2 3 46) (1 97 X 8 5) is in L, each of these representatives is one of the
cycles of an element of order 6 in L.)
We note that the permutations in our example for class d are just the
twelve clockwise 5-cycles Tj around the vertices i of the icosahedron shown
in Fig. 11.42.
It turns out that M 12 is actually generated by the corresponding
quotients TjT;I . In other words M\2 is the set of all permutations of the
form
twist untwist twist untwist ... twist untwist

x
Figure 11.42. An icosahedron.
330 Chapter 11

of the vertices of a kind of "Rubik" icosahedron in which the "twists" Tj

and "untwists" Til are physically possible.

Remarks. 0) The twist about one vertex followed by the untwist about the
opposite one is a generating rotation of the icosahedron. So the rotation
group of the icosahedron is contained in this M 12. Oi) Mathematically it is
simpler to regard MI2 as generated by the twelve permutations Tia, where
a = (ooO)(IX)(29)(38)(47)(56)

is the antipodal inversion of the icosahedron. It turns out that

Toa = (ooO)(129748365X)

(Memo: alternately double and negate!) is the permutation 11"0 mentioned

earlier in this section.

The Mathieu groups have now been known for a century and a quarter,
but are still capable of surprising us. Some references to them are [Ass2],
[Betl], [Carl], [Cur6], [Cur7], [GibI1, [HugI1, Don5], [KraI1, [MasI1,
[MatI], [Mat2], [PeeI1, [PleI5], [PleI5a], [Soil], [Sta12], [Ton3],
[WarI1, [Whil1.
 12
A Characterization of the Leech Lattice
J. H. Conway

We give a short proof that Leech's remarkable lattice is characterized by

some of its simplest properties.
The main result of this chapter is Theorem 7, which characterizes the
Leech lattice in four different ways. Although we must quote two
theorems to open and close the proof, the reader can take these on trust if
he wishes, and he will find that the proof is otherwise completely self-
contained. The argument immediately gives the order of the
automorphism group (Coo) of the lattice, and can be used to give other
information about the group and the lattice.
We proceed at once to the proof. We shall show that Leech's is the
only lattice in fewer than 32 dimensions that has one point per unit volume
and in which the square of every non-zero distance is an even integer
greater than 2. A lattice with one point per unit volume is called
unimodular, and then also even if every squared distance or norm is an
even integer (§2.4 of Chap. 2). In an even unimodular lattice of dimension
n we use N m for the number of vectors of norm m. Our main theorem
then asserts that Leech's is the only even unimodular lattice with n < 32
and N2 = O.
Theorem 1. If A is an even unimodular lattice with n < 32 and N 2 = 0,
then we have n = 24, N4 = 196560, N6 = 16773120, Ns = 398034000.
Proof Heeke's theorem (Theorem 17 and Cor. 18 of Chap. 7) says that if
A is an even unimodular lattice then n is a multiple of 8, and the theta
series e A (z) is a modular form of weight n/2 for the full modular group.
For n = 8 or n = 16 the space of such forms has dimension 1 and so the
condition No = 1 determines N 2m for all m, and in fact we find N 2 > 0 in
each case. For n = 24 the corresponding space has dimension 2 and now
the values No = 1, N2 = 0 determine N 2m . In fact we find that the
coefficients of the extremal theta series are
332 Chapter 12

where (J •• (m) is the sum of the ll-th powers of the divisors of m and T is
Ramanujan's function. The given values are readily computed from this .
From now on we shall suppose that A is a lattice satisfying the
hypotheses of Theorem 1. We call two vectors of A equivalent if their
difference is twice a lattice-vector - i.e. we consider A/2A. If
{b •. .. . . b 24 } is a basis for A then {2b •.. . . . 2b 24 } is a basis for 2A, and
so there are just 224 equivalence classes. We call a vector x short if it has
length at most .J8, and enquire which equivalence classes contain short
vectors. Since x and -x are always equivalent, the non-zero short vectors
in an equivalence class will occur in pairs (compare Theorem 2.8 of
Chap. 10).
Theorem 2. Each equivalence class contains a short vector. The
equivalence classes that contain more than a single pair of short vectors
are precisely those that contain vectors of length .J8. and these classes all
contain exactly 24 mutually orthogonal pairs of vectors of that length.
Proof Let x = OX and y = OY be equivalent short vectors, with
y ¢ ±x. Replacing y by -y if needs be, we can suppose that the angle
XOY ~ -rr/2. Then since OX2 ~ 8 and Oy2 ~ 8 we must have XY2 ~ 16
(Fig. 12. I) . But XY is twice a lattice vector, and so XY ~ 2J4, whence
XY2 ~ 16. It now follows that every inequality in the argument must in
fact have been an equality, so that two distinct equivalent short vectors not
forming a pair must be orthogonal and both of length .J8, and so there can
be at most 24 pairs of such vectors. The number of equivalence classes
that contain short vectors is therefore at least
No N4 N6 Ng
-+-+-+- .
I 2 2 48

Since the displayed number is exactly 224 each equivalence class is

accounted for, the other inequalities of our argument must also have been
exact, and the conclusions of the theorem follow.

-y
--~-----+~~----~ y

-x

Figure 12.1 Why XY2 ~ 16.

A Characterization of the Leech Lattice 333

Now let c = 1/../8. Then we can suppose that A contains all vectors of
the shapes c(±8,023 ) and c«±4)2,022), these being 24 mutually
orthogonal pairs of vectors of length ../8 and their halved differences. The
symbol «±4)2,022) denotes the typical vector with two coordinates ±4
and 22 coordinates zero. These vectors generate the sublattice Ao of all
vectors c(x\ • ...• X24) in which each Xi is a multiple of 4 and their sum
is a multiple of 8.
If X and yare vectors of A the scalar product x . y is necessarily an
integer, since it equals lh«x+y)2 - x 2 - y2). Letting
x = c(x\ . ...• X24) be the typical vector of A, and taking scalar products
with the generating vectors of Ao, we see that c 2 . 8Xi and C2(4xi - 4x)
are integral for all i, j, or in other words that the coordinates Xi are
integers all of the same parity. (This explains our choice of the scale
factor c.) If A has any vector x whose coordinates Xi are all odd, we can
suppose (by changing the signs of suitable coordinate-vectors) that it has a
vector x that satisfies Xi = 1 (mod 4) for each i.
We now investigate what congruence conditions modulo 4 are imposed
on the Xi' To this end, we let n = {I,2 •...• 24} and let C€ be the set of
all subsets C of n for which there exists a lattice-point X with Xi=2 or 0
(mod 4) according as i E C or i ¢ C (x is said to correspond to C). In
general, if A and B are subsets of n, we use A + B for the symmetric
difference (A\B) U (B\A), and define their distance to be the cardinal of
A + B. We call two sets close if their distance is at most 4. If X and y
correspond respectively to C and D, then x + y corresponds to C + D, so
that the set C€ is closed under symmetric difference.
Theorem 3. Each non-empty C€ -set has at least 8 members. The number
of C€ -sets with exactly 8 members is at most 759, and the total number of
C€ -sets at most 212.
Proof Let C ¢ IlJ be a C€ -set with n members, and x a corresponding
vector. By subtracting a suitable member of Ao we can reduce all but one
of the Xi to 0 or 2, and the remaining Xi to ±2. But then x . x = c 2 . 4n,
and since x . x ~ 4 we must have n ~ 8. Now no 5-element subset of n
can be contained in two distinct C€-sets of cardinal 8, for then their
symmetric difference would be a C€-set of cardinal at most 6. The number
of C€-sets of cardinal 8 is therefore at most (254)/(~) = 759.

Now suppose the set A is close to the distinct C€-sets A + X and

A + Y. Then X + Y is a C€-set and so X and Y must be disjoint sets
with 4 members each. It follows that if A is close to more than one C€-set
it is close to at most 6, and distant 4 from each of these. If we define
f (A, C) to be 1, 1/6, or 0 according as the distance from A to C is less
than, equal to, or greater than 4, we can express this result as
~ f(A, C) ~ 1 for each A ,
CE'C

whence
334 Chapter 12

~ ~ f(A, C) :::;; 224.

A~n CE<fi

On the other hand we have plainly

~ f(A, C) = 1 + 24 + (224) + (234 ) + ~ (244) ,

A~n

and since the displayed number is exactly 212 there can be at most
224/212 = 212 C€-sets.
We now enumerate the vectors x with x· x = 4. Each such x
corresponds to a solution of the equation ~ xl = 32 in which the Xi are
integers all of the same parity, the number of Xi == 2 (mod 4) being zero
or at least 8. The only possibilities are therefore of the shapes
c«±2)8,0 16), c(+3, (±J)23), and c«±4)2,0 22 ). In the first case the
non-zero coordinates must form a C€-set, and we obtain at most 27 such
vectors per C€-set since otherwise we should have two lattice-vectors
differing by a vector of shape c(±4,0 23 ), which is impossibly short.
Hence we have at most 759 . 27 vectors of this shape. In the second case,
the lower sign must be taken on a C€-set, as we see by subtracting a vector
with every coordinate congruent to 1 modulo 4, and since there are 24
positions for the exceptional coordinate +3, we have at most 24· 212
vectors of this shape. We have already observed that A contains all the
(224) . 22 possible vectors of the third shape.

Theorem 4. Every 5 -element subset of C€ is contained in just one 8-

element C€ -set. The set C€ is the closure under symmetric difference of
the set of 8 -element C€ -sets.
Proof The preceding argument has shown that A contains at most
27 . 759 + 212 . 24 + 22 . (24)
2

vectors x with x . x = 4, and since the displayed number is exactly

196560 every inequality of that argument must in fact be an equality. In
particular each 5-element subset of n is contained in just one of the 759
8-element C€-sets. If now C is any non-empty C€-set we can find an 8-
element C€-set with at least 5 members in common with C, and so C is the
symmetric difference of this 8-element C€-set and a C€-set of smaller
cardinal. Repeating the argument, we see that C is an iterated symmetric
difference of 8-element C€-sets.
Theorem 5. The vector x = c(x 1 • . . . • X24) is contained in A just if
(i) the Xi are integers all of the same parity.
(ii) the set of i where Xi takes any given value (mod 4) is a C€ -set. and
(iii) ~ Xi == 0 or 4 (mod 8) according as each Xi == 0 or 1 (mod 2).
Proof For each 8-element C€ -set, A contains 27 corresponding vectors
c( (± 2) 8, 0 16), which are either all those with evenly many + signs or all
the others. But we know that A contains the 24 vectors of the shape
c( - 3, 123 ), and it cannot therefore contain any vector of the shape
A Characterization of the Leech Lattice 335

d-2, 27, 0 16), for we obtain from this the impossibly short lattice vector
c (I 8, (- 1) 16) by subtraction of a suitable vector d - 3, 123 ). A must
therefore contain the 759 vectors d2 8, 0 16) that correspond to the 759 8-
element rc -sets.
Let L be the set of all x satisfying (i) - (iii), and let
x =c (x 1..... x 24) be any vector of either A or L. We shall show that x
is a linear combination of the vectors d-3, 123 ), the 759 vectors d2 8, 0 16)
just discussed, and the generators of Ao, all of which belong to both A and
L. Subtracting d-3, 123 ) if necessary, we may suppose that the Xi are
even. Subtracting a number of vectors d2 8, 0 16), we may suppose the Xi
to be multiples of 4. Finally, subtracting a member of Ao, we can reduce
all but one of the Xi to 0, and the remaining one to 0 or 4. But since
x . x < 4 every coordinate is now zero, our result is proved, and A has
been identified with L.
Theorem 6. A is unique to within isomorphism.
Proof The first sentence of Theorem 4 asserts that the 8-element rc
-sets
form a Steiner system S (5,8,24). To complete the identification of A we
appeal to a theorem of Witt ([Wit3]; [Liin1], [Mac6, Chap. 20, §5]),
asserting that the system S (5,8,24) is unique to within a permutation of
n. The lattice is therefore completely determined.
Our argument gives slightly more. There are N 8/48 sets of 24 mutually
orthogonal pairs of vectors of length .J8, and anyone of these sets could
have been used to define our coordinate-system. In other words, the
automorphism group of A is transitive on the N 8/48 natural coordinate-
frames. On the other hand, the number of automorphisms fixing any
coordinate-frame is exactly 212 times the order of the Mathieu group M 24,
since any such automorphism can be obtained by changing the signs of a
rc -set of coordinates and then permuting n by a permutation fixing the
system S (5,8,24), and Witt defined M 24 to be the group of all such
permutations. Hence the order of the automorphism group is N8/48 times
212 times the order of M 24 , namely
2223 9 5472 11' l3. 23 = 8315553613086720000 .
Theorem 7. A is the only even unimodular lattice L with n < 32 that
satisfies anyone of the following.
(i) L is not directly congruent to its mirror-image.
(ji) No reflection leaves L invariant.
(iii) N 2(L) = O.
(iv) N2m (L) = 0 for some m ~ O.

Proof Plainly (i) implies (ii), and (iii) implies (iv). We show that (ij)
implies (iii).
If L has a vector x that satisfies x . x = 2, then the reflection in the
hyperplane through the origin perpendicular to x takes y to
X·V
y-2~x,
X'X
336 Chapter 12

that is, to y - (x . y)x, which is a lattice-vector whenever y is. Hence

this reflection leaves L invariant.
To complete the proof of the theorem we need only show that condition
(iv) determines A. But N 2m (0 has the form
65520
691 (1J'1l (m)-kT(m) )

where since N 2 (L) ~ 0 we must have k ~ 1. But then

N (L) ~ 65520 ( 11_2 11/2) > 0
2m "'" 691 m m

for all m > 1.

Since the discovery of the above proof, Niemeier [Nie2] has completed
his enumeration of the even unimodular lattices in 24 dimensions. He finds
24 such lattices, which are characterized by their configurations of vectors
of length .J2, and confirms that Leech's lattice is the only one without such
vectors (see Chaps. 16 and 17). Another characterization of the Leech
lattice is given in §5 of Chap. 18.
 13
Bounds on Kissing Numbers
A. M. Odlyzko and N. J. A. Sloane

Upper bounds are given on the maximal number, Tn, of nonoverlapping

unit spheres that can touch a unit sphere in n-dimensional Euclidean
space, for n:::;; 24. In particular it is shown that Tg = 240 and
T24 = 196560.

1. A general upper bound

The kissing number problem, as discussed in §2 of Chap. I, asks for the
maximal number Tn of nonoverlapping equal n -dimensional spheres that
can touch another sphere of the same size. The following theorem leads to
upper bounds on Tn. We assume n ~ 3.
Theorem. If f(t} is a real polynomial such that (Cl) f(t} :::;; 0 for
-I :::;; t :::;; Y2, and (C2) the coefficients in the expansion of f (t) in terms
of Jacobi polynomials [Chap. 22 of AbrIl
k
f(t} = ~ fip/"a(t} ,
i-O

where a = (n - 3) /2, satisfy fo > 0, f1 ~ 0, ... ,h ~ 0, then Tn is

bounded by Tn:::;; f (I) / f o·
Although this follows from the general theory developed in Chap. 9 (see
especially Eq. (49), (50», for completeness we sketch a simplified proof.
A spherical code re(§2.3 of Chap. l) is any finite subset of the unit
sphere in n dimensions. For -I :::;; t :::;; I let

AI =
o
_1- (number of ordered pairs c, c' € re for which (c, c') = t},
Irei
where 0I is a Dirac delta-function, Irei is the cardinality of re, and ( , ) is
the usual inner product. Then
338 Chapter 13

For all k ~ 0 we have

I I
Atp{,a(t)d( = --
1
~ p{,a ( (c,c') ) ~ 0,
-I Irt'l c,c',C{j

since the kernel p{,a ( (x,y) ) is positive definite ((10), (J 1) of Chap, 9).
If there is an arrangement of T unit spheres S I, . . . , S touching T

another unit sphere So, the points of contact of So with S I, . . . , S T form a

spherical code rt' with At = 0 for 1/2 < ( < 1. It follows that an upper
bound on Tn is given by the optimal solution to the following linear
programming problem: choose the At (- 1 :::;; ( :::;; V2) so as to maximize
J At d( subject to the constraints At ~ 0 for -1 :::;; ( :::;; 1/2 , and

I
1/2
Atp{,a(t)d( ~ _p{,a(1), for k = 0,1, ....
-I

The theorem now follows by passing to the dual problem, and using the
fact that any feasible solution to the dual problem is an upper bound to the
optimal solution of the original problem.

2. Numerical results
For n = 8 we apply the theorem with

J(t) = 3~0 (( + 1)(( + 1/2)2(2(( - 1/2 )

=p +.Ii p + 200 p + 832 p + 1216 p + 5120 p + 2560 p (I)

o 7 I 63 2 231 3 429 4 3003 5 4641 6

where Pi stands for p/.5,2.5(t), and obtain T8 :::;; 240. For n = 24 we take

J(t) = 1490944 (( + 1)(( + V2)2(( + V4)2(2 (t - 1;4)2(t - V2)

15
= p + ~ p + 1144 P + 12992 P + 73888 P
o 23 I 425 2 3825 3 22185 4
+ 2169856 p + 59062016 p + 4472832 P
687735 5 25365285 6 2753575 7
+ 23855104 p + 7340032 p + 7340032 p (2)
28956015 8 20376455 9 80848515 10,

where Pi stands for Pi lO .5,lO.5(t), and obtain T24 :::;; 196560. Since each
sphere in the E 8 lattice packing in 8 dimensions touches 240 others (§8.1
of Chap. 4), and each sphere in the Leech lattice packing in 24 dimensions
touches 196560 others (§11 of Chap. 4), we have determined T8 and T24'
For other values of n below 24 we were unable to find such simple and
effective polynomials. The results are summarized in Table 1.5 of Chap. 1,
which also gives the degree of the polynomial used. The best polynomial
we have found for n = 4, for example, is the 9th degree polynomial
Bounds on Kissing Numbers 339

f(t} = Po + alP I + a2P2+a3P3+ ... + a9P9, where al = 2.412237,

a2 = 3.261973, a3 = 3.217960, a4 = 2.040011, as = 0.853848, a6 = a7
= ag = 0, a9 = 0.128520 (shown to 6 decimal places, although we actually
used 17 places), and Pi stands for PiO.S,o.S(t). This implies T4 ~ 25.5585.
This polynomial was found by the following method. First replace (Cn by
a finite set of inequalities at the points tj = -1 + 0.0015)
(0 ~ ) ~ 1000). Second, choose a value of k, and use linear
programming to find f~ , ... ,f~ so as to minimize
k
'~Jtpt,cx(1)
i-I

subject to the constraints

k
fi* ~ 0 (I ~ i ~ k), ~ ftPt,CX(tj) ~ -1 (0 ~ ) ~ 1000).
i-I

Let f* (t) denote the polynomial 1 + ~~_I ftPt,CX(t}. Of course this need
not satisfy (Cn for all points t in the interval [-1,1/2 ). Let E be chosen to
be greater than the maximal value of f* (t) on [-1, V2) (E may be
calculated by finding the zeros of the derivative of f* (t)). Then
f (t) = f* (t) - E satisfies (Cn and (C2), and so
~ f* (1) - E
Tn"" I-E'

All the upper bounds shown in Table 1.5, except those for n = 3 and 17,
were obtained in this way. The degree k was allowed to be as large as 30,
but in all the cases considered the degree of the best polynomial (given in
the third column of the table) did not exceed 14. For n = 8 and n = 24
the form of the polynomials obtained in this way led us to (1) and (2), but
for the other values of n no such simple expression suggested itself.
For n = 17 we made use of the additional inequalities

f -..J3/2 Atdt
-I ~ 1 and
f-.J2i3 At dt
-I ~ 2

to obtain TI7 ~ 12215. Other inequalities of this type could probably be

used to obtain further improvements of these results. Unfortunately, for
n = 3 our methods only gave T3 ~ 13, whereas the actual value is 12 (§2.1
of Chap. 1).
For comparison the values of Coxeter's bounds on Tn ([CoxI6), [Bar2))
for n = 4, 5, 6, 7 and 8 as given in [Leel0) are 26, 48, 85, 146 and 244
respectively. The lower bounds in Table 1.5 are obtained from the
packings in Table 1.2.
 14
Uniqueness of Certain Spherical Codes
E. Bannai and N. J. A. Sloane

We show that there is essentially only one way of arranging 240 (resp.
196560) nonoverlapping unit spheres in R8 (resp. R24) so that they all
touch another unit sphere 0 n, and only one way of arranging 56 (resp.
4600) spheres in R8 (resp. R24) so that they all touch two further, touching
spheres. The following tight spherical t -designs are also unique: the 5-
design in 0 7, the 7-designs in 0 8 and 0 23 , and the II-design in 0 24 .

1. Introduction
It was shown in the previous chapter that the maximal number of
nonoverlap ping unit spheres in R S (resp. R24) that can touch another unit
sphere is 240 (resp. 196560). Arrangements of spheres meeting these
bounds can be obtained from the Es and Leech lattices, respectively. We
now prove that these are the only arrangements meeting these bounds. In
[Ban7], [Ban8] it was shown that there are no tight spherical t-designs for
t ~ 8 except for the tight II-design in 0 24 (cf. §3.2 of Chap. 3). The
present chapter also shows that this and three other tight t -designs are
unique. (Other results on the uniqueness of these lattices and their
associated codes and groups may be found in [Con4] = Chap. 12, [Cur3],
[DelI3], [Jon5], [Kne4], [LunI], [Mac6], [Nie21, [Ple8], [PleI7], [StaI2],
[Wit3].)
Our notation is that 0 n denotes the unit sphere in R n and ( , ) is the
usual inner product. An (n, M, s) spherical code (§2.3 of Chap. 1) is a
subset «j of On of size M for which (u, v) ;:::;; s for all u, v E «j, u ~ v.
Examples of spherical codes may be obtained from sphere packings via
the following theorem, whose proof is trivial.
Theorem 1. In a packing of unit spheres in Rn let S 1, . . . ,Sk be a set of
spheres such that Si touches Sj for all i ~ j. Suppose there are further
Uniqueness of Certain Spherical Codes 341

spheres T I • . . . • TM each of which touches all the Sj. Then. after

rescaling. the centers of T I • . . . • TM form an (n - k + 1, M, l/(k + I)
spherical code.
Example 2. In the E 8 lattice packing in R8 (§8.1 of Chap. 4) there are
240 spheres touching each sphere, 56 that touch each pair of touching
spheres, 27 that touch each triple of mutually touching spheres, and so on.
From Theorem 1 the centers of these spheres give rise to (8, 240, 112),
(7, 56, 113), (6, 27, 114), (5, 16, 115), (4, 10, 116) and (3, 6, 117)
spherical codes.
Example 3. Similarly the Leech lattice in R24 (§11 of Chap. 4) gives rise
to (24, 196560, 112), (23,4600, 113), (22,891, 114), (21,336, 115), (20,
170, 116), ... spherical codes.
If f(j is an (n, M, s) spherical code and u E f(j, the distance
distribution of C with respect to u is the system of numbers
{AI(u): -1 ~ t ~ I}, where
AI(u) = I{v E f(j: (u, v) = tll.
and the distance distribution of f(j is the system of numbers
{AI: -1 ~ t ~ I}, where

Then (see Chap. 13) the AI satisfy Al == 1, AI == ° for s < t < 1,

~ AI = M -1,

~ AIPk(t) ~ -Pk(J), for k ... 1,2,3 •....

-1<;;1<;;.

where Pk(x) = pln- 3)/2,(n-3)/2(x) is a Jacobi polynomial [Chap. 22 of

Abrll. For a specified value of s an upper bound to M is therefore given
by the following linear programming problem.
(pI) Choose {AI: -1 ~ t ~ s} so as to maximize
}; AI
-l~t:e;;;s

subject to the inequalities

~ AIPk(r) ~ -Pk(J), for k - 1,2,3, ....

-l~t~s

Then, as we have seen in Eq. (49), (50) of Chap. 9 and Theorem 1 of the
previous chapter, the dual problem can be stated as follows.
(P2) Choose an integer N and a polynomial f (t) of degree N, say
342 Chapter 14

N
f (t) = "l;hPk (t),
k-O

so as to minimizef(l)lfo subject to the inequalities

fo> O,fk ~ 0 for k = 1,2, ... ,N,
f(t) ~ 0 for -1 ~ t ~ s.

Since any feasible solution to the dual problem is an upper bound to the
optimal solution of the primal problem, we have
M ~ f(J)lfo

for any polynomial f (t) satisfying (2) and (3).

2. Uniqueness of the code of size 240 in n8

Theorem 4 (see Chap. 13). If ~ is an (8, M, V2) code then M ~ 240.
Proof Use the polynomial

f (t) = 3~0 (t + I)(t + V2)2 t 2(t - V2) . (5)

Theorem 5. If (a) ~ is an (8,240, V2) code then (b) ~ is a tight

spherical 7 -design in n 8, (c) ~ carries a 4 -class association scheme, (!)
(d) the intersection numbers of this association scheme are uniquely
determined, and (e) the distance distribution of ~ with respect to any
u E ~ is given by
A1(u) = A_1(u) = 1, A 1/2(U) = A_ 1/2(U) = 56, Ao(u) = 126. (6)

Proof Let {At} be the distance distribution of ~. Then {At} is an

optimal solution to the primal problem (PI), and the polynomial f(t} in
(5) is an optimal solution to the dual problem (P2). The dual variables
f I, . . . ,f6 are nonzero, so by the theorem of complementary slackness
[SimOl the primal constraints (1) must hold with equality for
k = 1, ... ,6.
The dual constraints (3) do not hold with equality except for
t -1, ± \t2 and O. Therefore the primal variables must vanish
=
everywhere except perhaps for A_I> A±I/2 and Ao. From (J) these
numbers satisfy the equations
A-1Pk ( - I) + A- 1/2P k (-V2) + AOPk (0) + A 1/2Pk (V2) = -Pk (J), (7)

for k = 1,2, ... ,6. Thus

(OCr. [BanIIl, [De19l, [Goe4l, [Hig2l, [Sei3l, [S105l.

Uniqueness of Certain Spherical Codes 343

239
7 7 7 7
0
2 4 4 2
63 9 9 9 63
A_I
8 8 8 8 8
231 33 33 231
0 A-I/2
16 64 64 16
(8)
3003
- - - -429
- 143 429
Ao --3003
-
128 256 128 256 128
--9009
- 1287 o_ 1287
AI/2 --9009
-
256 1024 1024 256
51051 663 1105 663 51051
1024
----
2048
-
1024 2048
---
1024

The unique solution is

A_I = 1, A-1/2 = AI/2 = 56, Ao = 126. (9)

Since A_I (u) =:;;; 1 and A_I = 1, we have A_I (u) = 1 for all U E ee,
and
so the code is antipodal [De116, p. 373]. Therefore (7) also holds for
k = 7 and by [De116, Theorem 5.5] ee
is a spherical 7-design. By [De116,
Definition 5.13] the design is tight, since Ieel - 2 (130 ]. By [De116,
Theorem 7.5] ee carries a 4-class association scheme. Therefore
At (u) ... At is independent of u for all t. This proves (b), (c) and (e).
The numbers (9) are the valencies of the association scheme, and by
[De116, Theorem 7.4] determine all the intersection numbers. This proves
(d).
Theorem 6. If condition (b) of Theorem 5 holds then so do (a), (c), (d)
and (e).

Proof By definition I eel = 2 (130 ) . From [De116, Theorem 5.12] the

inner products between the members of ee are ± 1 and the zeros of
C 3 (x) = 160(x + Y2)x(x - Y2).

Thus all the At are zero except perhaps for A ± to A ± 1/2 and Ao. From
[De116, Theorem 5.5] Eq. (7) holds for k = 1,2 •...• 7. The rest of the
proof is the same as for Theorem 5.
In Example 2 we saw that the minimal vectors in the E s lattice form
an (8, 240, V2) code. Thus conditions (a)-(e) of Theorem 5 apply to this
code. Conversely we have:
Theorem 7. If eeis a tight spherical 7-design in Os there is an
orthogonal transformation mapping ee onto the minimal vectors of the Es
lattice.
344 Chapter 14

Proof From Theorem 6 the possible inner products in rc

are 0, ± \12, ± 1.
Let rc
= {Ul, . . . ,U240} and let L be the lattice in R8 consisting of the
vectors
240
2'; .Ji aj Uj , aj E Z.
j-l

Then L is an eight-dimensional even integral lattice generated by vectors

of norm 2, and is therefore (see §3 of Chap. 4) a direct sum of various
lattices An (n ~ 1), Dn (n ~ 4) and En (n = 6, 7, 8). The only lattice of
this type with at least 240 minimal vectors is E 8, so L is isometric to E 8
and rcis isometric to the set of minimal vectors in E 8.
By combining Theorems 5 and 7 we obtain:
Theorem 8. There is a unique way (up to isometry) to arrange 240
nonoverlapping unit spheres in R8 so that they all touch another unit
sphere.

3. Uniqueness of the code of size 56 in n7

Theorem 9. If rc is a (7, M, 1/3) code then M ~ 56.

Proof From (4), using the polynomial

f(t) = (( + I)(t + 1/3)2(( - 1/3).

Theorem 10. If (a) rc

is a (7, 56, 1/3) code then (b) is a tight rc
spherical 5-design in 0 7 , (c) rc
carries a 3-class association scheme, (d)
the intersection numbers of this association scheme are uniquely
determined, and (e) the distance distribution of rc
with respect to any
u E rc
is given by

(10)

Conversely (b) implies (a), (c), (d) and (e).

Proof The proof is parallel to the proofs of Theorems 5 and 6.
For example the (7, 56, 1/3) code given in Example 2 has properties
(a) -(e). Conversely we have:
Theorem 11. If rc
is a tight spherical 5-design in 0 7 there is an
orthogonal transformation mapping rc
onto the (7, 56, 1/3) code obtained
from the E8 lattice.
Proof Let rc
consist of the points u 1, . . . , U 56 lying on a unit sphere in
R7 centered at P. Choose a point 0 (in R8) so that -1:. ujOP = 7r/3 for all
i, and thus

for all i,j. Let v be a unit vector along OP (see Fig. 14.1). From
Uniqueness of Certain Spherical Codes 345

o
Figure 14.1. The construction used in the proof of Theorem 11:
1:. UiOP = 7r/3 for all i, lopl = 1/J3, I Ou,1 = IOu21 = 2/J3, and
cos cf> = (1 + 3 cos 8)/4.

Theorem 10 cos 1: UjPUj takes the values ± 1 and ± 1/3, so cos 1: UjOUj
takes the values 0, ± '/2 and 1. It follows that the vectors .J3fiOUj
(I ~ i ~ 56) span an even integral lattice, containing at least
2(56 + 1) = 114 minimal vectors (corresponding to ±~, ±v). This
lattice must therefore be either E 8 or E 7 ED A 1> and the latter is
incompatible with (I 0).
By combining Theorems 10 and 11 we obtain:
Theorem 12. There is a unique way (up to isometry) to arrange 56
nonoverlapping unit spheres in R8 so that they all touch two further,
touching, unit spheres.

4. Uniqueness of the code of size 196560 in n 24

Theorem 13 (see Chap. 13). If ~ is a (24, M, V2) code then
M ~ 196560.
Proof From (4), using the polynomial
f(t) = (t + I)(t + V2)2(t + 'J4)2t 2(t - ',4)2(t - '/2).

Theorem 14. If (a) ~ is a (24, 196560, V2) code then (b) ~ is a tight
spherical II-design in n 24, (c) ~ carries a 6-class association scheme,
(d) the intersection numbers of this association scheme are uniquely
determined, and (e) the distance distribution of ~ with respect to any
U E ~ is given by

A1(u) = A_1(u) = 1, A l /2(U) = A- 1/2(U) = 4600,

346 Chapter 14

Conversely (b) implies (a), (c), (d) and (e).

Proof The proof is parallel to those of Theorems 5 and 6.
In Example 3 we saw that the minimal vectors in the Leech lattice A
when suitably scaled form a (24, 196560, Ih) code. We shall require an
explicit description of this code, and take A to consist of the vectors in
(135), (136) of Chap. 4. Let A4 denote the set of 196560 minimal vectors
(of norm 4) - see §11 of Chap. 4. Then IhA4 is a (24, 196560, Ih) code
to which conditions (a)-(e) of Theorem 14 apply. Conversely we have:
Theorem 15. If C€ is a tight spherical 11-design in 0 24 there is an
orthogonal transformation mapping C€ onto V2 A 4.
Proof From Theorem 14 the distance distribution of C€ with respect to
any u E C€ is given by (I 1), and in particular the inner products in C€ are
0, ± If.!, ± Ih, ± 1. Let C€ = {u J, U2, . . . , u 196560}, and let L be the lattice
in R24 consisting of the vectors
196560
~ 2a; U;, a; E Z. (12)
;-1

Then

and L is a 24-dimensional even integral lattice. We shall establish

Theorem 15 by showing that there is an orthogonal transformation
mapping L onto 2A and C€ onto V2 A4.
Lemma 16. The minimal norm (v, v) for vEL, v ¢ 0, is 4.
Proof The minimal norm is even, so suppose it IS 2, with
(v,v) =2,v E L. Foru E 2C€wehave
l(u,v)1 = lul·lvl·lcos ~ uOvl ~ 25,

so (u, v) E {O, ± 1, ± 2} since L is integral. Suppose (u, v) = for ex

choices of u, (u, v) = 1 for fJ choices, and (u, v) = 2 for ')' choices, with
°
ex + 2fJ + 2')' = 196560. Without loss of generality we may assume
v = (.J2,0, 0, ... ,0).

Since C€ is an II-design,
1 196560 1 r
196560 .~ f(u;) = - J{}j(~)dw(~)
.-1 W24

holds for any homogeneous polynomial f(~J,~2' ... '~24) of total degree
~ 11, where W24 is the surface area of 0 24 (Eq. (81) of Chap. 3). Let us
choosef = h = ~t, for k = 2 and 4, so that
h (u;) = rk/2((u;, v ))k.

The right hand side of (14) can be evaluated from

Uniqueness of Certain Spherical Codes 347

1 ~ h(u)
196560 u E (J/2)A4

8190 945
- - - if k = 2, or if k = 4.
196560 196560

The equations (I4) now read

12 22 14 24
2{3' - + 2'Y' -8 = 8190 2{3' - + 2,)" - = 945
8 ' 64 64 '

which imply {3 = 33600, 'Y = -210, an impossibility.

Lemma 17. The set L4 of vectors of norm 4 in L coincides with 2C(i.
Proof By construction L4 contains 2 C(i. Conversely take u, v E L 4. Then
(u, v) ~ 3, or else
(u - v, u - v) = (u, u) - 2(u, v) + (v, v) = 2,

contradicting Lemma 16. Similarly (u, v) ~ -3. Therefore (u, v)

E to, ±1, ±2, ±4} and 1:. uOv ~ 7r/3 for u ~ v. From Theorem 13,
1L41 ~ 196560 = 12 C(il. Therefore L4 = 2 C(i.

For n ~ 3, the vectors

gl = .J'i(el + e2), g2 = .J'i(el - e2),
g3 = .J'i(e2 - e3), ... ,gn = .J'i(en-I - en), (I5)

with respect to an orthonormal basis {e I, . . . , en} for R n span a lattice Dn

(§7.1 of Chap. 4). There are 2n (n - 1) minimal vectors « ±.J'i) 20n-2) in
Dn. These lattices are nested: D3 ~ D4 ~ .. '. Ot must be understood
that all lattices Dn in this chapter are scaled to have minimal norm 4.)
Lemma 18. (j) For any pair of vectors u, v in A4 with 1:. u Ov = 7r 12 there
are 44 vectors w in A4 with 1:. uOw = 1:. vOw = 7r/3. (ii) The same
statement holds with A4 replaced by L4 = 2 C(i. (iij) There are 2n - 4
minimal vectors w in Dn such that 1:. glOw = 1:. g20w = 7r 13.
Proof (i) and (iij) are straightforward, and (ij) follows from (i) since A4
and 2 C(i are association schemes with the same parameters (Theorem 14).
Lemma 19. L contains a sublattice D 3.
Proof For the generators gh g2, g3 of D3 we can take any triple
u, v, w E L4 with 1:. uOv = 7r/2, 1:. uOw = 1:. vOw = 7r/3. Such a triple
exists by Lemma 18 (ij).
Lemma 20. L contains a sublattice Dn , for n = 3,4, ... , 24.
Proof We proceed by induction on n. Suppose the assertion holds for
n ~ 3. By choosing a suitable orthonormal basis el, ... , en, we see that
L4 contains vectors g}, ... , gn given by (I 5) and spanning Dn. By
Lemma 18 (ij) there are 44 vectors w in L4 with
348 Chapter 14

~glaw = ~g2aW = 7r/3. By Lemma 18 (iii) at least one of these is not a

minimal vector of Dn. Then this vector w is not in RDn. (For suppose
w = Clel+···+cnen. Since ~glaw = ~g2aw = 7r/3, CI = Ji and Cz = a.
For 3 ~ i ~ n,

and therefore
(w, Ji(el ± e») E {a, ± 1, ±2)

from (13). This implies C3 = C4="'=Cn = a, and contradicts (w,w) = 4.)

Choose en+1 so that {el.' ... en+d is an orthonormal basis for R<Dn,w>,
and suppose

The above argument shows that CI = -J2, C2 = ... = Cn = a, and

Cn+1 = ±Ji. Therefore <Dn,w> = Dn+1 ~ L.
Lemma 21. L is isometric to A.
Proof From Lemma 2a we may choose an orthonormal basis e I • . . . • e24
so that 2«j contains the vectors (±Ji)2a22 . Let U = (UI • . . . • U24)/.J8 be
any vector in 2 «j. From (13) the inner products of U with the vectors
(± Ji) 2a 22 are a, ± 1, ± 2, ± 4. By considering the inner products with
(Ji, ± Ji, a..... a) we obtain
uf + uf+'" +Uf4 = 32,
If2(UI ± uz) E {a, ±1, ±2, ±4),
U]'U2,'" E { a, ± 1, ±2, ±3, ±4, ±5 l.
Suppose UI = ±5. Then another Ui, say U2, is zero. The inner product of
u with (Ji, Ji,a . .... a) is 512, a contradiction. Proceeding in this way
it is not difficult to show that the only possibilities for the components of U
are

In particular UI • . • • • UZ4 are integers with the same parity.

It remains to show that these vectors are the same as those in A4 . To
see this we define a binary linear code g) of length 24 by taking as
codewords all binary vectors C for which there is a vector U E L with
U = (a + 2c + 4x)/.J8
for some x E Z24. Then as in Theorem 3 of Chap. 12 it follows that wt(c)
~ 8 for C ;=: a, and that there are at most 759 codewords of weight 8.
Therefore I g)1 ~ A (24,8) = 212, from Table 9.l. The argument following
Theorem 3 of Chap. 12 now shows that the only way that 2 g) can contain
Uniqueness of Certain Spherical Codes 349

196560 vectors u is for these vectors to coincide with the minimal vectors
in A4 .
This completes the proof of Theorem 15. By combining Theorems 14
and 15 we obtain:
Theorem 22. There is a unique way (up to isometry) to arrange 196560
nonoverlapping unit spheres in R24 so that they all touch another unit
sphere.
Remark. Since the Leech lattice has two distinct mirror image forms, the
isometry of Theorem 22 might necessarily involve a reflection.

5. Uniqueness of the code of size 4600 in n 23

Theorem 23. If C(j is a (23, M, 1/3) code then M ~ 4600.
Proof From (4), using the polynomial
f (t) = (t + l)(t + 1/3)2t 2(t - 1/3) .

Theorem 24. If (a) C(j is a (23, 4600, 1/3) code then (b) C(j is a tight
spherical 7-design in n 23, (c) C(j carries a 4-class association scheme, (d)
the intersection numbers of this association scheme are uniquely
determined, and (e) the distance distribution of C(j with respect to any
u E C(j is given by
A1(u) =A_1(u) = 1, A 1/3(U) -A- 1/3(U) -891, Ao(u) -2816.

Conversely (b) implies (a), (c), (d) and (e).

For example the (23, 4600, 1/3) code given in Example 3 has
properties (a) -(e). Conversely we have:

Theorem 25. If C(j is a tight spherical 7-design in n 23 there is an

orthogonal transformation mapping C(j onto the (23, 4600, 1/3) code
obtained from the Leech lattice.
Proof As in the proof of Theorem 11 we embed C(j = {u I, . . . , U4600} in
R24, choosing 0 so that ~ ujOP = 7r /3 for all i (cf. Fig. 14.1). Then

cos ~ ujOu j E {-Y2, 0, V-i, Y2, 1}.

Let L be the even integral lattice in R24 spanned by the vectors J3 OUj.
For convenience we set Uj = J30Uj.

Lemma 26. The minimal norm (v, v) for vEL, v ~ 0, is 4.

Proof Suppose vEL with (v,v) = 2, and write v = v' + v" with
v' II OP, v" ..l. OP, Iv'l = y, Iv"l = (2 - y2)1/2, and Uj = U'j + U"j with
U'j II OP, U'j ..l. OP, I U'jl = 1, 1 U"jl = 31/ 2 • Then
350 Chapter 14

(Vj,v) = (V'j,v') + (V"j,V") E {O, ±1, ±2},

COS.,..
~ V" jV
0 "E {O, ± 1, ±2} - Y
~~
,,3,,2 - y2

Since C€ is a tight 7-design, the set {cos 4:. V"jOv": 1::S:;; i ::s:;; 4600} is
symmetric about o. Therefore y E {O, ± \12, ± I}. First suppose y ... o.
Then

~ 0" {2 1 0 1 2} .
cos.,.. V" j v E - ..[6' - ..[6' , ..[6'..[6

Let these values occur 1', {3, a, {3, I' times respectively. Then by evaluating
the Oth, 2nd and 4th moments of C€ with respect to v", as in the proof of
Lemma 16, we obtain the equations a + 2{3 + 21' = 4600,
{3/3 + 41'/3'" 200, {3/8 + 81'/9 = 24, which imply I' = -14, an
impossibility. Similarly for the other values of y.

Lemma 27. L contains a sublattice isometric to Dn , for n = 3,4, ... ,24.

Proof. This is similar to the proof of Lemma 20, starting from the fact
that if we take u h U2 E C€ with 4:. u ,0 U2 = 7r /2, there are 42 vectors
Uj E C€ with 4:. u, 0 Uj = 4:. U 20 Uj = 7r /3. Furthermore the vector
v = 20P E L also satisfies 4:.u,Ov = 4:.U20V = 7r/3.

Lemma 28. L is isometric to A, and C€ is isometric to the (23, 4600, 1/3)

code obtained from the Leech lattice.

Proof. Let L4 denote the set of minimal vectors in L. From Lemma 27

we may assume that L4 contains all the vectors «±42022 »/.J8, and that
v - 2·0P is (440 ... 0)/.J8. As in Lemma 21 it follows that the vectors in
L4 have the form «±2)80'6)/.J8, «±42022 )/.J8 and «±1)23(±3)1)/.J8.
Furthermore the vectors Vj begin (2,2 .. J /.J8, (4,0 .. J /.J8,
(0,4 .. J /.J8, (3, I..J /.J8, or (I ,3 .. J /.J8. The code !!) is defined as in
Lemma 21: it is a linear code of minimum distance 8 containing at most
212 codewords. The zero codeword corresponds to the vectors Vj that begin
(4,0 .. J/.J8 or (O,4 .. J/.J8, and there are at most 2·2·22 of these. The
codewords of weight 8 that begin 11 ... correspond to the vectors Vj that
begin (2,2 .. J/.J8. The number of such codewords is at most 77, since
A (22,8,6) = 77 [Mac6, Fig. 3, p.688], and there are at most 25.77
corresponding Vj. The remaining Vj come from codewords beginning 10...
or 01..., and there are at most 2.2 10 of them (since A (22,8) ::s:;; 210 by
Table 9.1). But 2·2·22 + 25.77 + 2.2 10 = 4600, so all the inequalities in
the argument must be exact. In particular the codewords of weight 8 that
begin 11... must form the unique Steiner system S(3, 6, 22) ([Liinl1,
[Mac6, p. 645], [Wit3]), and hence L must be the Leech lattice.
This completes the proof of Theorem 25. By combining Theorem 24
and 25 we obtain:
Uniqueness of Certain Spherical Codes 351

Theorem 29. There is a unique way (up to isometry) to arrange 4600 unit
spheres in R24 so that they all touch two further, touching, unit spheres.
(Once again, the isometry might necessarily involve a reflection.)
Acknowledgements. We should like to acknowledge helpful conversations
with C. L. Mallows, A. M. Odlyzko and 1. G. Thompson.
 15
On the Classification of Integral
Quadratic Forms
J. H. Conway and N. J. A. Sloane

This chapter gives an account of the classification of integral quadratic

forms. It is particularly designed for readers who wish to be able to do
explicit calculations. Novel features include an elementary system of
rational invariants (defined without using the Hilbert norm residue
symboO, an improved notation for the genus of a form, an efficient way to
compute the number of spinor genera in a genus, and some conditions
which imply that there is only one class in a genus. We give tables of the
binary forms with -100 ~ det ~ 50, the indecomposable ternary forms
with Idetl ~ 50, the genera of forms with Idetl ~ 11, the genera of p-
elementary forms for all p, and the positive definite forms with
determinant 2 up to dimension 18 and determinant 3 up to dimension 17.

1. Introduction
The project of classifying integral quadratic forms has a long history, to
which many mathematicians have contributed. The binary (or two-
dimensionaO forms were comprehensively discussed by Gauss. Gauss and
later workers also made substantial inroads into the problem of ternary and
higher-dimensional forms. The greatest advances since then have been the
beautiful development of the theory of rational quadratic forms
(Minkowski, Hasse, Witt), and Eichler's complete classification of
indefinite forms in dimension 3 or higher in terms of the notion of spinor
genus.
Definite forms correspond to lattices in Euclidean space. For small
dimensions they can be classified using Minkowski's generalization of
Gauss's notion of reduced form, but this method rapidly becomes
impracticable when the dimension reaches 6 or 7. However there is a
geometric method used by Witt and Kneser which (after the work of
Niemeier) is effective roughly until the sum of the dimension and the
On the Classification of Integral Quadratic Forms 353

(determinant) 1/2 exceeds 24, beyond which point it seems that the forms
are inherently unclassifiable. The situation is summarized in Fig. 15.1.
There are several novel features of this chapter.
(1) We present On §5) an elementary system of rational invariants for
quadratic forms, defined without using the Hilbert norm residue symbol,
and whose values are certain integers modulo 8. The modulo-8 version of
the 2-adic invariant seems to have first arisen in topological investigations
(see [CasI], [Hir4]). Practitioners in the subject know that the effect of
the "product formula" is to yield congruence conditions modulo 8 on the
signature. With the invariants we use these congruences emerge
immediately rather than at the end of a long calculation (see Eqs. (15) and
(16».
(2) We also give a simply described system of p-adic invariants for
integral forms that yields a handy notation for the genus of a quadratic
form (§7).
(3) In terms of this new notation we enumerate Gn §8.1 and Table 15.4)
all genera of quadratic forms having determinant of magnitude less than
12, and
(4) classify the genera of p-elementary forms for all p (§8.2).
In view of Eichler's theory of spinor genera (see Theorem 14), these
results actually give the integral equivalence classes in the case of indefinite
forms of dimension ~ 3.
(5) We also give a simple description of the spinor genus, including a
notation for the various spinor genera in the genus of a given form, and an
easy computational way of finding their number (§9).

Dimension Definite Indefinite

1 Trivial Trivial

2 Gauss: Gauss: cycles of

reduced forms. reduced forms.

3 Minkowski:
reduced
forms
...
7 Eichler:
Kneser- spinor
Niemeier genus
gluing
method.
24 ...

Impracticable.
Figure 15.1 How quadratic forms are classified.
354 Chapter 15

(6) We include some theorems gIVIng conditions under which a genus

contains just one class (Theorem 20 and its Corollaries). In particular we
show that, if f is an indefinite form of dimension n and determinant d,
and there is more than one class in the genus of f, then
[~l n
4 2 d .
IS d'IVISI
. 'bl e by k (2)

for some nonsquare natural number k

(7) We also give a number of tables:
=
0 or 1 (mod 4).

- reduced binary quadratic forms whose determinant d satisfies Id I ~ 50

for definite forms, Id I ~ 100 for indefinite forms (Tables 15.1, 15.2),
- indecomposable ternary forms with Id I ~ 50 for definite forms and
Idl ~ 100 for indefinite forms (Tables 15.6, 15.7),
- genera of quadratic forms with Id I < 12 (Table 15.4),
- definite quadratic forms of determinant 2 and dimension ~ 18 (15.8),
- definite quadratic forms of determinant 3 and dimension ~ 17 (15.9).
Our treatment is addressed to a reader who wishes to be able to do
explicit calculations while gaining some understanding of the general
theory. The chapter is arranged as follows. §§2 and 4 contain definitions
and other mathematical preliminaries. §3 deals with binary forms. §§5,6
treat the classification of forms over the rational numbers, and §7 the
classification over the p -adic integers and the associated notion of the
genus of a form. §8 gives some applications of the results of §7. The
spinor genus of a form and the classification of indefinite forms are dealt
with in §9, and definite forms in §1O. The final section discusses the
computational complexity of the classification problem.

2. Definitions
2.1 Quadratic forms. Let x be the row vector (x), X2) and A the
symmetric matrix [b ~). Then the expression

is called the binary quadratic form with matrix A. Calculations with this
form often involve the associated bilinear form
f(x, y) = xAytr = aX\YI + bXlY2 + bx2Yl + CX2Y2'
Replacing A by a symmetric matrix of arbitrary dimension n we obtain
the notion of an n-ary quadratic form and of an n-ary symmetric bilinear
form (see §2.2 of Chap. 2). From the very large number of references on
quadratic forms let us mention in particular [Bor5], [Cas3], [Cas4],
[Coh5], [Dav2a], [Dic2], [Dic3], [Eicl], [Gaul], [HsilJ-[Hsi9], [Jon3],
[Knel]-[Kne9], [LamIl, [MiI7], [MinO]-[Min3], [O'Mel]-[O'Me4],
[Orzl], [Ran3], [Ran5], [Ran5a], [Riel], [Rysll-[RysI4], [SchO]-[Sch2],
[Sch 13], [Sch 14], [Serl], [Smi5], [Smi6], [Tau2], [Wae5], [Wat3]-
[Wat22], [Witl], [Zagll.
On the Classification of Integral Quadratic Forms 355

We are concerned with the classification of integral quadratic forms

under integral equivalence. There are two definitions of integrality for a
quadratic form. The binary form (2) is integral as a quadratic form if a,
2b and c belong to Z, and integral as a symmetric bilinear form if its
matrix entries a, b, c belong to Z. The latter is the definition used by
Gauss [Gaul]; Cassels [Cas3] calls such a form classically integral.
Although some authors hold strong opinions about which definition of
integrality should be used (Watson [Wat3] refers to "Gauss's mistake of
introducing binomial coefficients into the notation"), it makes very little
difference for the classification theory. The two definitions are not really
rivals but collaborators. For if f is integral in either sense then 2f is
integral in the other sense, and f is equivalent to g if and only if 2f is
equivalent to 2g. Since for algebraic purposes a form is usually most
conveniently specified by its matrix entries, we prefer the second definition,
and so in this book we call f an integral form if and only if its matrix
entries are integers (i.e. if and only if it is classically integral, or integral as
a symmetric bilinear form).
2.2 Forms and lattices; integral equivalence. Some geometric ideas are
appropriate. We consider a rational vector space V with inner product
( , ) and refer to (x, x) as the norm of x. If V is 2-dimensional, and
spanned by vectors el and e2 with

then the norm of the vector x = xlel + X2e2 is just f(x), and its inner
product with the vector y = Ylel + Y2e2 isf(x, y).
The vectors x = xlel + X2e2 for which XI and X2 are integers form a
lattice in V, and el, e2 is an integral basis for this lattice. The other
integral bases for this lattice have the form ae I + {3e2, 'Ye I + oe2 , where
a, {3, 'Y, 0 are integers with ao - {3'Y = ± 1. We shall therefore call two
binary forms f and g with matrices A and B integrally equivalent. or say
they are in the same class. and write f - g, if there exists a matrix

with integer entries and determinant ± 1 for which B = M A M lr •

Geometrically, f and g refer to the same lattice with different integral
bases. The equivalence is proper if det M = + 1, improper if
det M =-1.
The general case is exactly similar (§2.2 of Chap. 2). A symmetric
matrix A = (ai) with integer entries determines a (classically) integral
form f, whose values are the norms of the members of a lattice A (in an
n -dimensional vector space) that has an integral basis e I • . . . • en with
(ei, ej) = aij' If M = (mij) has integer entries then the vectors
el'=!,mljej, ... ,en'=!,mnjej will generate a sublattice of A, and this
will be all of A if and only if M- I has integer entries; this happens if and
only if det M = ± 1.
356 Chapter 15

We therefore say that two n-ary forms f and g with matrices A and B
are (properly or improperly) integrally equivalent if their matrices are
related by
B =MAMlr

for some M with integer entries and determinant + 1 or -1 respectively.

It is clear that the number d = det A is an invariant of f for integral
equivalence (j.e. if f and g are integrally equivalent, det A = det B), and
we shall call d the determinant of f. The reader should be aware that
many authors use the term discriminant for this number multiplied by
certain powers of 2 and -1 that depend on the dimension (and the
author).
In practice, when transforming f into an equivalent form, we derive the
matrix B from A by performing elementary row operations (mUltiplying a
row by a unit or adding multiples of one row to another), followed by the
exactly corresponding column operations.
These notions can be immediately generalized to arbitrary rings R
(with 1). A form f is defined over R if it is represented by a matrix A
with entries from R, and two forms f and g are equivalent over R if (4)
holds for some M with entries from R and with a determinant which is a
unit of R (j.e. an element of R with an inverse in R). For example,
taking R to be the rational numbers Q, we see that the forms x 2 + y2,
2z2 + 2t2, although not integrally equivalent, are rationally equivalent, as
shown by the formulae
x=z+t, z=Y2(x+y),
y=z-t, t=Ydx-y).

If f and g are rationally equivalent, the ratio of their determinants is the

square of a nonzero rational number.

3. The classification of binary quadratic forms

Almost everything one can say about the classification of binary quadratic
forms was already said by Gauss in his Disquisitiones arithmeticae
[Gaul]. The complete classification in the indefinite binary case uses
cycles of reduced forms, and is totally unlike Eichler's complete
classification of indefinite forms in dimensions ~ 3. A very clear account
has recently been given by Edwards [Edwl], who also discusses the
connections with the ideal theory of quadratic number rings, and so here
we shall only present the results, together with a brief indication of these
connections. Other treatments of the binary case may be found in [Cas3],
[DavI], [Jon3], [LeVI], [Wat3], [ZagI], etc.
3.1 Cycles of reduced forms
Theorem 1 (Quoted with changes in notation from [Edwl, p. 3251.)

Let [~~ ~~] (abbreviated aobo a\) be an integral binary quadratic form of
On the Classification of Integral Quadratic Forms 357

determinant d = aoal - b6, and suppose that -d is not a square. We

define a sequence

(5)

(abbreviated

of binary forms of determinant d by the rules:

b·2+d
ai and bi determine ai+1 as -'--,
ai
bi and ai+1 determine bi+! as the largest solution of
bi + bi+! == 0 (mod ai+l)

for which
b/'+I + d < 0 (7)

if such solutions exist, and otherwise as the smallest solution of

(6) in absolute value, taking bi+! positive in case of a tie.
Then the sequence of forms derived in this way from the given form is
ultimately periodic, the forms in the period being called a cycle of reduced
forms. Furthermore two binary forms are properly equivalent if and only
if they lead to the same cycle of reduced forms.
The condition that -d be not a square implies that the ai are nonzero,
and so ai+1 and bi+1 are well-defined. Only minor modifications are
required when -d is a square - see §3.3.
As an example, the sequence of forms arising from x 2 - 67y2 is

0087527725788
+1 -67 +1 -3 +6 -7 +9 -2 +9 -7 +6 -3 +1 -3 ... (8)

There are ten reduced forms in the cycle. On the other hand -x 2 + 67y2
leads to the disjoint sequence obtained by changing the signs of the lower
terms in (8) (the ai's), and is therefore an inequivalent form.

3.2 Definite binary forms. In the case of a definite (say positive definite)
binary form, it follows from Theorem 1 that the cycle contains either a
single form

[b ~), with b = 0 or 2b = la 1 ,

or else two forms

[a b) [c-b -b)a '

b c '
(10)
358 Chapte r 15

The term reduced is normally applied only to the single form (9) or to
whichever of the pair (10) satisfies
a < C , or a = c and b > 0.
It is not difficult to show that a positive definite form (b ~ 1is reduced in
this sense if and only if it satisfies
-a < 2b ~ a ~ c, with b ~ 0 if a = c
[Dic3, Theorem 99], [Jon3, Theorem 76]. Furthermore every posItive
definite form is properly equivalent to a unique such reduced form. Since
(I 1) implies b 2 ~ d 13, all reduced forms are easily enumerated: for each b
with Ibl ~ .Jd13 we factor d+b 2 = ac in all possible ways consistent with
(I1) (see Table 15.1).
The reduction condition (11) expresses the fact that
a is the absolutely smallest value taken (at a vector e I say) by the form,
and
c is the absolutely smallest value that can be taken by the form at a
vector e2 independent of el .
An appropriate generalization of these conditions to higher dimensions
leads to the notion of a Minkowski reduced form (see §1O. 1).
(I 1) is also equivalent to the assertion that the root z = x Iy of
f (x, y) = 0 in the upper half plane lies in the region Iz I ~ 1,
-lj2 ~ Re z ~ Ij2 shaded in Fig. 15.2. As the form undergoes unimodular
transformations

(~ ~l E SL 2 (Z) ,

i.e. with a, {3, 'Y, 0 E Z and a{3 - 'YO = 1, this root transforms into az +(3 ,
'Yz+o

-1 o 1
2"

Figure 15.2 The fundamental region for PSL 2 (Z).

On the Classification of Integral Quadratic Forms 359

and the region mentioned is a fundamental region for the action of

SL 2 (Z) on the upper half plane (see for example [LeVI, Vol. 1, Chap.
I]). This notion also generalizes to higher dimensions, and it is an
important theorem that in the generalization the fundamental region has
only finitely many walls (which correspond to inequalities on the matrix
entries of the form).
3.3 Indefinite binary forms. In the indefinite case, supposing that -d is
not a square, it can be shown from Theorem I that the reduced forms
(those in the cycle) are precisely the ones satisfying
0< b < r-fi < min {b+lal, b+lcl} (12)

[Gaul, §183]. Again the reduced forms are easily found: for each positive
integer b < r-fi we factor d+b 2 = ac in all possible ways satisfying
(12) .
When -d is a square, Theorem 1 must be modified slightly, and in
particular the inequality in (7) must be replaced by ~. Then the process
of Theorem I terminates in a form (b g] with b = r-fi. If the process
is extended backwards it terminates in a f~rm (~ ~]. Gauss [Gaul,
§§206-21O] proved that two forms (b g], (~ g] are properly equivalent
if and only if a == a' (mod 2b), and are improperly equivalent if and only
if
aa' == gcd(a, b)2 (mod 2b gcd(a, b» .

We therefore extend the notion of reduced form in this case to include,

besides the forms satisfying (12), all forms of the shape
ob a or abO (-b < a ~ b) .

Then the "cycle" of reduced forms becomes a finite sequence

Gauss declared that tables of binary quadratic forms should not be

published, since they are so easily computed [Lehl, p. 691. Nevertheless
we feel the usefulness of our paper is enhanced by Tables 15.1 and 15.2,
which enumerate all reduced binary quadratic forms with
- 100 ~ d ~ 50. The notation is that introduced in Theorem 1. Table
15.1 gives the positive definite forms with d ~ 50, and Table 15.2 the
cycles of reduced indefinite forms. In Table 15.2 four related cycles

... +ab-cd+ef-g h .. .
... -ab+cd-ef+g h .. .

... h +g f -e d +c b -a .. .
... h_gf +ed-cb+a .. .
Table 15.1. Reduced positive definite binary forms.
d Forms

10 1
10 2
1 0 3, 2 I 2
4 1 0 4, 2 0 2
1 0 5, 2 I 3
1 0 6, 2 0 3
1 0 7, 2 I 4
I 0 8, 2 0 4, 3 I 3
9 I 0 9, 3 0 3, 2 I 5
10 1 0 10, 20 5
II 1 0 I I, 2 I 6, 3 ± 14
12 I 0 12, 2 0 6, 3 0 4, 4 24
13 I 0 13, 2 I 7
14 1 0 14, 2 0 7, 3 ± 1 5
15 I a 15, 3 a 5, 2 I 8, 4 I 4
16 I a 16, 2 a 8, 4 a 4, 4 2 5
17 1 0 17,2 1 9, 3±16
18 1 0 18,2 0 9,3 0 6
19 1 0 19, 2 I 10, 4 ± 15
20 I 0 20, 2 a 10, 4 a 5, 3 ± I 7, 4 2 6
21 1 0 21,3 0 7,2 1 11,5 2 5
22 I a 22, 2 a 11
23 I a 23, 2 I 12, 3 ± I 8, 4 ± I 6
24 I 0 24, 2 a 12, 3 a 8, 4 a 6, 5 I 5, 4 2 7
25 I a 25, 5 a 5, 2 I 13
26 1 0 26,2 0 13, 3±19, 5±26
27 1027,3 0 9,2 1 14,4±1 7,6 3 6
28 1 0 28,2 0 14,4 0 7,4 2 8
29 1029,2 1 15,3±IIO,5±1 6
30 I a 30, 2 a 15, 3 a 10, 5 a 6
31 1 0 31,2 1 16, 4±1 8, 5±27
32 1 0 32,2 0 16,4 0 8,3 ±III, 4 2 9, 6 2 6
33 1 0 33,3 0 11,2 1 17,6 3 7
34 1 0 34, 2 0 17, 5 ± 17
35 1 0 35,5 0 7,2 1 18, 3±1 12, 4±1 9, 6 1 6
36 I a 36, 2 a 18, 3 a 12, 4 a 9, 6 a 6, 4 2 10, 5 ± 2 8
37 I a 37, 2 I 19
38 1 0 38,2 0 19, 3±1 13, 6±27
39 I 0 39, 3 a 13, 2 I 20, 4 ± I 10, 5 ± 1 8, 6 3 8
40 I a 40, 2 a 20, 4 a 10, 5 a 8, 4 2 II, 7 3 7
41 1 0 41,2 1 21, 3±114, 6±17, 5±29
42 1 0 42, 2 a 21, 3 a 14, 6 a 7
43 I 0 43, 2 I 22, 4 ± I I 1
44 1 0 44,2 0 22,4 0 11, 3±115, 5±19, 4 2 12, 6±28
45 1045,3015,509,2123,727,639
46 1 a 46, 2 a 23, 5 ± 2 10
47 1 0 47,2 1 24, 3±1 16, 4±1 12, 6±18, 7±3 8
48 I a 48, 2 a 24, 3 a 16, 4 a 12, 6 a 8, 7 I 7, 4 2 13, 8 4 8
49 1 0 49, 7 0 7, 2 1 25, 5 ±IIO
50 1 a 50, 2 a 25, 5 a 10, 3 ± I 17, 6 ± 2 9
On the Classification of Integral Quadratic Forms 361

are all represented in the table by a single entry

... abcde f g h ....

In restoring the signs it is helpful to note that the lower digits alternate in
sign. Some care is needed in recovering the original cycles, because some
of the four cycles just mentioned may coincide, and so an entry in the table
may represent one, two or four cycles. We have used ordinary parentheses
( ) and curly brackets { } to further reduce the size of the table.
Entries bounded by parentheses indicate whole or half periods. Thus for
d = - 99 the entry

represents the four distinct cycles

... 58_7693_10758_7 .. .

... -5 8 7 6 -9 3 10 7 -5 8 7 .. .
... 7 8 -5 710 3 -9 6 7 8 -5 .. .
... -7 8 5 7 -10 3 9 6 -7 8 5 .. .

However, the entry

for d = -37 yields only two inequivalent cycles, namely

... 35_4374_3543_7435_4 ...

and its reversal. Most entries in the table are contained within curly
brackets, which indicate reflections about the outermost digits. Thus for
d = -13 the entry

represents the single cycle

... 13_4132-3143_1341-3231-4113_4 ...

while for d - -14,

represents the two cycles

... 13_5222-5313_5 ...

and
... -1 3 5 2 -2 25 3 -1 3 5 ....
 Table I5.2a. Reduced indefinite binary forms.

d Forms

-1 OI}. OIl}
-2 {II}
-3 {112}
-4 0 2 I. 0 2 1 I. 0 2 2 )
-5 { 1 2}. {2I}
-6 { 1 2 2}
-7 {I 2 3 1 2}
-8 { 1 24 }. {2 2 }
-9 0 3 l. 0 3 ! l. 0 3 2 I. 0 3 3 )
-10 {I 3 }. {223 I }
-11 { 1 3 2}
-12 {I 3 31. {22 4 }
-13 (I3 4 1 3 2 1. {2 3 }
-14 (I 3 5 2 2)
-15 {I 3 61. {2 3 3}
-16 0 4 1. 0 4 II. 0 4 21. 0 4 3 2 41. 0 4 4}
-17 {I 4 1. {2 3 4}
-18 (I 4 21. {3 3 }
-19 (I 4 3 2 5 3 2)
-20 { 1441. {2 41. {4 2}
-21 {I 4 5 1 4 3 31. {2 3 6}
- 22 { 1 46 2 3 4 2 }
-23 {I 4 7 3 2}
-24 {I 4 81. {2 4 41. {33 5 2 4}
-25 0 5 1. 0 5 11. 0 5 21. 05 3 4}. 0 5 4 3 1. 055}
- 26 { I 51. {2 4 51}
-27 (1 5 21. (3 3 6)
- 28 { I 5 3 44 I. (2 4 6 2 4 )
- 29 { 1 54 3 5 21. {2 5 }
- 30 { I 5 5 I. {2 4 7 3 3 }
-31 {1 5 6 1 5 4 3 5 2}
- 32 { I 5 7 24 I. {2 4 8 I. {4 4}
- 33 { I 58 3 3 I. {2 5 4 3 6 }
- 34 { 1 5 9 42 I. {5 3 4 6 2 53}
- 35 { I 5 10 I. (2 5 5 )
-36 06}. 0 6 11. 0 6 21. 0 6 31. 0 6 41. 0 6 5 4 41. 06 6}
- 37 { I 61. {2 5 6 1 I. (3 5 4 3 74)
-38 {1 6 2}
- 39 { I 6 3 I. {2 5 7 2 5 3 6 }
-40 { I 6 4 }. {2 61. {5 5 3 4 8 I. {4 4 6 2 }
-41 { I 6 5 4 I. {2 5 8 3 4 5 }
-42 { I 6 6 }. {2 6 3 }
-43 { I 6 7 I 6 5 3 4 9 52}
-44 {1682537441.{264}
-45 {1 6 9 3 4 5 51. {2 5 101. {3 6 }. {6 3 }
-46 { 1 6 10 4 3 5 7 2 6 4 5 6 2 }
-47 { I 6 II 52}
-48 {I 6 121. {2 6 61.13 6 41. {4 4 8}
-49 0 7 I. 0 7 I }. 0 7 2 I. 0 7 3 5 8 3 5 7 O. 0 7 4 5 6 7 0. 0 7 7 }
-50 { I 7 I. {2 6 7 I I. {55}
 Table 15.2b. Reduced indefinite binary forms.
d Forms
-51 11 7 2},13 6 5 4 7 3 6}
-52 I 1 7 3 5 9 4 4}, 1 2 6 8 2 6 4 J. 14 6 }
-53 I I 7 4 5 7 2 J. 127}
-54 I I 75 3 9 6 2 J. I 3 6 6 }
-55 I I 7 6 5 5 }, 12 7 3 5 10 }
-56 I I 77 J. I 2 6 10 4 4 J. I 46 5 4 8 }
-57 I I 78 I 7 6 3 J. I 2 7 4 5 8 3 6 }
-58 I I 7 9 2 6 4 7 3 J. I 2 6 II 5 3 7}
-59 I I 7 10 3 5 7 2 }
-60 I 1 7 11 4 4}, 12 6 12}, 13 6 8 2 7 5 5 J. 14 6 6}
-61 { I 7 12 5 3 7 4 5 9 4 5 6 J. I 2 7 6 5 }
-62 11 7 13 6 2}
-63 I 1 7 14l, 12 7 7}, 13 6 9 3 6}
-64 0 8 J. 0 8 I }, 0 8 2 J. 0 8 3 7 5 8 0, 0 8 4 J. 0 8 6 4 8 J. 0 8 7 6 4 J. 088}
-65 II 8 J. 12 7 8 I }, 14 5 10 J. I 4 7 J. I 5 5 8 3 7 4 }
-66 I I 8 2 }, I 3 6 10 4 5 6 6 }
-67 I I 8376 572972 }
-68 I 18 4 J. 12 8 }, 14 6 8 2 }
-69 I 1 8 5 7 4 5 11 6 3J. 12 7 10 3 6}
-70 I I 8 6 4 9 5 5 J. I 2 8 3 7 7 }
-71 11 8 7 6 5 4 11 7 2}
-72 I 1 8 8J. 1 2 8 4 J. 13 6 12}, 14 6 9 3 7 4 8}, 1 6 6}
-73 11 8 9 1 8 7 3 8 J. 127125476583}
-74 11 8 10 2 7 5 J. 1 2 8 5 7}
-75 I 1 8 11 3 6}, 12 7 13 6 3J. 15 5 1O}
-76 I I 8 12 4 5 6 8 29 7 3 8 4 J. I 28 6 4 10 6 4 }
-77 I I 8 13 5 4 7 7 J. 12 7 14}
-78 11 8 14 6 3J. {2 8 7 6 6)
-79 I I 8 15 7 2 J. (3 8 5 7 6 5 9 4 7 3 10 7 )
-80 I 1 8 16J. 12 8 8), 14 6 11 5 5}, 14 8 J. 18 4 }
-81 0 9 J. 0 9 I J. 0 9 2 J. 0 9 3 }, 0 9 4 7 8 9 0, 0 9 5 6 9 3 8 5 7 9 0, 0 9 6 J. 0 9 9 }
-82 I 1 9 }, 1 2 8 9 1 J. (3711 4 6 8 )
-83 I 1 9 2}
-84 11 9 3J. 12 8 10 2 8 6 }, 14 6 12}, 14 8 7 5 }
-85 I 1 9 4 7 9 2 J. 12 9 }, 15 5 12 7 3 8 7 6 J. (16 5 1O)
-86 I I 9 5 6 10 4 7 3 II 8 2 }
-87 I I 9 6 }, I 2 9 3 }
-88 I 19 7 5 9 4 8J. 12 8 12 4 6 8 4}, 14 6 13 7 3 8 8J. 14 8 6}
-89 I I 9 8 7 5 8 J. I 2 9 4 7 10 3 8 5 }
-90 I 1 9 9J. 13 9 J. 12 8 13 5 5J. {6 9 9 3 }
-91 11 9 10 1 9 8 3 7 14J. 12 9 5 6 11 5 6 7 7}
-92 11911286784J. 12 8 14 6 4}
-93 11912374117493J. 12 9 6}
-94 119134685792108371582}
-95 I 1 9 14 5 5J. 12 9 7 5 1O}
-96 11 9 15 6 4}, 13 9 5 6 12J. 12 8 16}, 14 8 8}, 16 6 10 4 8}
-97 I 19167381138594}, {29876512749}
-98 '{ I 9 I7 8 2},I 77}
-99 I I 9 18), I 2 9 9 J. I 3 9 6 }, (5 8 7 6 9 3 10 7 )
-100 OIOJ. OlOl), 0 10 2}, 010381247100, 0104J. 0105J. 0106 8 },
0108 6 J. 01098 4J. 0101O}
364 Chapter 15

When -d is a perfect square, the cycles become chains terminated by O's.

Thus for d = -16,

represents

and

Numerous other tables exist in the literature (see [Bral], [Cas3,

p.3571, [Edwl, p.333], UncI], [JonI], [Legl], [Lehl, pp.68-72]'
[Som2]), but most of these omit some classes of forms (for example the
imprimitive forms, or those with -d a square, or indefinite forms).
3.4 Composition of binary forms. For binary forms, under suitable
restrictions, there is a notion of composition found by Gauss, that gives the
forms a group structure ([Gaul, §234], [Cas3, Chap. 4], [Dic2, Chap. 3],
[Edwl, §8.6], [Jon3, §4.4]). There is no generalization to dimensions ~ 3.
Composition is best understood in terms of the multiplication of ideal
classes in the corresponding quadratic number rings.
For simplicity we shall suppose that -d is a square-free number not
congruent to I (mod 4), since then the set Z[.J=d] of algebraic integers in
Q (.J=d) is precisely the set of numbers of the form r +s.J=d for
r, s E Z. It is also natural to restrict attention to properly primitive
quadratic forms j(x, y) = ax 2 + 2bxy + cy2 (those for which the
numbers a, 2b, c have no common factor) of determinant ac - b 2 = d.
Any ideal § in Z [.J=d] has a two-member basis (over Z), so that
§ = < r + s.J=d, t + u.J=d > ,
and its norm is usually taken to be the positive integer Iru -sf I [Reil, pp.
293, 3301. A principal ideal has a single generator r + s.J=d (say) over
Z [.J=d], while over Z it is generated by r + s.J=d and
.J=d (r + s.J=d). This ideal has norm r2 + s2d, and will be denoted by

< r + s.J=d > .

To get the proper correspondence between Gauss's group of forms under
composition and the group of ideal classes we must introduce the notion of
a normed or oriented ideal, an ideal § of norm N corresponding to two
normed ideals §N and §-N' A principal normed ideal is
< r + s.J=d > r2+s2d '

rather than
On the Classification of Integral Quadratic Forms 365

Multiplication of normed ideals is defined by

§N·;M = §;NM . (13)

Two (normed) ideals § and; are said to be in the same (normed) ideal
class if and only if there exist principal (normed) ideals ~ and !!l for
which §~ = ;!!l, and under composition these ideal classes form a group,
the (normed) ideal class group.
The proper (i.e. determinant + 1) equivalence classes of properly
primitive forms of determinant d correspond to classes of normed ideals as
follows. (We remind the reader that we are supposing -d to be square-
free and not congruent to 1 mod 4). One solution of the equation
f(x, y) = ax 2+2bxy +cy2 = 0 is

-=
x -b+r-d
y a

and we shall say that the normed ideal

<a, -b+r-d >a

corresponds to f.
Composition may now be defined as follows. To compose two quadratic
forms we first pass to the corresponding normed ideals, multiply them
using (13), and then, working modulo principal normed ideals, convert the
product to a normed ideal of shape
<a, -b+r-d >a ,

and reduce the corresponding quadratic form (b ~) of determinant d by

the cycle method of Theorem 1.

For example, the quadratic form (~ _~), which belongs to the

reduced cycle
2 2
3 -1 3 -2 3
corresponds to the normed ideal
< 3, -2+.J7 >3.
The square of this is the normed ideal of norm 9 whose ideal part is
generated by
32 = 9, 3(-2+.J7) = -6+3.J7, (-2+.J7)2 = 11-4.J7,

and this is easily seen to be the same as the ideal <9, -5+.J7>9
corresponding to the form (~ ~), which gives the reduction sequence
5 1 2 2 1 1
9 2 -3 -3 2 -3···
366 Chapter 15

Thus the composition of the equivalence class containing [~ _~) with

itself yields the class containing (i _13) . In fact every form of
determinant -7 is in one of these two classes, and so the group is cyclic of
order 2.
It is perhaps worthwhile to mention that the forms improperly
equivalent to f belong to the inverse class in the group (see [Cas3, Chap..
14, Theorem 2.1]). Thus the inverse class is represented by either (Z
~)
or [-'b -:). A form improperly equivalent to itself is usually called an
ambiguous form. Edwards [Edwl, Chaps. 7, 8] gives a clear account of
the group structure in the general case, with many examples. He works,
however, with an axiomatically defined notion of divisor. in place of our
normed ideals.
3.5 Genera and spinor genera for binary forms. The theory of genera (§7)
was originally developed by Gauss only for the binary case, where it
presents special features and is intimately connected with the group of
forms under composition. In fact two forms are in the same genus if and
only if their quotient is a square in this group. The genus of a binary
quadratic form is usually indicated by certain "characters" which can be
computed from the numbers represented by the form. We shall not
describe the easy conversion from this notation into the one used in §7.
The theory of spinor genera described in §9 assumes throughout that
the dimension is at least three, and in the binary case there are several
differences. Our treatment is no longer appropriate since there do not exist
spinor operators corresponding to all sequences (r _ h r2, r3, .. J of p -adic
unit square classes. Estes and Pall [Est!] have given a full investigation of
spinor genera in the binary case and have shown in particular that two
forms are in the same spinor genus if and only if their quotient is a fourth
power in the group.

4. The p-adic numbers

We now return to the general case of a quadratic form of arbitrary
dimension. The notion of integral equivalence (defined in §2) is a very
subtle one, which is best approached by studying the weaker notions of
equivalence over the larger rings
Q, the rational numbers,
Qp, the p-adic rational numbers, and
Zp, the p-adic integral numbers,
where p ranges over the "primes" -1, 2, 3, 5, .... The multiplicative
group of nonzero rational numbers is a direct product of its cyclic
subgroups generated by -1 and the positive prime numbers 2, 3, 5, .... In
this chapter we shall refer to -1 as a prime number, although it presents
several special features which arise from the fact that the subgroup it
generates is cyclic of order 2 rather than of infinite order. By convention
On the Classification of Integral Quadratic Forms 367

Q-l and Z-l are both equal to the ring R of real numbers. The number
theory of Q or Z involves an infinity of prime numbers; by passing to Qp or
Zp we concentrate on just one prime at a time.
4.1 The p-adic numbers. We give a brief description of the main
properties of the p-adic numbers that will be needed later. Further
information is readily available in a number of books [Bacl1, [Bor5],
[Cas3], [Kobl], [Mah4], [O'Men
Any real number is the limit of a Cauchy sequence of rationals in the
ordinary metric

If p is any positive prime, we can also equip the rationals with the p -adic
metric

where pk is the exact power of p in the factorization of x-y. For instance

1 1
d 5 (2, 52) = 52 = 25 '

1 1
d 3 (27'0) - r3 .. 27.
We note that for x ;z! y the product of dp (x ,y) over all primes p
(including -1) is + l.
The p -adic rational numbers Qp are the limit points of Cauchy
sequences of ordinary rational numbers with respect to the p -adic metric.
The p -adic integers (or rational p -adic integers) Zp are obtained by
requiring the terms in the sequence to be ordinary integers.
For example, the difference between the nth and any later term in the
sequence 4, 34, 334, 3334, ... is divisible by 5n (in fact by IOn), and so
this sequence is 5-adically convergent to a 5-adic integer a say. In this
case a is actually a rational number. For 3a is the limit of the sequence
12, 102, 1002, 10002, ... which 5-adically converges to 2, since its nth
term differs from 2 by a multiple of 5n • Thus a = 2/3.
In a similar way we see that any rational number whose denominator is
not divisible by p is a p-adic integer.
4.2 p-adic square classes. As another example, we see that 6 has a square
root among the 5-adic integers, since 12 == 6 (mod 5) ,
(-9)2 == 6 (mod 25) , (16)2 == 6 (mod 125) , (-109)2 == 6 (mod 625) ,
... , the sequence 1, -9, 16, -109 ,... being capable of being continued
indefinitely in such a way that the n-th term is congruent to all later terms
modulo 5n • It is therefore a 5-adic Cauchy sequence whose limit is a
square root of 6. Similarly, for any positive odd prime p, an integer not
divisible by p that is a quadratic residue modulo p is a p-adic square
[Bac1, p. 59].
368 Chapter 15

Two p -adic rational numbers are said to be in the same p -adic square
class if their ratio is a p-adic rational square. The square classes for the
various values of p may be described as follows (cf. [Bac1, pp. 59-60],
[Cas3, p. 40], [Wat3, p. 33]):
+ u, - u for p = -1 ,

U" U3, us, U7, 2u" 2U3, 2us, 2U7 for p = 2,

U+, U_, pu+, pu_ for p ~ 3,

where
for p = -1, U is any positive real number (a (- I}-adic unit),
for p == 2, Ui represents any 2-adic unit congruent to i (mod 8),
for p ~ 3, U+ (resp. u_) is any p-adic unit that is a quadratic residue
(nonresidue) mod p .
4.3 An extended Jacobi-Legendre symbol. In agreement with our policy of
regarding - 1 as a prime, it is natural to define the greatest common
divisor of two integers

with a = 0 or 1, a -= 0 or 1, and b, {3, c, /" ... = 0,1,2, ... to be

(n, v) == (_l)min (a, a) 2 min (b, (:J) 3min (c, 1') ... .

Note that this implies that two negative numbers cannot have (n, v) = 1
and so will not be counted as coprime. We can define the Jacobi-Legendre
symbol [~] in all cases where (n, v) = 1 as follows:

± 1 for a prime p ~ 3, according as v is or is not

= congruent to a square mod p .

== I (since in this case v > 0) , and

= 1 if v == ± 1 (mod 8) , -1 if v == ± 3 (mod 8) .

(Note that [; ] is equally well defined if v is a p-adic integer not divisible

by p, since every p -adic integer is congruent modulo p to a rational
[~ ] for n given by (14) is then defined to be

r[; r[; r
integer.) The symbol

[:1
On the Classification of Integral Quadratic Forms 369

With this definition the law of quadratic reciprocity asserts that

whenever either symbol is defined, unless n and v are both congruent to - 1

(mod 4), when

4.4 Diagonalization of quadratic forms. It is well-known that any

quadratic form over a field of characteristic ¢ 2 (for example Q or Qp)
may be diagonalized, and in fact any vector at which the form takes a
nonzero value may be taken as the first term of a diagonal basis (see for
example [Jon3, Theorem 2]).
Furthermore for p ¢ 2 any form can be diagonalized over Zp. For
p = - 1 this is covered by the previous assertion. Otherwise we proceed as
follows. We first find a matrix entry that is divisible by the lowest power
of p. If this is a diagonal entry, say all, then we can start the
diagonalization by subtracting multiples of the first row from the others so
as to clear the rest of the first column, following this by the corresponding
column operations to clear the rest of the first row. On the other hand if a
nondiagonal entry, say a \2, is divisible by the least power of p, and all
diagonal entries are divisible by a higher power, we can reduce to the first
case by adding the second row to the first and then the second column to
the first. This replaces all by all+2aI2+a22, which, since p ¢ 2, is now
divisible by the lowest power of p to occur in any entry.
The same method works if p = 2 unless we arrive at a stage when some
off-diagonal entry a 12 say is divisible by the least possible power q = 2k,
while all diagonal entries are divisible by 2k+1. In this case the leading
2x2 submatrix has the form

[qa qb]
qb qc '

where a and c are divisible by 2 but b is not, so that d = ac-b 2 is not

divisible by 2. This implies that any pair of integers (x, y) is a 2-adically
integral linear combination of (a, b) and (b, C), so that (since all entries
are divisible by q) we can subtract suitable multiples of the first two rows
from the others (followed by the corresponding column operations), so as
qa qb]
to remove [qb qc as a direct summand. Thus we have proved the
following result.

Theorem 2. For p ¢ 2 any p -adically integral form can be diagonalized

by a p -adically integral transformation. For p = 2 there is a p -adically
370 Chapter 15

integral transformation expressing the form as a direct sum of forms with

matrices

(qx) , [qa qb]

qb qc '

where q is a power of 2, a and c are divisible by 2, but x, band

d = ac -b 2 are not.
Note. The reader who consults other works on quadratic forms will notice
that what we call the prime -1 is usually given the name 00. Over more
general algebraic number rings there will be several such "primes",
corresponding to different archimedean valuations, and in Hasse's original
publications they were given symbols such as 1', 1", .... The most
appropriate name for them is "unit primes", since they arise from
properties of the group of units of the underlying ring. Unfortunately the
pernicious habit has grown up of calling them "infinite primes" instead.
When we started to write this chapter we hesitated between the notations
00 and - 1 for the archimedean prime in our case, but eventually found

that the unconventional name -1 made things so much more simple that
its omission would be indefensible.

5. Rational invariants of quadratic forms

In §§5 and 6 we shall investigate when two integral quadratic forms are
equivalent over the rational numbers; the main results are stated in
Theorems 3, 4 and 5. Rational invariants for quadratic forms are usually
defined via the Hilbert norm residue symbol ([Cas3, Chap. 61, [Jon3,
Chap. 31, [Wat3, Chap. 3]). Our treatment avoids the use of this symbol
and furthermore transforms the standard "product formula" into the
readily usable sum formula (I5) or "oddity formula" (I 6). Since we are
working over Q we may assume that the form has already been
diagonalized (cf. §4.4).
5.1 The invariants and the oddity formula. Any rational or p-adic integer
A can be written uniquely in the form A = paa where a is prime to p
(meaning that a is positive if p = -1, cf. §4.3). Then p (A) = po. is called
the p-part of A, and p'(A) = a is the p'-part. We shall introduce the
term p -adic antisquare to mean a number of the form podd· u _ when
p ~ 3, and 2odd 'u±3 for p = 2, since both the p- and p'-parts of such a
number are non-squares. (There are no (- I}-adic antisquares.) In terms
of our extended Jacobi-Legendre symbol, paa is a p-adic antisquare if and
only if

po. is not a square and [; 1= - 1.

The p-signature of the integral quadratic form

f = diag {paa, p(3b, p'Yc , ... } is defined to be
On the Classification of Integral Quadratic Forms 371

pa + pfJ + p'Y + ... + 4m (p ¢ 2) ,

a + b + c + ... + 4m (p = 2) ,

where m is the number of p-adic antisquares among paa, pfJb, p'Yc , ....
Thus the (-I)-signature of f is just its ordinary signature, which (by
Sylvester's law of inertia, §6.2) is an invariant for real equivalence. For
p ~ 2 the p -signature is only to be regarded as defined modulo 8, and we
shall see that the p-signature is an invariant for rational equivalence. The
2-signature, which often behaves specially, is also called the oddity of f.
What is usually termed the product formula relating the different p-
adic invariants [Cas3, p. 76], [Jon3, Th. 29] becomes in this notation the
sum formula
2-signature - dimension == ~ p -signature - dimension (mod 8) (15)
oddp

or
~ p -excess (J) == 0 (mod 8) ,
all p

where we define the p-excess to be

p -signature - dimension, (p ¢ 2) ,
dimension - p -signature, (p = 2) .

For practical calculations it is better to treat the contributions from

p = -1 and 2 separately, leading to the oddity formula:
signature (J) + ~ p -excess (J) == oddity (J) (mod 8). (16)
p~3

Example. For the form f = diag {I, 3, - 3} we compute:

for p = -1, the signature = 1+1-1 = 1,
for p = 3, the 3-excess = 0+2+2+4 = 8

(since - 3 is a 3-adic antisquare), and

for p ~ 5, the p-excess = 0+0+0 = 0,

while

for p = 2, the oddity = 1+3-3 = 1,

so that (16) reads

1+8+0+0+ . .. == 1 (mod 8) .

Then our first main theorem, the proof of which will be given in §6, is
the following.
372 Chapter 15

Theorem 3. Two nonsingular forms of the same dimension are equivalent

over the rationals if and only if
(j) the quotient of their determinants is a rational square, and
(ij) for each p they have the same p -excess.
Condition (ii) may be replaced by the equivalent condition:
(ij)' they have the same signature, the same oddity, and, for all p ~ 3,
the same p -excesses modulo 8.
It is obvious that if two forms are equivalent over the rationals then
they must be equivalent over Qp for all p. Since on the other hand the
invariants used in Theorem 3 are p -adic invariants, this theorem can be
restated as follows.

Theorem 4 (The weak Hasse principle). A necessary and sufficient

condition for two rational forms to be equivalent over Q is that they be
equivalent over Qp for all p.
5.2 Existence of rational forms with prescribed invariants. The next
theorem states that the oddity formula is essentially the only relation
between the p -adic invariants for all p.

Theorem 5 (The strong Hasse principle). If for each p we are given a

p -adic form f(P) of determinant d, satisfying
signature <.r(- 0) + ~ p -excess <.r(P» == oddity <.r(2» (mod 8) ,
p~3

then there exists a rational form f which is equivalent to f(p) over Qp for
each p.
By Theorem 4, if such a form exists it is unique up to equivalence.
Sketch of proof (For further details see for example [Cas3, Chap. 6,
Theorem 1.3] or [Jon3, Theorem 291.) The theorem is first reduced to the
case when f is a binary form. Then we wish to find a rational form
f = diag {A, B} of prescribed determinant d and with given signature and
p -excesses for all p ~ 2. The idea is to choose a large prime q and to
take f - ± diag {q, qd}, the sign being determined by the signature. The
value of the p -excess for each prime p dividing d is controlled by the
residue class of q modulo p, or if p = 2 by the residue class of q modulo
8. All these values may therefore be simultaneously adjusted by requiring
q to be in a suitable arithmetic progression modulo 4d. The existence of
such primes q is guaranteed by Dirichlet's theorem (1837) on primes in
arithmetic progressions (see for example [Apol]). For primes p ¢ q that
do not divide d, the p -excess is trivial, while for q itself the q -excess must
be correct by the oddity formula.
Remarks. (1) This proof essentially dates back to Legendre, at a time
when Dirichlet's theorem was an unproved conjecture. Gauss later
eliminated the dependence on that conjecture by finding a proof using the
genera of integral binary quadratic forms (see [Cas3, Chap. 14, §5).
(2) The possible values of a given p-adic invariant for an n-dimensional
form can be computed from its possible values for I-dimensional forms,
On the Classification of Integral Quadratic Forms 373

which are easily listed. For instance, p-excesses are always even, and are
divisible by 4 if P == 1 (mod 4).

5.3 The conventional form of the Hasse-Minkowski invariant. In the

literature on quadratic forms the subtle part of the p-adic invariant for a
form is usually expressed as a number equal to ± 1, called the Hasse-
Minkowski invariant (rather than our p-excess). There are several
different conventions, but a common one [Cas3, p. 55] is that the Hasse-
Minkowski invariant for f -= diag (AJ, A 2, ••• , An} is
(J)p = (AJ, A 2,· .. , An)p = II (AI, A)p (17)
\"I<j~n

where (x, y ) p is the so-called Hilbert norm residue symbol. This invariant
can be recovered from the p -excess as follows. (J) p is equal to
or -1

according as
p -excess (J) - p -excess (Jo)

is congruent to
o or 4 (mod 8) ,

where

is the "standard" form having the same determinant and dimension as f.

6. The invariance and completeness of the rational invariants

This section is devoted to proving Theorem 3.

6.1 The p-adic invariants for binary forms. The equivalence class of a
binary form over a field is completely determined by its determinant d
(modulo squares) and by any nonzero number a that it represents. For if
f(e\) = a, we can take e\ to be the first vector of a diagonal basis, with
respect to which we must have f - diag (a, d fa).
For the p-adic rationals there are only finitely many square classes, and
we can enumerate all possible forms (see Table 15.3), and thereby check
that two forms have the same determinant and p-adic invariants
(signature, p-excess, and oddity) if and only if they are equivalent. As an
example we consider the 2-adic forms of determinant 2U3' Each such form
is equivalent to one of

for which the oddities are respectively

1+3+4 == 0, 3+1 == 4, 5+7 == 4, 7+5+4 == 0
374 Chapter 15

Table 15.3. Correspondence between the p-adic invariants and the

numbers represented by binary forms.

p Det Square classes represented Signature

-1 +u +u 2
-u -2

-u +u, -u 0
p Det Square classes represented Oddity
2 UI UI. us, 2UI. 2us 2
U3, U7, 2U3, 2U7 6
U3 UI. U3, us, U7 4
2UI. 2U3, 2us, 2U7 0

Us UI. us, 2U3, 2U7 6

U3, U7, 2UI. 2us 2

U7 all 0

2uI UI. u3, 2UI. 2U3 2

us, U7, 2us, 2U7 6
2U3 UI. U7, 2U3, 2us 0
U3, us, 2UI. 2U7 4

2us UI. U3, 2us, 2U7 2

us, U7, 2UI. 2U3 6
2U7 UI. U7, 2UI. 2U7 0
U3, us, 2U3, 2us 4

p Det Square classes p -excess, for p (mod 8) -

represented 1 3 5 7

~3 U+ all 0 - 0 -
U+, u_ - 0 - 0
pu+, pu_ - 4 - 4

u_ all - 0 - 0
U+, u_ 0 - 0 -
pU+,pu_ 4 - 4 -
pu+ u+,pu+ 0 2 4 6
U_,pu_ 4 6 0 2

pu_ U+, pu_ 4 6 0 2

U_, pu+ 0 2 4 6
On the Classification of Integral Quadratic Forms 375

(mod 8). Since x 2+6y2 (the first form) represents 7, it is equivalent to the
fourth form, and similarly the second and third forms are equivalent. But
an easy calculation reveals that x 2+6y2 does not represent any number of
the form 4m (8k+3), and so these two pairs of forms are not equivalent.
Thus there are two distinct 2-adic binary forms of determinant 2u 3. The
first has oddity 0 and represents the numbers u h U7, 2u 3 and 2u 5, while
the second has oddity 4 and represents the numbers U3, U5, 2Ul and 2U7.
This explains the entries for p - 2, det = 2U3 in Table 15.3.
After a similar discussion for all the cases with p - 2 in Table 15.3 we
conclude that the oddity is a 2-adic invariant, and together with the
determinant is a complete invariant for 2-adic rational equivalence.
Similarly for the other primes. This completes the proof of Theorem 3 for
binary forms.
The reader will notice that there is, for each p, one form that
represents all nonzero numbers. This is the isotropic form diag {A, -AJ,
of determinant -I, so called because it also represents zero nontrivially.
6.2 The p-adic invariants for n-ary forms. It is easy to extend the above
remarks to show that our invariants really are invariants for forms of all
dimensions. The proof reduces to showing that any equivalence between
diagonal forms can be broken down into a chain of binary equivalences
(i.e. ones effecting just two diagonal terms).
To see this, note that for
f = diag {a,b,c, ... J, f' = diag {a',b', c', .. J

to be equivalent, a' must be representable by f. We choose a

representation involving the smallest number of terms, say
a' = ax 2 + by2 + cz 2 + dt 2 ,
and then we have the binary equivalences
diag {a,b,c,d,e, ... } - diag {a2,b*,c,d,e, ... }
-diag {a3,b*,c*,d,e, ... } -diag {a4,b*,c*,d*,e, ... } ,

where a2 = ax 2 + by2, a3 = ax 2 + by2 + cz 2, a4 = '" are nonzero.

These show that f and f' are equivalent to diag {a',b*,c*, ... J, and so by
Witt's cancellation theorem (which we shall prove in a moment) the forms
diag {b, c , d, .. J and diag {b *, c *, d* ,... } are equivalent. Since by induction
the latter equivalence reduces to a chain of binary ones, we deduce the
same for the equivalence of f and f'.
The remainder of this section will be devoted to the proof that two
forms having the same p -adic invariants for all p (including -1) are
equivalent over the rationals.
Let f and g be two rational quadratic forms with the same nonzero
determinant (modulo a square factor), and the same signature, oddity and
376 Chapter 15

p-excess for all p ~ 3. Then we shall prove that for a suitable

nonsingular form h, f E9 h is equivalent to g E9 h and will deduce that f
is equivalent to g by repeated application of Witt's cancellation theorem.

Theorem 6 (Witt's cancellation theorem [Witl], [Cas3, p. 21], [SchO, p.

22], [Sch2, Chap. 1]). Over any field of characteristic ;Z! 2, if
diag {a, b, c, .. J - diag {a, b', c', ... } and a ;Z! 0, then diag {b, c, ... } -
diag {b', c', ... J.
Proof This is equivalent to the following geometrical assertion. Let V
be a vector space over the field, equipped with the bilinear form
f(x, y) = aXIYl + bX2Y2 + CX3Y3 + .... Then if vectors v and w in V
have the same nonzero norm a, there is an automorphism of V fixing f
and taking v to w. Now for any rEV of nonzero norm, the reflection
2 f(x, r) (18)
x - x - f(r, r) r

is an automorphism of V preserving f. Not both the vectors r = v ± w

can have zero norm, and so one of the corresponding reflections exists and
takes v to ± w, and can be followed by negation if necessary. This
establishes the theorem.
Remark. Sylvester's law of inertia (the invariance of signature under real
equivalence [Jon3, Theorem 2]) follows immediately from Theorem 6,
since if

over the reals, then we can deduce that

diag {(+ 1)k} - diag {(_1)k} ,

implying k = 0 (since one form is positive definite, the other negative

definite) .
We shall say that a form has trivial invariants if all its p-excesses are
congruent to zero modulo 8 and its determinant is a perfect square. Then
the desired result will follow from:

Theorem 7. If F has trivial invariants then F is equivalent over the

rationals to a form of the shape
diag {± 1, ± 1, ... , ± 1} .

To see that the result we want is a consequence of Theorem 7 we argue

as follows. Since p-excesses are always even, if f and g have the same
invariants then
(19)

will both have (the same) trivial invariants, and by Theorem 7 will be
equivalent to forms of the shape diag {± 1, ± 1, .. J. Since signature (J)
On the Classification of Integral Quadratic Forms 377

= signature (g), the numbers of positive and negative terms agree, and the
two forms (I9) are equivalent. By Witt cancellation (Theorem 6) we
deduce that f - g.

6.3 The proof of Theorem 7. We shall suppose throughout that F is a

diagonal form with square-free integer entries and with trivial invariants,
and will actually show that, for a sufficiently large N,
FEB diag {(+t)N, (-ON} - diag {±1, ±l, ...}

over the rationals. We shall do this by gradually reducing the primes

appearing in the entries of F. Let p be the largest such prime, and call
the entries of F divisible by p the p -terms. Note that any p -term has the
form PQIQ2 ... qk where -1 ~ Qi < P for all i. We suppose first that
p ~ 3.

Theorem 8. (The replacement lemma). Assume p ~ 3. If

-1 ~ a, b < p. and ab is congruent to a square (mod p). then we may
replace any p -term "pat" by "pbt" without introducing any prime larger
than p.
Note. The replacement process involves adjoining more direct summands
± 1 to F.
Proof (Based on [Con8, p. 4011.) We can write ab = x 2-py , with
Ix I < If2 p, and so Iy I < p. Then the identity
padb/p)2_pbt(x/p)2 = -ybt

shows that the form diag {pat, - pbt} represents - ybt, and therefore, by
the observation at the beginning of §6.1, is equivalent to diag {Yat, - ybt}.
Also x 2_y2 = diag {I, - I} represents all numbers, in particular - pbt, and
so
diag {I, -I} - diag {-pbt,pbd.

Then we have
diag {pat, I, - I} - diag {pat, - pbt , pbd - diag {Yat, - ybt , pbd ,

which is the desired replacement. This completes the proof of Theorem 8.

We now suppose p > 2. By repeated use of the replacement lemma we
can replace each p -term by puk, and so by p or pu, where u = r + I is
the least positive non-residue modulo p. But also diag{p ,pr} represents
pu, and so is equivalent to diag{pu ,pur}. We may therefore replace
p, P by p, pr, then pu, pur, then pu, pu , (20)

or vice versa, so that the first of two or more p-terms may be chosen
arbitrarily.
The determinant condition tells us there exist evenly many p-terms.
We may replace the first one by -p and the second by p or pu (and
378 Chapter 15

definitely by p if there are more than two p -terms). If the second is p we

can now eliminate the first two p-terms using
diag {-p,p} - diag {-I, l} .

If not, the only p-terms are -p,pu, and the p-excess differs by 4 from
that of diag{-p,p} (which is zero), and so is non-trivial, contradicting the
supposition.
So when p > 2 we have been able to reduce the size of p by repeated
application of the replacement lemma. If p = 2 all p-terms are ± 2, and
there are an even number of them since the determinant is a square. They
can then be eliminated using the equivalences diag {2, 2} - diag {I, I} ,
diag {2, -2} - diag {I, -I}, diag {-2, -2} - diag {-I, -I} . This
completes the proofs of Theorems 7 and 3.

7. The genus and its invariants

Two integral quadratic forms are said to be in the same genus if they are
equivalent over the p -adic integers for all primes p (including -I). As we
shall see in §9, for indefinite forms of dimension n ~ 3, there is usually
only one equivalence class of forms in a genus. In fact this holds whenever
Idet I I < 128, and when it fails 4[ nl21 det I must be divisible by k (2) for
some nonsquare natural number k == 0 or 1 (mod 4).
In this section we give a complete system of invariants for p -adic
integral equivalence for each p, and then show how to combine them to
obtain a handy characterization of the genus.
No proofs will be offered. For p ;z! 2 several accounts are readily
available (e.g. Cassels [Cas3]), and all cases are handled by O'Meara
[O'Mel], who gives the invariants for forms over arbitrary number fields.
The correctness of the simple system of invariants and transformation rules
for p - 2 given here was originally verified (by J.H.C'> by showing that
they suffice to put every form into B. W. Jones' canonical form [Jon21, yet
are consistent with G. Pall's complete system of invariants [Pa1l1
We remark that much of the importance of the genus - for instance in
topological investigations - arises from the fact that two forms In
and g
are in the same genus if and only if I (B [n] and g (B ln] are
integrally equivalent. This follows from properties of the spinor genus.
7.1 p-adic invariants. As we saw in Theorem 2, any form can be
decomposed over the p-adic integers as a direct sum
(22)

in which each Iq is a p-adic unit lorm. meaning a p-adic integral form

whose determinant is prime to p (if p ~ 2) or a positive definite form (if
p = - 1). The summands q{q in (22) are called Jordan constituents of I,
and (22) itself is a Jordan decomposition of I. We call q the scale of the
On the Classification of Integral Quadratic Forms 379

constituent qJq. Note that, when p = -1, (22) is the familiar result that
any form can be written as a sum of definite forms:
I = III E9 (-1)/-1,

where I I and I -I are positive definite.

For p ¢ 2 the set of values of q occurring in (22), together with the
dimensions nq = dim Iq and signs

~q [de~/q ] ,
=

form a complete set of invariants for I (see Theorem 9). The case p = 2
presents additional complexities and will be discussed in §§7 .3-7 .6. In the
case p = - 1, ~q = + 1 since det Iq is positive, and so n+ I and n_1 are the
only invariants (this is Sylvester's law of inertia, §6.2). By convention, + 1
and -1 will usually be abbreviated to + and -, and so we write n+ for
n+b n_ for n_I'
7.2 The p-adic symbol for a form. For p = -1 we express the fact that
n+ = a and n_ = b by the (- J) -adic symbol
+Q _b .

For other odd p we shall use a p -adic symbol which is a formal product of
the "factors"

For example if p = 3, the symbol

I- 2 3+ 5 9+ 1 2T 3 (23)

represents a form

with dim/l = 2, dim!J = 5, dim/9 = 1, dim 127 = 3, where the

determinants of 13, 19 are quadratic residues modulo 3 and those of I I, 127
are nonresidues.
In these symbols we may adopt certain obvious abbreviations, such as
replacing (23) by 1- 2 35 9 2T3.

Theorem 9. For p ¢ 2, two quadratic lorms I and g are equivalent

over the p -adic integers if and only if they have the same invariants nq , ~q
lor each power q 01 p, or equivalently if and only if they have the same
p -adic symbol.
The proofs of Theorems 9 and 10 are omitted (see the remarks at the
beginning of this section). Theorem 9 makes two assertions. First, q, ~q,
nq are a complete set of invariants for the Jordan constituents qJq, and
380 Chapter 15

second, the invariants of the Jordan constituents are invariants of f. Both

assertions must be modified when p = 2.
7.3 2-adic invariants. Let the 2-adic Jordan decomposition of f be
(24)

Then the invariants of qfq are the quantities

q, the scale of qfq,
Sq = lor II (see below), the type of fq, which is the scaled type of qfq,
nq = dim fq (the dimension of fq or qfq),
det f q
[ --2- ] (the sign of fq or qfq), and
~q = -

tq, the oddity of fq (see §5.I).

We define Sq to be I if qfq represents an odd multiple of q, and otherwise

II (compare §2.4 of Chap. 2). Equivalently, Sq is I if and only if there is
an odd entry on the main diagonal of the matrix representing fq, and
otherwise II. If fq (of type I) has been diagonalized, then tq is its trace,
read modulo 8. If fq has type II, tq = O.
7.4 The 2-adic symbol. The 2-adic symbol representing a given Jordan
decomposition (24) of f is a formal product of factors

where the former indicates a constituent qfq for which fq has type I and

det fq ] .
[ --2- .
= ~q' dim fq = nq , oddity (Jq) = tq ,

while the latter indicates a constituent qfq for which fq has type II and

[de~fq ] = ~q, dim fq = nq , oddity (Jq) = 0.

We shall sometimes write

(The value of t q is often unimportant).

For example 1- 22t 343" 18+ 4 (or liI 2 2t 3 43"1 8~4) represents a form
having a Jordan decomposition

III which f J, 12, f 4, f 8 have dimensions 2, 3, 1, 4 and determinants

congruent to ± 3, ± 1, ± 3, ± 1 (mod 8) respectively, f 1 and f 8 have
type II, while f 2 and f 4 are of type I and can be put into diagonal form
with traces congruent to 5 and 3 (mod 8) respectively.
On the Classification of Integral Quadratic Forms 381

7.5 Equivalences between Jordan decompositions. So far we have described

the invariants for a Jordan constituent qfq. Unfortunately a form may
have several essentially different Jordan decompositions, and so can have
several different 2-adic symbols. The precise rule giving all such
equivalences is as follows.

Theorem 10. Two forms f and l' with respective invariants

are 2-adically equivalent just if

(i) nq = n'q, Sq = S'q for all q, and
(ii) for each integer m (including negative integers) for which f 2m has type
II, we have
~ (tq-t'q) == 4(min(a,m)+min(b,m)+ ... ) (mod 8)
q <2 m

where 2a , 2b ,... are the values of q for which fq ¢ f'q.

Although Theorem 10 completely describes all the 2-adic equivalences,
it is often simpler to use the following ideas.

Compartments and trains. Suppose f has a Jordan decomposition (24).

By an interval of forms we mean all the forms qfq, even those of zero
dimension, for which ql ~ q ~ q2, where qlo q2 are powers of 2. A
compartment is a maximal interval in which all forms are of scaled type I,
and a train is a maximal interval having the property that for each pair of
adjacent forms at least one is of scaled type I. Thus in
1+2[262453]8+0[16tl132+2: 64- 2: 128- 4 [2563"1]512+ 0 (25)

the square brackets enclose the compartments and the trains are separated
by colons. Notice that here one train has two compartments, while another
train has none.
There are two ways in which such symbols may be altered and yet still
represent 2-adically equivalent forms.

(i) Oddity fusion. Two 2-adic symbols represent the same form if the only
change is that the oddities have been altered in a way that does not affect
their total sum over any compartment.
So we may replace the individual oddity markers in a compartment by
their total (modulo 8), written as a subscript to the entire compartment.
For example we may replace [26 2 45 3] in (25) by [r 2 4+ 3h.

(iO Sign walking. A form is unaltered if the signs fq of any two terms in a
train are simultaneously changed, provided certain oddities are altered by
4. Let fql' fq1! ql < q2, be the signs we wish to change. We imagine
walking along the train from the term for ql to that for q2' Our walk
consists of a number of steps between adjacent forms fq and f2q, and each
382 Chapter 15

such step involves just one compartment, since at least one of fq, f 2q is of
type I, and if they are both of type I they are in the same compartment.
Then the rule is that the total oddity of a compartment must be changed
by 4 modulo 8, precisely when the number of steps that involve that
compartment is odd.
Suppose for example we wish to change the signs corresponding to f 2
and f 16 in the train
(26)

The walk has three steps, from f 2 to f 4, f 4 to f 8, and f 8 to f 16. The first
two steps affect the first compartment and the third step affects the second
compartment. The resulting symbol is therefore
12 [2 2 43 h 8°[16- 1]5 32 2 (27)

Alternatively, a walk of just one step from f 2 to f 4 in (26) would lead to

12[22 4- 3 h 8° [161]1322. (28)

All of (26), (27), (28) represent equivalent forms.

The effect of the three-step walk could also be achieved by three
separate one-step walks, except that at an intermediate stage the symbol
would contain a factor 8-°, corresponding to an impossible Jordan
constituent. Transformations involving such impossible constituents are
quite legal provided the end results are meaningful.
7.6 A canonical 2-adic symbol. Using these rules we can arrange that
there is at most one minus sign per train, which can be attached to any
form of nonzero dimension. One convenient rule is to put this sign on the
earliest nonzero dimensional form of a train. If this convention is adopted
and only the total oddities of the compartments are given, the resulting
symbol is absolutely unique and may be taken as a canonical symbol for
the form. Thus for (25) the canonical symbol is
1- 2[2+ 24+ 3 h 8+°[16+1]132+2: 64- 2: 128+4[256+ 1h 512+°,

which would be further abbreviated in practice to

1- 2[2 2 43h [16]1 322: 64- 2: 128 4 [256h .

Then two forms are 2-adically equivalent if and only if their canonical
symbols are identical.
7.7 Existence of forms with prescribed invariants. It is important to
specify exactly which conceivable systems of invariants actually correspond
to quadratic forms. There are three sets of conditions.

The determinant conditions for each p. The p -adic square classes of the
determinant as computed from the p-adic symbols must agree with its
known value. In other words
On the Classification of Integral Quadratic Forms 383

the product of all the signs ~q in the p -adic symbol must be [;] (29)

where det (j) = paa and (a, p) == 1.

The oddity condition, relating aU p. From the p -adic symbols we can

compute the invariants appearing in the oddity formula. Thus for p - -I,
signature (j) ... r-s ,

if the (-1) -adic symbol is +' - s; and for p ~ 3,

p -excess (j) == 2; nq (q-1) + 4kp (mod 8) ,
q

where the nq are the dimensions of the Jordan constituents and kp is the
number of antisquare terms (i.e. q not a square and ~q - - 1) in the p-
adic symbol; and for p = 2,
oddity (j) == 2; tq + 4k2 (mod 8) ,
q

from the 2-adic symbol. Then these quantities must be related by the
oddity formula
signature (j) + 2; p -excess (j) == oddity (j) (mod 8). (30)
p~3

The existence condition for each Jordan constituent. Each term in the p-
adic symbol must correspond to an existing form. If p ¢ 2, for each
Jordan constituent qfq of dimension n and sign ~ we must have
if n == 0 or p = - 1 then ~ == +. (31)

For p - 2 the following must hold:

for n = 0, type = II and ~ == +, (32)

{ ~~ ==== +_ .. t == ± 1 (mod 8) ,
for n = I, . . t == ± 3 (mod 8) , (33)

for n - 2 {~ == + .. t == 0 or ± 2 (mod 8) ,
and type I ~ = - . . t == 4 or ± 2 (mod 8) ,

while for general n we have t == n (mod 2), and

t == 0 (mod 8) for type II, so that n odd" type I. (35)

Theorem 11. If a system of putative p -adic symbols for each p satisfies

the determinant, oddity and p -adic existence conditions (29)-(35), then
there exists an integral quadratic form with these p -adic symbols.
384 Chapter 15

When working with the abbreviated form of a 2-adic symbol, it is

helpful to know that the only existence condition on a compartment
containing two or more Jordan constituents is that its total oddity must
have the same parity as its total dimension.
7.8 A symbol for the genus. We can combine the significant portions of
our p-adic symbols to give a handy notation for the entire genus (that
happily generalizes some notation we have used elsewhere [ConI3],
[Con33], [Con34]). This symbol has the form
Ir , .(- .. ) or II r , J .. ) ,
where the Roman numeral is the type of the entire form, i.e. the type of its
2-adic Jordan constituent f I; the subscripts indicate the (- 1) -adic symbol
+r - s; and the parenthesis contains the usual symbols

for the powers q > 1 of all primes 2, 3, .... The subscripts t, I or II may
be omitted when their values can be deduced from the oddness of m or the
oddity formula (30).
The symbols

corresponding to the constituents f 1 in each p -adic Jordan decomposition

have been omitted. However
the sign ± can be recovered from det f,
the number m can be recovered from dim f = r+s,
the type I or II (for p = 2) is displayed, and
the oddity t (when relevant) can be computed from the oddity
formula (30).
For example I r ,s(2) has determinant (-1) s 2, and so its p-excess is 0 for
p ~ 3. From (30) the oddity is r-s, and therefore the 2-adic symbol is
II±(r+s-1l 2l = [I+(r+S-1l 2 I ]r_.. ,

using (29). Similarly II r , s (3) (whose determinant is (-I)S 3 = ± 3) has

3-adic symbol 1± (r+s-\)3 1 (with the same ±), yielding a 3-excess of 2, and
2-adic symbol Iil(r+s), since det f == ± 3 (mod 8). The 2-adic symbol for
I r , s (3) would be I;-~ti).
A variant of this notation indicates the total dimension r + s explicitly
and treats - 1 like any other prime, thus writing
Ir+s (_S X) or IIr+s (_S X)

for

respectively.
On the Classification of Integral Quadratic Forms 385

8. Classification of forms of small determinant and of p-

elementary forms
8.1 Forms of small determinant. Using the notation introduced in the
previous section, the distinct genera of forms of any given determinant can
be classified in a systematic way. One first writes down all possible p-adic
symbols for p = - 1, 2 and all p dividing the determinant (it is best to
handle p = 2 last). The determinant condition (29) and the rules for
manipulating trains (§7.5) are used to control the signs. Then the oddity
formula (30) and the existence conditions (§7.7) lead to congruences
(modulo 8) relating the signature to the oddity parameters. We illustrate
this process by classifying the genera of forms with determinants ± 1 and
±3.

Determinant ± 1. Let the (-1)-adic symbol be +,_s, with r+s = n.

Since the determinant is == ± 1 (mod 8), the possible 2-adic symbols are
Ij+<'+S) and

for which the respective oddity conditions read

r-s == t and r-s == 0 (mod 8).

The former determines and corresponds to the genus symbol

I"s ... 1"s(1), which exists for all possible signatures except r = s = O.
The latter gives II" S ... II" S (1), which exists only for signatures divisible
by 8. (The existence conditions are trivially satisfied.) This explains lines
1 and 2 of Table 15.4 below.

Determinant ± 3. The possible p -adic symbols for

p= -1 3 2
are
1"s(3): +,_s 1± <r+s-1l3 1 Ii" (,+s)
I" S(3-1): +'_S 1~<r+s-llr1 Ii" (,+s)
II" S(3): +,_s 1± <r+s-1l3 1 1- <r+s)

1I"s(3-1): +,_s 1~(,+s-l)r1 1-('+s) ,

where the ambiguous sign ± is _s, and the 2-adic sign is - since the
determinant is == ± 3 (mod 8). These lead to the respective oddity
conditions
r-s +2 == t,
r-s +6 == t,
r-s +2 == 0,
r-s +6 == 0,
386 Chapter 15

the first two of which determine t, while the last two give conditions on the
signature. The existence conditions on the Jordan constituents are
automatically satisfied except for small n, when they are best handled by
consideration of all pairs (r, s) with r+s = n. Thus in the case Ir. s (3)
and for (r, s) equal to

(0,0), 0, 0), (O,U, (2,0), o,u, (0,2)

we find the 2-adic symbol to be

respectively, for which the existence condition holds in only three cases:

x J x J J x

The three failures are classified as nr - s = 0, L], 2-2 in Table 15.4.

Similar arguments lead to the following theorem.

Theorem 12. All genera of forms with Ideterminant I ~ II are as

indicated in Table 15.4.
The theory of spinor genera (see Corollary 21 below) shows that for
indefinite cases of dimension ~ 3 these are also the integral equivalence
classes.
In Table 15.4 we have used the convention that all ambiguous signs in
any row of the table are linked. The last column lists the exceptions, if
any, in the form n u , where n is the dimension and (5 the forbidden
signature. If all signatures of dimension n are excluded the subscript is
omitted. The asterisk * towards the end of the table indicates that when
n = I, this genus must be interpreted as IIr. s ( r l x 5- 1).
8.2 p-elementary forms. The same methods enable as to classify the so-
called p -elementary forms, that is, forms for which L' /L is a nontrivial
elementary abelian p-group, where L is the lattice of the form and L' is
the dual lattice. These forms were partially classified by Rudakov and
Shafarevich [Rudl], [Rud2], and the results used by Nikulin in [NikJ]-
[Nik6] (see also [Doll]).
Theorem 13. (a) For p ~ 3 the distinct genera of p -elementary forms
are
Ir.s(PH) for all signatures r-s,

and
IIr,s(P±k) for r-s == ±2-2-(P-Uk (mod 8),

except that in either case when k = n (= r+s) the sign must be [ -pi ) s
On the Classification of Integral Quadratic Forms 387

Table 15.4. All genera of forms with Idetl ~ 11.

Det Genus Signature Exceptions

(mod 8) nr- s
± 1 Ir, s all 0
± 1 IIr, s 0 none
±2 Ir,s(2) all 0,1
±2 II r , s (2) ±I none
± 3 Ir , s (3± I) all 0,1'1'1> 2'1'2
± 3 II r , s (3± I) +2 none
±4 I r , s (4± I) all 0, 1,2'1'2
±4 II r , s (4 1) ±I none
±4 II r , s (4- 1) ±3 none
±4 I r, s (2[) all 0,1,2
±4 II r , s (2[) 0, ±2 0
±4 I r , s(2ft) all 0, 1,2, 3±3, 4±4
±4 II r , s (2fl) 0 0
±4 II r , s (2ii 2) 4 none
± 5 I r, s (5) all 0
± 5 I r , s (5- 1) all 0, 1,20
± 5 II r , s (5) 4 none
± 5 II r , s (5- 1) 0 0
±6 Ir ,s(2 x 3±1) all 0,1
±6 II r ,s(2 x 3±1) odd I± 1
±7 I r ,s(7±1) all 0, I'l'l> 2'1'2
± 7 II r ,s(7±1) ±2 none
± 8 I r , s (8± I) all 0, I, 2:1'2, 3'1'3
± 8 I r ,s(8;j) all 0, I, 20 , 2'1'2, 3'1'1
± 8 IIr, s (8) ±I none
± 8 IIr, s (8- 1) ±I 1
± 8 I r,s(2 x 4) all 0,1,2
± 8 II r ,s(2x4) even 0
± 8 I r ,s(2 3 ) all 0,1,2,3
± 8 II r , s (2 3) odd 1
±9 Ir, s (9± I) all 0, I'l'l
±9 II r , s (9± I) 0 0
±9 Ir, s (32) all 0, 1,20
±9 Ir, s (3-2) all 0, I, 2±2
±9 II r , s (32) 4 none
±9 II r , s (3-2) 0 0
±1O Ir, s (2 x 5 ± I) all 0,1
± 10 II r ,s(2 x 5) ±3 none
± 10 II r , s (2 x 5- 1) ±I none*
± 11 Ir, s (11 ± I) all 0, I'l'l> 2'1'2
± 11 II r , s (11 ± I) +2 none
388 Chapter 15

Table 15.5. The 2-elementary forms.

I T • s (2{), 0 < k < n;

IT. s (2A), k even < n, and

if k = n-I then r-s == ± I (mod 8),
if k = n-2 then r-s =1= 4 (mod 8);

lIT. S<2f), 0 < k == n (mod 2), and

if k = I then r-s == ± I (mod 8),
if k = 2 then r-s =1= 4 (mod 8);

lIT s (2A), nand keven, r-s == 0 (mod 8);

lIT s(2iik), nand keven, 0 < k < n, r-s == 4 (mod 8).

When these forms are indefinite of dimension ~ 3, each genus contains

just one spinor genus, and so, by Theorem 14, just one class.
(b) The distinct genera of 2-elementary forms are as shown in Table
15.5. In all indefinite cases these genera again contain just one class.

9. The spinor genus

9.1 Introduction. The spinor genus, introduced by Eichler ([Eicl]; see also
[Earl], [Ear3]-[Ear5], [Hsi2]), is a refinement of the notion of genus, and
its importance stems from the following remarkable result.

Theorem 14. (Eichler [Eicl]; see also [Cas3, Chap. II, Theorem 1.4],
[Wat3, Theorem 63]). For indefinite forms of dimension at least 3, a
spinor genus contains exactly one integral equivalence class of forms.
The current use of the term differs slightly from Eichler's (by replacing
the orthogonal group by the special orthogonal group in some places). The
description given here is in essence due to Watson [Wat3], although since
Watson's description is rather complicated our treatment is based on the
version given by Cassels [Cas3], which we recommend to the reader who
would like to see the proofs. Cassels does not however give quite enough
information about spinor norms to justify the claim that the spinor genus is
a practicably computable invariant. We have therefore quoted from the
relevant theorem of Watson, and have taken some pains to produce a
mechanical rule for computing the spinor kernel.
The invariants we described in the previous section are invariants of the
genus. If two forms are in the same genus we can find a rational
transformation relating them whose denominator can be made relatively
prime to any given integer. If this transformation is integral (i.e. if the
On the Classification of Integral Quadratic Forms 389

denominator is 1) the forms are in the same class. The concept of spinor
genus arises when we apply local arguments to this rational transformation
to see what obstruction there is to making it an integral one.
Each genus is partitioned into a number of spinor genera (the number
is always a power of 2). There is a group of spinor operators acting
transitively on the spinor genera, so that we can obtain any spinor genus in
the genus from a fixed one by applying a suitable spinor operator. Thus
the division into spinor genera is determined once we name the spinor
operators and are able to decide when a given spinor operator acts trivially,
i.e. is in the spinor kernel. Our Theorems 15-17 serve as operational
definitions for the notions of spinor operator and spinor kernel.
The computations depend on the calculation of the spinor norms of
operations in certain orthogonal groups. Excellent references are Kneser
[Kne31, Hsia [Hsill, Earnest and Hsia [Ear2l These references have the
additional merit of discussing the problem over more general rings and of
giving the best possible conditions on the indices of prime divisors of the
discriminant to ensure that an indefinite genus contains just one class.

9.2 The spinor genus. If f and g are forms of determinant d in the same
genus, then they are rationally equivalent by some transformation whose
denominator is prime to 2d [Cas3, Chap. 91, [Wat3, p. 78]. Hence we can
find corresponding lattices Land M for which
[L: L nM] = [M: L nM] = r (say),

for some number r which is prime to 2d. We may paraphrase Watson's

redefinition of spinor genus by saying that f and g are in the same spinor
genus if and only if r is an automorphous number (as defined below).

Theorem 15. (Obtained from Theorem 70 of [Wat3].) (a) The spinor

genus of g (or M) is determined by that of f (or L) together with the
number r. Using SG to denote spinor genus we shall write
SG(g) = SG{J) • Mr) ,
SG(M) =SG(L) .Mr).

and call "Ll(r) " a spinor operator.

(b) Furthermore, if the dimension is at least 3, then SG{J) • Ll(r) is
defined for every natural number r prime to 2d.
To complete the definition we need to know how to find the r's for
which Mr) is in the spinor kernel, i.e. Ll (r) fixes every spinor genus. This
is done in the next section.
Remarks. (j) Part (b) of Theorem 15 fails for dimension 2. (ij) The
theorem is usually applied to indefinite forms of dimension ~ 3, in which
case by Theorem 14 we need not distinguish between the spinor genus of a
form and the form itself.
390 Chapter 15

9.3 Identifying the spinor kernel. Any element of the orthogonal group of
f, over a field not of characteristic 2, can be written as a product of
reflections in certain vectors v" V2, ... , Vk. The spinor norm (defined only
up to multiplication by square factors) of this operation is f(v) ... f(Vk).
This operation is proper (j.e. of determinant 1) or improper (determinant
- 1) according as k is even or odd. The proper operations form the special
orthogonal group of the form. Only some elements of the orthogonal
group of f have integral matrix entries: these are the integral
automorphisms of f .
We call r E Q automorphous if it is the spinor norm of a proper
integral automorphism of f. Similarly a p -adic number A = paa E Qp is
p -adically automorphous if it is the spinor norm of a proper p -adic
integral automorphism of f.
The following theorems are due to Eichler [EicIl and Watson [Wat31
We have obtained them by translating Cassels's versions into our notation
- see in particular Theorem 3.1 of Cassels [Cas3, Chap. III and its
corollary.

Theorem 16. The spinor kernel consists of the spinor operators a (r) for
which the positive integer r is an automorphous number not divisible by
any prime divisor of 2d.
It is important that we can compute the spinor kernel "locally", in fact
by performing a simple calculation for each prime in a certain finite set II.
In the terminology of the next section we have:

Theorem 17. The spinor kernel is generated by the spinor operators

ap (A) for which p E II and A is a p -adically automorphous number.

9.4 Naming the spinor operators for the genus of f. Let II be any finite
set of primes that contains -1, 2 and all primes dividing d = det f, where
f is some form in the genus. (We shall see later that often some primes
can be removed from II with no loss of information).
Since the spinor operators depend only on the square class of r, we can
name them by sequences C., rp, .. .)pElh in which each rp is a p-adic unit
square class. In this notation the group operation is componentwise
multiplication. Rational or p -adic integers can be regarded as spinor
operators in the following way.
(j) To any rational integer r not divisible by any p E II there corresponds
the spinor operator whose p -coordinate is the p -adic square class of r for
each p. We shall write this as
a{r) = (r, r, .. .) (37)

(the square class being understood).

On the Classification of Integral Quadratic Forms 391

(iO To each p-adic integer A = paa there corresponds the spinor operator
l1p (A) whose p I-coordinate (for PI ¢ p) is the P I-adic square class of pU,
and whose p-coordinate is the p-adic square class of a:
(38)

9.5 Computing the spinor kernel from the p-adic symbols. In view of
Theorem 17 the spinor kernel is determined once we know the p -adically
automorphous numbers. A vector v is called a p -adic root vector for f if
and only if the reflection in v is a p -adically integral automorphism of f.
Of course this reflection has determinant - 1.

Theorem 18. (Based on Theorem 81 of [Wat3]). A p-adic number is

p -adically automorphous for f if and only if it is the product of an even
number of norms of p -adic root vectors for f.

Remark. For odd p it follows from [Cas3, p. 115, Corollary 1] that every
p -adically integral automorphism is the product of p -adically integral
reflections. For p = 2 this is usually true but not always [O'Me4].

An algorithm for finding the p -adically automorphous numbers for a form.

In the rest of this section we describe a mechanical rule for finding the p-
adically automorphous numbers, that can be justified using Theorem 18.
§9.6 contains some examples and simplifications.
The algorithm is rather complicated to state, but is completely mechanical and
very easy to apply. If p ;z!: 2 we first diagonalize f, or if p = 2 we express f as a
direct sum of forms of the shapes

(qx) and [qa qb]

qb qc '

where q is a power of 2, a and c are divisible by 2, but x, band ac-b 2 are not
(see Theorem 2). We now prepare a list in two parts:

(J) If P ;z!: 2, all the diagonal entries, or if p = 2 the diagonal entries qx. (These
numbers are not yet to be interpreted modulo squares, and of course the list may
contain repeated entries.)

(II) Only if p = 2: the numbers 2quJ, 2qU3, 2qus, 2qU7 for every direct summand

[~~ ~n
Then the group of p-adically automorphous numbers is generated by the p-adic
square classes of the ratios (or products) of all pairs of numbers from the total list,
supplemented by:
(j) all p -adic units if

either p ~ 3 and fq has dimension ~ 2 for any q,

or p = 2 and fq Ell f2q Ell f4q Ell fSq has dimension ~ 3 for any q,
and
(ij) the square classes
392 Chapter 15

whenever p = 2 and part (I) of the list contains two entries whose ratio
has the form
(I or 4 or 16)uodd, {2 or 8)uI o's, {2 or S)U30,7

respectively.
Example. For the form f = diag {3, 16} and the prime p = 2, part (I) of the list
consists of {3 = U3, 16 - 16uI} and part (II) is empty. Since the ratio of 16 to 3
has the form 16u3, we supplement the total list by Us, according to (ii). Thus the
2-adically automorphous numbers are generated by the square classes of
{16ul> U3, us}, i.e. are {Ul> U3, us, U7l.
The supplementation rules correspond to the possibilities for root vectors not
necessarily in the basis. Thus e I +e 2, of norm 2 = 2u I> is a 2-adic root vector for
(6 ~), illustrating the first rule in (ii).
9.6 Tractable and irrelevant primes
The following considerations may be used to simplify the calculations. If there is a
prime p such that, for each p-adic unit u, the spinor operator
<1p{u) - (1, 1, ... ,1, u, 1, .. ') ... ,p,.. (39)

lies in the spinor kernel, then plainly the p-coordinate can be deleted, since it
conveys no information modulo the spinor kernel. Such p are called tractable. For
example, - I is always tractable, as is the prime 2 in the above example.
However, a tractable prime may still have some effect on spinor genus
calculations, since if p itself is automorphous, the spinor kernel will contain <1p (p),
which has nontrivial values in coordinates other than the p-th. For example, if f is
indefinite then -1 is {- 1)-adically automorphous, so that <1-1 (- 1) =
{+ 1, -1, -1, ".)-1,2,3, ... is in the spinor kernel.
If the p -adically automorphous numbers are precisely the square classes of the
p-adic units (as happens when p ¢ -lor 2 and p 1det(f», then p is not only
tractable but irrelevant. Irrelevant primes do not affect the computation of the
spinor genus in any way.
Example. We consider the form
2 1 0
f - [1 2 0
o 0 IS
(discussed in [Wat3, p. 115]). For p ¢ 2 we can diagonalize f, obtaining
diag {2,3/2, IS}. To find the spinor kernel we proceed as follows.
p = -1: list (I) = {2 = u, 3/2 = u, IS = u}, so U is the only {- 1)-adically
automorphous number.
p = 2: list (I) = {Is - 2u}, list (II) = {2ul> 2U3, 2us, 2U7}, so the 2-adically
automorphous numbers are {UI> U3, us, U7l. and 2 is tractable.
p = 3: list (I) = {2 = U_, 3/2 = 3u_, IS - 9u_l. so the 3-adically automorphous
numbers are {u+, 3u+l.
We need retain only the 3-coordinate since 3 is the only intractable prime, and the
spinor kernel is generated by
On the Classification of Integral Quadratic Forms 393

A2 (Uh U3, Us or U7) - (U+)3 ,

A3 (u+ or 3u+) - (U+)3

SO (U-)3 is not in the spinor kernel, and therefore the genus of f contains two
distinct spinor genera, one containing f, the other containing f.(u-h. A
representative for the second spinor genus is ([WatJ, p. lIS])

01
6 3
[3 6 0 .
002

9.7 When is there only one class in the genus? In practice one usually
finds that all primes are tractable and so the spinor genus coincides with
the genus (and therefore, in the case of indefinite forms of dimension at
least 3, the genus contains only one class). This section gives some
conditions which guarantee that this will happen.

Theorem 19. Iff is indefinite and the genus off contains more than one
class, then for some p (possibly - I), f can be p-adically diagonalized
and the diagonal entries all involve distinct powers of p.

Proof Suppose the contrary. Then dim f ~ 3, since otherwise in the

(- I) -adic (real) diagonalization the terms involve distinct powers of - 1.
We now quote Eichler's theorem (Theorem 14) to see that the class
coincides with the spinor genus and there must therefore be an intractable
prime p. If p ~ 3 we know that none of the p -adic Jordan constituents fq
can have dimension ~ 2 and the result follows. So we may conclude that
2 is the only intractable prime. No nontrivial 2-adic Jordan constituent
may have scaled type II, since then all 2-adic units are automorphous.
Thus f is 2-adically diagonalizable. If any two of the diagonal terms
involve the same power of 2, the algorithm implies that 5 is 2-adically
automorphous, and so (U5)2 is in the spinor kernel. But since -1 is (- 1)-
adically automorphous, (U7)2 is also in the spinor kernel, and these
generate all the possibilities. Therefore the diagonal entries may only
involve distinct powers of 2. This completes the proof.
With a little more care the argument can be refined to give the
following result.

Theorem 20. Suppose f is an indefinite form of dimension nand

determinant d.
(a) If p ~ 3 is an intractable prime then
(n)
d is divisible by p 2 (40)

(b) A prime p == 3 (mod 4) cannot be the only intractable prime.

394 Chapter 15

(c) If 2 is an intractable prime then

[~l n
4 2 d is divisible by 4 (2) (41)

(d) If 2 is the only intractable prime then

[~l n
4 2 d is divisible by 8 (2) (42)

(We recall that a prime p is tractable if dp(U) is in the spinor kernel for
every p -adic unit U.)

Proof (a) If an odd prime p is intractable then from part (j) of the
algorithm the powers of p in the diagonal terms of f are all distinct, and
so, when arranged in increasing order, must be at least po, pI, p2, ... ,
yielding a product of at least p (~.

(b) If p = 3 (mod 4) is the only intractable prime then every element of

the spinor kernel has a name (u ±) p. But - 1 is (-1) -adically
automorphous and is a non-residue modulo p, so (-1) p is in the spinor
kernel. Therefore the spinor kernel has order 2 and p is tractable, a
contradiction.

(c) If 2 is intractable then from part (II) of the algorithm there is no type
II summand, i.e. the form is diagonalizable. Moreover no three powers of
2 in the diagonal terms can lie in the range 21 to 21+3 (inclusive), for any
t. Therefore, when arranged in increasing order, the powers of 2 must be
at least

When multiplied by 1, 4, 1, 4, ... this sequence becomes

4°, 4 1, 4 2 , 43, ... ,

which implies (41).

(d) If 2 is the only intractable prime then every element of the spinor
kernel has a name (UI)2, (U3)2, (U5)2 or (U7)2, and since -1 is (-1)-
adically automorphous, (U7)2 is in the spinor kernel. Moreover, if any two
powers of 2 in the diagonal terms have ratio 1 or 4 or 16, then by part (ij)
of the algorithm (u 5) 2 is in the spinor kernel and all possibilities are
generated. So the even powers of 2 are at least
20, 26, 212, ... ,

and the odd powers of 2 are at least

2 1,2 7,2 13 , ....
On the Classification of Integral Quadratic Forms 395

The least possible values are therefore

which on multiplication by 1,4, 1,4, ... become

establishing (42).

Theorem 21. If f is an indefinite form of dimension n and determinant

d. with more than one class in its genus. then
[.!!.] (n)
4 2 d is divisible by k 2 (43)

for some nonsquare natural number k == 0 or 1 ( mod 4).

This is a strengthening of [WatJ, Corollaries 1 and 2 to Theorem 691.

Compare [Kne31, [Hsill, [Ear21. Without going into too much detail it is
worth mentioning that this result is nearly best possible. If 4[n/2]d is
divisible by k (2) for such a k, then there is a genus of forms with
determinant d or some small multiple of d that contains more than one
class.
Proof If the dimension is 2 the assertion is trivial. For certainly
d ~ ± 1, or else there is only one class in the genus. Therefore some
prime p ~ 2 divides d, and (43) holds with k = 4p.
For dimensions n ~ 3, as in the proof of Theorem 19, there must be at
least one intractable prime p ~ 2. If some p == 1 (mod 4) is intractable
then, from (40), (43) holds with k = p. If two primes p and q congruent
to 3 (mod 4) are intractable then (43) holds with k = pq. If both 2 and
p == 3 (mod 4) are intractable then k = 4p will do, from (40) and (41); if
2 is the only intractable prime then k = 8 will do, from (42); and by part
(b) of the theorem this has exhausted all the possibilities.
Corollary 22. Suppose f is an indefinite form of dimension nand
determinant d. with more than one class in its genus. Then Idl ~ do.
where do is given by the following table.
n 2 3 5,7,9, ...

17 128

Proof For large n, (42) implies that det f is divisible roughly by 8 (i),
whereas if 5 is intractable then from (40) we need only 5 (i) I det f. The
cutoff point turns out to be at n = 4, and for n ~ 4 the smallest
determinant (for an indefinite form with more than one class in its genus)
is that of the form
396 Chapter 15

omitting the final term if n is even. This determinant is

5(~) if n is even, 2'5(~) if n is odd. (44)

But (42) is better than (44) when n = 3. For n = 2, we find from Table
15.2 that the binary forms

[23 -43] and

[-23 43]
are in distinct classes although both belong to the genus 11 1,1 (I7). The
pair of ternary forms with determinant -128 is given at the end of this
chapter.
For definite forms the question of when there is only one class in the
genus behaves completely differently, and is the subject of a series of
papers by Watson [Wat5]-[Wat7], [Wat14]-[Wat221

10. The classification of positive definite forms

10.1 Minkowski reduction. We shall not say very much about this
important notion, since our main interest is in forms of large dimension
where it is impracticable. (For further information see the references on
reduction algorithms listed in §1.4 of Chap. 2,) Let f be a positive definite
n-dimensional form. f is said to be Minkowski reduced if it has been
expressed in terms of an integral basis e!>.oo, en such that for each t,
1 ~ t ~ n,
f(e t ) ~ f(v) for all integral vectors v for which e!>.oo, e t-!> v
can be continued to an integral basis. (45)

In other words each successive et is chosen so that f (e t) is as small as is

possible. By letting v range over all integral vectors, the condition (45)
implies inequalities on the matrix entries aij' It turns out [Cas3, p. 256,
Theorem 1.3] that only finitely many of these inequalities are necessary
("the fundamental region has finitely many walls"), but unfortunately their
number tends to infinity very rapidly with the dimension.
Some of these inequalities may be easily written down. (j) It is
immediate from (45) that
o< all ~ a22 ~ ... ~ ann. (46)

(ij) If we let v = et - ~ fse s (for some set S of subscripts s < t and

sES
coefficients fs = ± 1) the inequality f (et) ~ f (v) becomes
21 ~ fsast - ~ frfsa rs 1 ~ ~ ass·
sES r, sES sES
r<s
On the Classification of Integral Quadratic Forms 397

The cases S = {s}, {r, s}, {q, r, s}, ... lead to

(47)

21 ars ± art ± astl ~ arr+ass (r < s < r) , (48)

21 a a qt + {3 art + -y ast - a {3a qr - a -ya qs - {3 -ya rs I

~ aqq+arr+ass (q <r <s < r) , (49)

with a, {3, -y = ± 1, etc.

It is a theorem of Minkowski (see for example [Cas3, p. 257, Lemma
1.2D that for dimension ~ 4 a reduced form may be defined using vectors
v with coefficients equal to 0 or ± 1. In fact the inequalities

(46), (46).(47), (46)-(48), (46).(49)

(in which q, r, s, t ~ n) define a Minkowski reduced form for

n = 1, 2, 3, 4

respectively.
For n = 5, 6, 7 and 8, defining systems of inequalities for Minkowski
reduced forms have been given by Minkowski, Ryskov, Tammela and
Novikova - see [AffI], [Aff2], [Grul], [NovI], [Rys2], [Rys3], [Rys8],
[RysI4], [TamI]-[Tam4]. But the coefficients of v can no longer be
restricted to 0 and ± 1.
Many theorems in the geometry of numbers are consequences of the
inequalities (45) (see the references mentioned at the beginning of this
section). In particular they can be used to show that there are only finitely
many classes of forms of any given determinant [Casl, p. 256, Theorem
1.1] and in some case to enumerate these forms [Wae51 However, even in
dimension 3, the process is tedious for moderately large determinants, and
for much higher dimensions it is out of the question.
Tables 15.6 and 15.7 give all indecomposable, reduced, definite ternary
forms with determinant Idl ~ 50 and the indefinite forms with Idl ~ 100.
An entry abe f g h represents the form having matrix

a b h
b c f ,
h f g

and h is omitted when it is zero. Table 15.6 gives the positive definite
forms with d ~ 50, and was computed using the inequalities (46)-(48).
Table 15.7 gives the indefinite forms with Id I ~ 100, and was computed
using the theory of spinor genera.
398 Chapter 15

Table 15.6. Indecomposable positive definite ternary forms.

d Forms
4 2] 2] 2
2] 2] 3
2] 3 ] 2
]0 2] 2] 4
]2 2]4]2,3]2]3
]3 2 ] 2 ] 5, 2] 3 ] 3
]6 2 ] 5 ] 2, 2] 2 ] 6, 3] 3 ] 3_]
]7 3]2]4
]8 2] 3 ] 4
]9 2]2]7,2]4]3

20 2 ] 4 2 4, 3] 3 ] 3 ]
2] 3]3]3
22 2 ] 2 ] 8, 3] 2 ] 5
23 2]3]5
24 2 ] 7 ] 2, 4] 2 ] 4, 3] 3 ] 4_]
25 2 ] 2 ] 9, 2] 5 ] 3
262]4]4
27 3 ] 2 ] 6, 2] 4 2 5
28 2 ] 8 ] 2, 2] 2] ] 0, 2] 3 ] 6, 2] 5 2 4. 3] 3 ] 4 ]
29 3]3]4
30 3]4 ]3
3] 2]2]]],4]2]5
32 2]9]2,3]2]7,3]3]5_],3]4 2 4,4 2 4 2 4
33 2] 3 ] 7, 2] 4 ] 5
34 2 ] 2 ] ]2, 2] 5 ] 4, 2] 4 2 6
35 3]4]4_]
36 2]]0]2,2]6 2 4,2]5]5_],3]3]5],4]4]4_ 2
37 2 ] 2 ] ]3, 2] 7 ] 3, 3] 2 ] 8, 2] 5 2 5, 3] 3 ] 5
38 2 ] 3 ] 8, 4] 2 ] 6
39 3 ] 5 ] 3, 3] 4 ] 4 ]
40 2]]]]2,2]2]]4,2]4]6,5]2]5.3]3]6_],4]3]4
4] 2] 4 2 7. 3] 4 ] 4
42 3]2]9
43 2 ] 2 ] ] 5, 2] 8 ] 3, 2] 3 ] 9, 2] 5 ] 5, 3] 4 2 5
44 2 ] ]2 ] 2, 2] 7 24, 3] 3 ] 6 ], 3] 5 2 4, 4] 4 2 4, 4 2 4 2 5
45 4]2]7,2]5]6_],3]3]6
46 2 ] 2 ] ]6, 2] 5 26, 3] 4 ] 5_]
47 3 ] 2 ] 10, 2] 4 ] 7, 2] 6 2 5
48 2 ] ]3 ] 2, 2] 3 ] ] 0, 2] 4 2 8, 2] 6 3 6, 3] 6 ] 3, 3] 3 ] 7_],
4] 5 ] 4_ 2 , 4 2 5 2 4
49 2]2]]7,2]9]3,5]2]6,5]3]5_ 2
50 2 ] 7 ] 4, 3] 5 ] 4 ] , 4] 4 ] 4_]
On the Classification of Integral Quadratic Forms 399

Table 15.7. Indecomposable indefinite ternary forms.

d Forms
+8 ±21 ±21 +2
+28 ±21 +6 1 ±2
+32 ±21 +21 ±6
+56 ±21 ± 14 I +2
+64 ±41 ±41 +4
+64 ±42±42+4
+68 ±21 ±6 1 +6
+72 ±6 1 ±21 +6
+72 ±21 +21 ±14
+84 ±21 +20 1 ±2
+92 ±2 I +22 1 ±2

10.2 The Kneser gluing method. The integral lattices generated by vectors
of norm 1 and 2 are completely classified. Such a lattice can be written as
a direct sum of the particular lattices

Certain other lattices can be found by gluing these (and possibly other)
components together. This technique, due to Witt and Kneser [Kne41, is
described in §3 of Chap. 4. In Chaps. 16 and 17 we shall describe how, by
a combination of gluing and other methods, the unimodular lattices of
dimension n ~ 25 have been enumerated. However it is worth pointing
out that the enumerations of unimodular lattices can be used to find
lattices of other determinants in a fairly simple way, as the following
section will illustrate (cf. [Kne41, [Ple12]).
10.3 Positive definite forms of determinant 2 and 3. By following the
method used by Kneser [Kne41, and making use of the results of Chap. 16,
in this section we classify the forms of determinant 2 up to dimension 18
and determinant 3 up to dimension 17. This is enough to demonstrate the
techniques used, and since beyond this point the tables become unwieldy, is
a good place to stop. The results for determinant + dimension ~ 17 agree
with Kneser's.
Theorem 23. All positive definite forms of determinant 2 and dimension
~ 18, or determinant 3 and dimension ~ 17 are as shown in Tables 15.8
and 15.9.
Note. The tables explain these lattices in terms of unimodular lattices
taken from Chap. 16. As in that chapter a unimodular lattice is specified
by its component root lattices.
Outline of proof. Determinant 2. If Ln has determinant 2, then
L; / Ln has order 2 and there is a vector vEL; \ Ln with 2v E Ln and
400 Chapter 15

v'v == V2 (mod I). In the particular case L I = A h write w instead of v,

with w . w = V2. Then Ln EB A I can be extended by the glue vector v+w to
give a lattice L n+1 (say) of determinant 1. Conversely, Ln = w-L
(in Ln+l ) = {x E L n+l : X·W = O}, Thus all n-dimensional lattices Ln
of determinant 2 are uniquely obtained as the orthogonal lattices to norm 2
vectors w in (n + I)-dimensional lattices L n + 1 of determinant 1.
All such L n + 1 can be found (for n+ I ~ 23) in Chap. 16. Suppose
Ln+1 = M n+l - k EB h, where M n+l - k has minimal norm ~ 2. There are
now two possibilities for w.
(a) w Elk (if k ~ 2), so that Ln = w-L = M n+l - k EB Al EB I k - 2 • If Ln
has minimal norm 2 then we must have Ln = M n- I EB A I' These lattices
are shown in column (a) of Table 15.8.
We know from Table 15.4 that there are just two genera with
determinant 2, namely In (2) and lIn (2). Ln = M n- I EB A I is even (i.e. in
lIn (2» exactly when M n - I is. For n = 17 there are three lattices in

Table 15.8. Positive definite lattices of determinant 2, minimal norm

~ 2.

Dim. (a) (b) <2>-L in nI nIl nlol

0 - - 0 0 0
1 Al - 0 I 1
2 - - 0 0 I
3 - - 0 0 I
4 - - 0 0 1
5 - - 0 0 I
6 - - 0 0 I
7 - E8 0 1 2
8 - - 0 0 2
9 E8 EBA I - 0 1 3
10 - - 0 0 3
11 - D12 1 0 4
12 - - 0 0 4
13 D12EBAI Er 2 0 6
14 - A is 1 0 7
15 Er EB Al D 16 , E~, D~ 2 2 11
16 AlsEBAI AllE6 3 0 14
17 E~ EB Al A\7A I
DI6 EBA I DlOE 7A I
D~ EB Al Dg
A~ 6 4 24
18 AllE6 EB Al EgOI
A ll D 7 0 1
ArDs 6 0 30
On the Classification of Integral Quadratic Forms 401

column (a), two of which are even, although the table only indicates their
number. The actual components can be found from Chap. 16.

(b) Alternatively we can take w to be any norm 2 vector in a lattice M,,+I

of determinant 1 and minimal norm 2. For example there are exactly two
inequivalent choices for w in the I8-dimensional lattice A II E 6 • These
lattices are shown in column (b) of Table 15.8. In Tables 15.8, 15.9 the
symbol < c > denotes a one-dimensional lattice u Z with U'u - c.
The last three columns of Table 15.8 give nI (resp. nil), the number of
odd (resp. even) lattices with determinant 2 and minimal norm ~ 2 in
each dimension, and n" the number of lattices with determinant 2 and
minimal norm ~ 1.
Determinant 3. If L" has determinant 3 there is a vector vEL: \ L"
with 3v E L" and v'v == ± 1/3 (mod 1). It is easy to see that if
V'v == I/] (mod 1) then the 3-adic symbol for L" is 1',-1 31, and otherwise
it is 1,,-1 rl.
First we consider the case v'v == I/] (mod 1). We take A2 with
w E Ai \ A 2, w·w - 2/3 (Chap. 4, Eq. (55», and extend L" ED A2 by the
glue vector v+w to obtain a lattice L,,+2 of determinant 1. Conversely,
L" = Ai- in L,,+2' If L,,+2 - M,,+2-k ED Ik there are two possibilities.
(a) A2 C lko so (if L" has minimal norm ~ 2), L" = M,,_I ED < 3 >,
where M,,_I has minimal norm 2 and determinant 1. (b) A2 c M,,+2-k'
For example Ai- in E8 gives L6 = E 6.
Second, consider V'v == -If3 (mod 1). We take a I-dimensional lattice
MI (say) - <3>, with generator w of norm 3, and extend L" ED MI by
the glue vector v + 1/3 W to get a lattice L,,+I of determinant 1.
Conversely, L" is the orthogonal lattice to a norm 3 vector w in an (n +1)-
dimensional lattice L,,+I of determinant 1. If L,,+I "" M,,+I-k ED Ik there
are now three possibilities to consider. (c) w E lko (d) the projection of
w onto M,,+I-k has norm 2, while the projection onto Ik has norm 1, and
(e) w E M,,+I-k.
In case (d), suppose L,,+I - M" ED IJ, where M" has minimal norm 2.
Let w = v+e where v EM", v·v = 2, e E IJ, e'e = 1, and let K,,_I = v.L
in M", so that det K,,_I = 2 (from the first part of the proof). Then
L" = w.L contains K,,_J, and also the vector u = v-2e, which generates a
I-dimensional lattice < 6 > orthogonal to K,,_I' In fact L" is
K,,_I ED < 6 > extended by a glue vector t + 112 u, where t is a nonzero
glue vector for K:_ I / K,,_I'
Table 15.9 shows the lattices of minimal norm ~ 2 in the five cases,
and the number of even (nil) and odd (nI) lattices of minimal norm 2 in
each dimension. There are four genera of forms of determinant 3 (see
Table 15.4), namely 1,,(3±I) and 1I,,(3±I). Columns (a) and (b) belong
to either I" (31) or II" (31), and columns (c), (d) and (e) to either I" (3-1)
or 11,,(3-1). The final column, ntoto gives the number of lattices with
determinant 3 and minimal norm ~ 1.
402 Chapter 15

Table 15.9. Positive definite lattices of determinant 3, minimal norm

~ 2.

Dim (a) (b) (c) (d) (e)

<3>"- in <3>"- in
n, nil
n'OI I
At in
0 0 0 0
1 <3> 0 1
2 A, 0 1 2
3 0 0 2
4 0 0 2
0 0
6 E8 0
0 0
£8 Gl I, 0 4
9 E,Gl<3> 0
10 DI2 E8 Gl A, 1 7
11 DI2 0 8
12 Ef DI2 Gl I, 0 10
13 D 12 Gl<3> A,s Ej 0 13
14 D'6, Ei;, Di; DI2 Gl A2 Ej Gl I, A" 4 2 19
15 EjGl<3> A"E6 A"GlI, Di; 0 24
16 A"Gl<3> A"A, Ef Gl A, Ei Gl I, A"E.
DIOE,A, DI6 Gl I,
DJ D?; Gl I,
A? 12 0 36
17 EiGl<3> Elo, A,s Gl A, A"E6 GlI, A"A,
D'6 Gl <3> A"D,O, D,oE,A,
Di; Gl < 3 > AjD s Dt
A~ 17 0 53

11. Computational complexity

Finally we give a brief discussion of the complexity of the classification
problem. (The complexities of other questions connected with lattices are
discussed in §I.4 of Chap. 2.) The following are some of the principal
questions that we have encountered.

(CI) Find the number of classes of integral quadratic forms of dimension

n and determinant d. (C2) Exhibit one form in each class. (C3) Given
two forms, determine if they are in the same class. (C4) If they are, find
an explicit equivalence. (G I) Find the number of genera of forms of
dimension n and determinant d. (G2) Exhibit one form in each genus.
(G3) Given two forms, determine if they are in the same genus. (G4) If
they are, find an explicit rational equivalence whose denominator is prime
to any given number. A closely related problem is to find which numbers
are represented by a form. (S) Given a form f of dimension nand
determinant d, and an integer k, is there an integral solution to
f (x) = k? If so, find all solutions.
We shall not say much about (S). For n = 2 it was solved by Gauss
[Gaul, §§I80, 205, 212] (see also [BorS, p. 142], [CohS, p. 1], [EdwI, pp.
317, 330], [LagI], [Lag2], [Mor6]). The complexity of Gauss's solution
appears to be dominated by the complexity of factoring k, which is at most
exp(c .Jlog k log log k) for some constant c [Mor8], [PomI]. (Certainly
On the Classification of Integral Quadratic Forms 403

Gauss's solution is more efficient than the 0 (.Jk) solutions proposed in

[Dij11 and [Bac3] in the case f = X2+y2,) For general n not much is
known. This problem includes the determination of the minimal nonzero
norm of a lattice - see §IA of Chap. 2.
For problem (CO, in the case n = 2 there are explicit formulae for the
class number, which can be evaluated in a number of steps that is a
polynomial in d [Cas3, p. 371], [Dic2, Chap. vI].
For positive definite forms of fixed dimension n, all of (CO-(C4) and
(GO-(G4) can be solved by algorithms whose running time is a polynomial
in d. (By using Minkowski-reduced forms (§lO.O, all entries in the
matrices and vectors involved can be bounded by simple functions of d
(compare §3.2)') However these polynomials typically behave like d n2, so
the growth as a function of n is likely to be worse than exponential.
Furthermore the mass formula (see Chap. 16) shows that the class number
for definite forms grows at least as fast as nn 2 [Mi17, p. 501. Since it seems
unlikely that one can find the class number for n > 2 without determining
all the classes, the complexity of (CO and (C2) for definite forms as a
function of n seems to be worse than exponential.
In the remainder of this section we consider indefinite forms. If the
determinant is given in factored form, problem (CO is easy. Using the
method of §§7,8, the number of steps required is a polynomial in the
number of factors of d. If d is a prime, for example, the answer is given
in Theorem 13.
Our invariants for the genus and spinor genus also provide quick (and
polynomial-time) solutions to (GO, (G3) and usually (C3). We illustrate
with a notorious example. Dickson and Ross in 1930 were unable to decide
whether the ternary forms x 2-3y 2_2yz-23z 2 and x 2-7y 2-6yz-llz2 were
equivalent [Dic3, p. 1471. It is now known that they are equivalent
[Ben2], [Cas3, pp. 132, 2511, but we shall establish this using our
invariants. Replacing the forms by their negatives for convenience, the
problem is to decide the integral equivalence of

[} 2: J and [} ~ -~l
We first compute the genus in each case. By rational transformations of
denominators 3 and 7, respectively, the forms may be diagonalized to

diag {3, 638 , -I} and diag {7, 678, -l} .

Therefore, since 3 and 7 are odd and prime to the determinant -68, we
can read off the relevant p -adic symbols
p = -1: +2_ +2 _

P = 17: 1- 2 IT' 1- 2 IT'

P = 2: 12" 24 3' It24i'

404 Chapter 15

and since the first 2-adic symbol is converted to the second by a 2-step
walk, the forms are indeed in the same genus.
Let us compute the spinor kernel for the first form. Plainly 2 is the
only intractable prime, although -1 is still relevant since it tells us that
-1 is (- U-adically automorphous and so (U7)2 is in the spinor kernel.
Since two of the terms in the 2-adic diagonalization have ratio - 3 = U 5,
(U 5h is also in the spinor kernel, which therefore includes everything, and
so there is only one class in the genus. Thus the two forms are integrally
equivalent. In fact the unimodular matrix
-3 2
M = -2 3 10]
14 (50)
-1 5

transforms the first form into the second.

There do not seem to be good algorithms for the remaining problems
(C2) -(C4) , (G 2) and (G4). Logically there is no difficulty: the proofs of
the theorems on genus and spinor genus (Theorems 9, 10, 14 etc.) are at
bottom computationally effective. For example, for problems (C3), (C4)
and (G4) , if two forms are known to be in the same genus, we can in
principle search through all rational matrices until a rational equivalence of
denominator r prime to 2d is found. The forms are then in the same class
if and only if d(r) is in the spinor kernel. If they are in the same class we
can continue the search until an integral equivalence is found. (The
matrix (50) was essentially found by this procedure, after using the
diophantine equations resulting from Eq. (4) to restrict the search.) For
problems (C2) and (G2) we search through all integral matrices in turn,
applying these techniques, until we have found the correct number of
distinct classes or genera of forms of determinant d. But these are only
logicians' solutions, and it is not clear to us that in general they can be
converted into practical algorithms.
Often there is some convenient artifice. Sometimes one can make use
of Gauss's complete theory of binary forms. We illustrate by finding two
inequivalent indefinite ternary forms of determinant - 128 of the same
genus (see Corollary 22). It is easy to see that the genus of
f = diag {-I, 64, 2}, namely 12,1 (2X64), contains two spinor genera and
hence two classes, namely f and f * dO). We must find a representative
for the second class. Thus we wish to find two lattices Land M in this
genus whose intersection has index 3 in each of them. Among the binary
forms of determinant - 64 we find the form [-?
~), which remains
integral when its first row and column are divided by 3 and simultaneously
its second row and column are multiplied by 3, yielding [-l
61) So the
ternary forms

(a) [- ~ 1
63 ~] and (b) [-9 0]
170
002
002
On the Classification of Integral Quadratic Forms 405

represent lattices Land M for which generators can be chosen in the form
eh e2, e3 for Land 3eh e 2i3, e3 for M. Now L also possesses the
diagonal basis eh e\+e2, e3, showing that it corresponds to the original
form f. The matrix (51 b) is therefore a representative for the second class
in this genus.
 16
Enumeration of Unimodular Lattices
J. H. Conway and N. J. A. Sloane

In this chapter we state explicit formulae for the Minkowski-Siegel mass

constants for unimodular lattices. We give Niemeier's list of 24-
dimensional even unimodular lattices, use the mass constant to verify that
it is correct, and then enumerate all unimodular lattices of dimension
n ~ 23.

1. The Niemeier lattices and the Leech lattice

The even unimodular (or Type II) lattices in 24 dimensions were
enumerated by Niemeier in 1968 [Nie2]. There are 23 such lattices with
minimal norm 2, and one, the Leech lattice A24 , with minimal norm 4.
(A 24 was already known to be the unique 24-dimensional Type II lattice
with minimal norm 4-see Chap. 12.)
The Niemeier lattices are shown in Table 16.1. The first column labels
each lattice with a single Greek letter, for use in Chap. 17. Each Niemeier
lattice A is obtained by gluing certain component lattices together by
means of glue vectors, as described in §3 of Chap. 4. The second column
gives the component lattices, which belong to the list An (n ~ 1), Dn
(n ~ 4), E 6, E7 and Eg. In this chapter we usually refer to lattices merely
by their components, for example A llD7E 6 represents one of the Niemeier
lattices. Elsewhere we should write (A ll D 7 E 6 )+ to indicate the presence of
glue. If A has components L I, . . . , L k , the glue vectors for A have the
form Y = (YI, ... ,Yk), where each Yi can be regarded as a coset
representative (or glue vector) for Lt modulo L j • These coset
representatives, labeled [0], [11. ... , [d -1] for a component of
determinant d, are listed in Chap. 4.
The set of glue vectors (y I, . . . ,Yk) for A forms an additive group
called the glue code. The third column of Table 16.1 gives generators for
the glue code. If a glue vector contains parentheses, this indicates that all
vectors obtained by cyclically shifting the part of the vector inside the
Enumeration of Unimodular Lattices 407

Table 16.l. The 24-dimensional even unimodular lattices L. The order

of the glue code is I G~I. Also Aut(L) has order IGol IGd IG11, where
IGol can be found from (2). h is the Coxeter number, 24h is the number
of vectors of norm 2, V is the number of Leech lattice points around the
corresponding deep hole (see Chap. 23).
Name Components Generators for glue code IG~I IGII IG21 h V

[I] I 46 25
"{3 D24
D16 E g [10] 2 I 30 26
I' EJ [000] I 6 30 27
0 A'4 [5] 2 I 25 25
DI, [12], [21] 4 I 2 22 26
I A 17 E, [31] 6 2 18 26
~ DiOEj [1101, [301J 4 18 27
0 A15 D 9 [21J 8 2 16 26
DJ [(122) J 6 14 27
A I, [15J 13 2 2 13 26
A A llD,E 6 [III] 12 2 12 27
[I (012)J 24
" Et 9 12 28
A~ D6 [2401, [SOIl, [053J 20 2 2 10 27
Dt [even perms of (0123) J 16 24 10 28
AJ [(114) ] 27 2 6 9 27
7r Aj D5 [ II 121, [1721 J 32 2 4 28
p At [I (216) J 49 2 12 7 28
(J At D4 [2(024)01, [330011, [30302J, [30033] 72 2 24 6 29
DS [111111], [0(02332) J 64 3 720 6 30
AS [I (01441)] 125 2 120 30
1> A~ [3(200101 I)] 256 2 1344 4 32
X Aj' [2(11211122212)J 729 2 IMd 3 36
y; AI4 [I (0000010100110011010111 I)] 4096 IM241 48
w Leech 0

parentheses are also glue vectors. For example, the glue vectors for the
Niemeier lattice DJ in Table 16.1 are described by [(J 22»), indicating that
the glue words are spanned by
[122) = «V2)8, (011), (011»
[212] = «(011), (Vl)8, (011))
[221) = «(011), (011), (Vl)8)

- the glue vectors for Dn are given in §7.1 of Chap. 4. The full glue code
for this example contains the eight vectors [000), [122), [212), [221),
[033], [303], [330), [Ill). The number of glue vectors, IGool (this
notation is explained at the end of Chap. 24), is given in the fourth
column.
The automorphism group Aut(A) is compounded of groups Go, G" G 2
and has order
408 Chapter 16

as described in Eq. (4) of Chap. 4. The group Go is the direct product of

subgroups of order
(n+I)!, 2 n - I n!, 27 34 5, 2 1°3 4 5'7, 2 14 35 52 7, (2)

for components An, D n , E 6, E 7, E8 respectively (these groups are the Weyl

groups for the corresponding root lattices). The orders of G I and G 2 are
given in columns 5 and 6 of the table. Column 7 gives the Coxeter number
h for each lattice (which is the Coxeter number of any of its components).
Then the lattice contains 24h vectors of norm 2.
The glue codes for the Niemeier lattices Et, DS, AF, A [4 are
respectively the tetracode C(j4, the hexacode C(j6, and the Golay codes C(j12,
C(j 24 (§§2.5.1, 2.5.2, 2.8.5, 2.8.2 of Chap. 3). The glue codes for A [2 and
AllD7E6 are given in full in Chap. 24. For AS the group G 2 (A9) is
isomorphic to PGL 2 (5) acting on {oo,O,I,2,3,4}. For Af the group
G 2 (Af) is isomorphic to 23 . PGL 2 (7) acting on the extended Hamming
code of length 8 over the integers modulo 4. The glue codes are also
described (often using different coordinates) by Venkov in Chap. 18.
Other references dealing with the Niemeier lattices are [Erol1-[Er03l
The Niemeier lattice A [4 is equivalent to the lattice obtained by
applying Construction A (§2 of Chap. 7) to C(j24, has p = 1/.J2, 0 = T12,
T = 48 and is of course an even integral lattice.

The lattices in Table 16.1 are arranged in decreasing order of h, and in

tied cases by increasing size of G 2 . This ordering has the property that
occurrences from anyone of the families An, Dn, En are in descending
order of n.
In §3 we give two different verifications that Niemeier's list is correct
(and another is given in Chap. 18). First we state the mass formulae.

2. The mass formulae for lattices

As mentioned in §2.4 of Chap. 2, unimodular lattices have the property
that there are explicit formulae, the mass formulae, (0 which give
appropriately weighted sums of the theta-series of all the inequivalent
lattices of a given dimension. Here we just give the formulae for the
number of inequivalent lattices, the Minkowski-Siegel mass constants.
Theorem 1. Let n be the set of all inequivalent odd unimodular lattices
of dimension n. Then

is equal to V2 for n = 1, and otherwise to

(OThe German original "Massformel" actually means "measure formula".

The standard mistranslation is a happy one.
Enumeration of Unimodular Lattices 409

(I-rk)(I+21-k) IB ·BB···B
2. k! k 2 4
I If
2k-2 ,
n=2k==0(mod8),

k~k2~~1 IB 2B4···B2k l, if n = 2k + 1 == ± 1 (mod 8),

(k-l;!2 2k + 1 IEk-l·B2B4···B2k-21, if n = 2k == ±2 (mod 8),

k~k2~~1 IB 2B C· B 2kl, if n = 2k + 1 == ±3 (mod 8),

(I -rk) (I _2 1- k )
-'-----'---'----"'-IBk·B2B4···B2k-21, if n = 2k == 4 (mod 8).
2·k!

Here Bk and Ek are the k th Bernoulli and Euler numbers. respectively:

Bo=l. B 1 =-1/2. B2=1/6, ... , Eo=l. E 2 =-I. E4=S, ... [Abrl.
p.810).
Theorem 2. Let n be the set of all inequivalent even unimodular lattices
of dimension n. Then

if n = 2k == 0 (mod 8). the right-hand side being interpreted as 1 when

n = o.
Theorem 3. Let n be the set of all inequivalent unimodular Eisenstein
lattices of dimension n. let

and let an = 3n- 1IB n (I/3)I/n, where Bn(x) is the nth Bernoulli
polynomial [Abrl. Chap. 23]. Then Ml = 1/6,
3n/2 + (_ I)n/2
Mn = Mn-1Bn 2n if n even,

Remarks. (i) These theorems have a long history. The versions given here
are taken from [Con34], [Fei2], [Serl], [SlolO]. For further information
and generalizations see [Bra3], [BroS], [Bro6], [Cas3], [CosIl, [ Eicl],
[Hsi6a], [Ko2], [Mag3], [Mil7], [O'MeIl, [Pall], [Pa12], [PfeIl-[Pfe3]'
[Que3], [SieIl-[Sie3). The reader is warned that there are mistakes in
many of the published formulae for the mass constants. (ij) Numerical
values of the mass constants in Theorems 1 and 2 for n ~ 32 are given in
Tables 16.2-16.5. (iii) Examples illustrating the use of these theorems
were given in §2.4 of Chap. 2. Other examples appear in §§3,4 below. (iv)
Analogous theorems for codes may be found in [BroS], [Mac4], [Mac6,
Chap. 19, §6], [Mac8], [MaI2], [Ple7], [Ple8], [Plel6], [Plel7], [Tholl.
410 Chapter 16

Table 16.2. Exact values of the Minkowski-Siegel mass constants for

Type I lattices.

n mass

1/2

1
2
8

3
48

4
384
1
5
3840

6
46080

7
645120
1
8
10321920
17
9
2786918400

10
2229534720
31
11
735746457600
31
12
5885971660800
691
13
765176315904000
42151
14
192824431607808000
29713
15
385648863215616000
505121
16
12340763622899712000

3. Verifications of Niemeier's list

As an application of Theorem 2 we prove:
Theorem 4 [Nie21. The even unimodular lattices of dimension 24 are as
shown in Table 16.l.
Proof ([Con341; see also [Ero3]). We shall verify that

f
1 1027637932586061520960267 (3)
IAut{A) I 129477933340026851560636148613120000000 '
Enumeration of Unimodular Lattices 411

Table 16.2 (cont.)

n mass

642332179
17
18881368343036559360000
692319119
18
15105094674429247488000
8003636403977
19
77489135679822039613440000
248112728523287
20
619913085438576316907520000
593468652605200909
21
216969579903501710917632000000
50904295073459007001
22
1507367607750643465322496000000
1015740532498234470066371
23
1317439289174062388691861504000000
701876707956280018815862361
24
21079028626784998219069784064000000
84715059480304651623612272842147
25
30465396080006318014267329085440000000
14616335635894388876188472684851927
26
31871491283698917307233513504768000000
1894352751772146867430486995462923265007
27
12429881600642577749821070266859520000000
10345060377427694043037889482223023950203227
28
99439052805140621998568562134876160000000
4285009823959590682115628739356169586687220752159
29
28837325313490780379584883019114086400000000
156429914319579070270102710292957201465725850451195039
30
346047903761889364555018596229369036800000000
447543572700878404232772149927275573042725639059585587715489
31
150184790232659984216878070763546161971200000000
416022631331982837344253774635747627157753429574979915187099394241
32
9611826574890238989880196528866954366156800000000

where the sum is taken over all lattices A in Table 16.1. IAudA) I is given
by (1), where I GIl and I G 2 1 can be found in columns 5 and 6 of the table,
and IGol is the product of the IGol's for the components of A, which can
be obtained from Eq. (2) (except for the Leech lattice, for which
IAudA) I = ICool is given in §3.3 of Chap. 10). The second column of
Table 16.6 gives IAut (A) 1-1 times the denominator of the right-hand side
of (3). Since the sum is equal to the numerator of the right-hand side of
(3), the list is complete.
412 Chapter 16

Table 16.3. Decimal expansions of the mass constants for Type I lattices.

n Mass n Mass n Mass n Mass

0.5 9 6.100xlO- 9 17 3.402 x 10- 14 25 2.781 x 10- 6
2 0.125 10 4.485 x 10- 10 18 4.583 x 10- 14 26 4.586 x 10- 4
3 2.083 x 10-2 II 4.213 x 10- 11 19 1.033 x 10- 13 27 1.524 x 10- 1
4 2.604 X 10- 3 12 5.267 x 10- 12 20 4.002 X 10- 13 28 1.040 x 10 2
5 2.604 X 10-4 13 9.031 x 10- 13 21 2.735 x 10- 12 29 1.486 x 10'
6 2.170 X 10- 5 14 2.186 x 10- 13 22 3.377 x 10- 11 30 4.520 X 10 8
7 1.551 X 10-6 15 7.705x 10- 14 23 7.710x 10- 10 31 2.980x 10 12
8 9.688 X 10-8 16 4.093 x 10- 14 24 3.330 x 10-8 32 4.328 x 10 16

Table 16.4. The Minkowski-Siegel mass constants for lattices of Type II.

n mass

o
8
696729600
691
16
277667181515243520000
1027637932586061520960267
24
129477933340026851560636148613120000000
4890529010450384254108570593011950899382291953107314413193123
32
121325280941552041649762780685623131486814208000000000

Table 16.5. Decimal expansions of the mass constants for Type II

lattices.

n o 8 16 24 32
Mass 1.435xlO-9 2.489 x 10- 18 7.937xlO- 15 4.031x10 7

A more illuminating proof appears in Chap. 18. We sketch yet another

proof, which uses some results from later chapters. Let G denote the
group of all autochronous automorphisms of the even unimodular 26-
dimensional Lorentzian lattice II 25 ,]' and let H be the reflection subgroup
of G. The groups G and H are given in Chap. 27, where in particular it is
shown that H is a Coxeter group whose graph is isomorphic to the Leech
lattice. From Vinberg's work [Vin5] it follows that the even unimodular
24-dimensional Euclidean lattices of minimal norm 2 are described by
those maximal subdiagrams of the Leech lattice that are unions of the
extended Coxeter-Dynkin diagrams An (n ~ ]), Dn (n ~ 4), E 6, E 7 and E 8·
But in Chap. 23 it is shown that there are precisely 23 such subdiagrams,
which are exactly those corresponding to the Niemeier lattices of minimal
norm 2. This result, plus the fact that the Leech lattice is the unique even
unimodular lattice of minimal norm ~ 4 (Chap. 12), provides another
Enumeration of Unimodular Lattices 413

Table 16.6. Verification that Niemeier's list is complete.

D" 24877125
E~ 63804560820
D,.E. 271057837050
A'4 4173688995840
D;, 67271626831500
A 17 E, 3483146354688000
DIO E ; 4134535541136000
A "D. 33307587016704000
D~ 156983146327507500
A;, 834785957117952000
E: 373503391765504000
A"D,E. 8082641116053504000
D: 19144966823230248000
AiD. 106690862731906252800
A~ 225800767686574080000
AiD; 2700612462901377024000
A: 8361079854908571648000
D~ 1196560426451890500000
A~D4 52278522738634063872000
A~ 180674574584719324741632
A~ 437599241673834240000000
A ~' 312927932591898624000000
A;4 31522712171959008000000
Leech 15570572852330496000
Total 1027637932586061520960267

proof of Niemeier's result. It also clarifies the one-to-one correspondence

between the deep holes in the Leech lattice (Chap. 23) and the Niemeier
lattices. We must emphasize, however, that in no sense is this a short-cut
to Niemeier's result, for the proofs in Chap. 23 require extensive
computations.

4. The enumeration of unimodular lattices in dimensions

n ~ 23
References to earlier work on the classification of unimodular lattices are
given in §2.4 of Chap. 2, and the known results are summarized in
Table 2.2 of that chapter. The odd lattices in dimensions 24 and 25 are
discussed in Chap. 17. In the rest of this chapter we shall enumerate the
unimodular lattices of dimension n ~ 23.
Theorem 5 [Con34]. The unimodular lattices of dimension not exceeding
23 and containing no vectors of norm 1 are as shown in Table 16.7.
Remark. The unimodular lattices of minimal norm 1 are now easily
determined, since they have the form Zk ED L, where the minimal norm of
L is at least 2 (seeEq. (4)).
Proof We refer to the lattices in Table 16.1 as the Niemeier lattices. The
proof of Theorem 5 is based on the remark (justified below) that any
lattice of the desired type is associated with a certain Niemeier lattice, and
414 Chapter 16

consists in finding all lattices that can be obtained from the Niemeier
lattices.
The lattices in Table 16.7 have components Vb V 2,... (shown in the
column headed V). The V; are taken from the list On (n ~ I), an empty
component (that is, one containing no vector of norm ~2); Z, the one-
dimensional lattice of integers; and An (n ~ 1), Dn (n ~ I), E 6, E 7 and E 8.
As usual the subscript on a component indicates its dimension. We
sometimes use 1m to denote the m -dimensional integer lattice zm.
Remark. The notation used in this chapter to specify the components of a
lattice differs slightly from that used in [Con27]. For example the lattice
El 0 1 was there called El[3l. The latter notation is more informative but
less general (it fails when the empty component has dimension greater than
one), and somewhat confusing, since [3] is also the name of a glue digit.
We therefore recommend the ... On notation for general use.
Chains of lattices. Every unimodular lattice A appears in a uniquely
determined chain of the form

where the initial lattice An does not represent 1. We call An the reduced
version of A. (Beware: elsewhere in this book An usually denotes a
laminated lattice,) The summand 1m is the sublattice of A generated by
vectors of norm 1, and An is its orthogonal complement. In these
circumstances we have

Our enumeration process considers all lattices of a chain simultaneously.

For some purposes the most appropriate lattice to consider is the initial
lattice An, but for other purposes it is the 23-dimensional lattice
An ffi h3-n'

The associated Niemeier lattice. We now describe how an odd unimodular

lattice An of dimension n ~ 23 is associated with a uniquely determined
Niemeier lattice N. For brevity we omit the easy justifications of some
statements, since in any case the mass formula provides an independent
verification. Let m = 24-n. Both An and 1m have sublattices of index 2
containing only vectors of even norm, denoted by A~ and I~, respectively.
The dual of A~ consists of A~ and three cosets, say A~, A;, A~, and
similarly the dual of I~ consists of the cosets I~, I~, I';', I~. The notation
can be chosen so that

and the set of all vectors

{x+y: x E A~,y E I!, k = O,1,2,3}

Enumeration of Unimodular Lattices 415

is an even unimodular 24-dimensional lattice. The latter is the Niemeier

lattice N associated with An. (A and N are neighbors, as defined in the
following chapter.)
Constructing An from N. The lattice I~ is better known as Dm, and seen
from the point of view of N the process we have just described appears as
follows. We look for a copy of the lattice Dm in N and observe that An is
the set of vectors x in the orthogonal complement Dj, for which either
x + [0] or x + [2] is in N, where [OJ, [11, [2], [3] are the glue vectors for
Dm in Dj" [2] being a glue vector of norm 1 (see §7.1 of Chap. 4). An
will be the initial member of its chain if Dm is not contained in a Dm' in N
with m > m using the same glue vector [2].
I

Alternatively we can refer to the 23-dimensional member of the chain,

for which the appropriate" Dm is a D\ generated by a single vector v = 2e
of norm 4 and [2] = e. This 23-dimensional lattice is the set of vectors
x E e.L for which x + ne E N for some integer n. The other members of
the chain can be obtained by removing summands I k •
Thus our process can be viewed from either end of the chain containing
An. On the one hand we search for a maximal sublattice Dm (and glue
vector [2]) in the Niemeier lattice N, and locate the initial member of the
chain in Dj,. Alternatively we find each possible vector v = 2e of norm 4
in N and locate the 23-dimensional member of the chain in the subspace
v.L.
The Niemeier lattices of minimal norm 2. Let N be one of the 23
Niemeier lattices with minimal norm 2, and let WI. W 2,... be its
components. The norm 4 vector v used in the construction may then be
written as

where VI. V2, V3,'" are either (1) minimal representatives of glue digits
corresponding to a glue word of norm 4; (2) minimal vectors r, s of norm 2
in two distinct components Wi, W;, and 0 in the other components; or
(~3) 0 except for one Vi which is a vector of norm 4 in its component W;.
(Again we omit the easy proof of this statement.) The cases have been
numbered so as to correspond with the value of m mentioned above.
In case (1) the D\ generated by v is maximal and we obtain a 23-
dimensional lattice not representing 1. Our symbol for this case is the glue
word mentioned.
In case (2) the D2 generated by rand s is maximal with
[2] = V2(r+s), and so the reduced lattice is 22-dimensional. The symbol
for this case is a word with *'s in positions i and j and O's elsewhere.
The cases (~3) can be further subdivided as follows. Case (3): v - Vi
is the vector (1, 1, -1, -1, Ok-3) in a component Wi - Ak for some
k ~ 3. The maximal Dm is the sublattice A3 of Ak supported in the four
nonzero coordinates of Vi' The reduced lattice is 21-dimensional, and the
symbol for this case has a + in position i and O's elsewhere.
 Table 16.7. The unimodular lattices of dimension ~ 23 that contain no
vectors of norm 1. All are odd (or of type 1) except for four lattices: the
empty lattice, E 8, Ei, and D I6 , which are even (or of type II). The
notation is explained in §4.

dim N w V c(w) gig, 12

0 D24 IZl 0

8 D\6E S -0 E. 240

12 Di2 -0 D12 264

14 DlOE~ -00 E~ 252

---.-.---~

15 A\sDy 0- A" 2 240

16 E~ -00 E~ 2 480
16 D'6 E • 0- D'6 1 480
16 D: -00 D~ 224

17 Al1D7E 6 0-0 All E 6 2 204

18 A17E 7 0- A"A, 1 308

18 DlOE~ 0-0 DIOE 7A , 2 1 308
18 D! -000 D~ 4 6 180
18 A~D6 00- A~ 4 180

19 E! -000 E~O, 4 12 216

19 Al1D7E 6 00- A l1 D7 O, 2 216
19 A~D~ 00-0 A~D5 2 4 152

20 D24 + D20 760

20 D'6 E • +0 DI2Ep. 1 504
20 DI2 +0 D 12 D. 2 376
20 DIOE~ +00 E~D6 1 312
20 A"D 9 0+ A"D, 280
20 D: +00 D~D4 2 248
20 Al1D7E 6 0+0 AllE6A 3 2 216
20 D: +000 D~Ai 4 6 184
20 A~6 00+ A~Ai 1 4 184
20 A;D~ 00+0 A~DsOl 2 4 152
20 D~ +0 5 D~ 6 120 120
20 A~D4 04 + A~ 1 16 120

21 A24 + A 20O, 2 420

21 Al7E 7 +0 A l3 E7O, 308
21 A"D 9 +0 A l1 D9 O, 2 276
21 Ai2 +0 A l2A.O, 2 228
21 AIID7 E 6 +00 D7A 7E 60 , 2 212
21 A~D6 +00 A 9 D 6A,O, 2 180
21 A~ +00 A~A40, 3 4 164
21 A~D; +000 A 7D;A,O, 2 4 148
21 A! +000 A~A20, 4 6 132
21 A~D4 +0000 A~D4A,O, 4 12 116
21 A~ +0' A~O, 6 40 100
21 A~ +0 7 Aj 8 336 84

22 D'6 E • D'4E7A , 492

22 E: **0 E.E~ 2 492
22 Di2 VioAI 2 364
22 A17E 7 A"D6O, 2 300
22 DlOE~ 0 .. DIOD~ 2 300
22 DlOE; --0 D.E7DoA, 2 1 300
22 AlSD9 A l3 D7A,O, 1 2 268
22 D: ..0 D8D~Ai 236
22 Ai2 Alo02 1 4 220
22 E: ..00 E~Al 6 8 204
22 Al1D7E 6 0** A l1 D,A,A, 2 204
22 AllD7E6 -0- A 9 D 7A,O, 2 204
 Table 16.7 (concluded)

dim N w V c(w) glg2 12

22 A l1D7E. ..0 A.EoD,AIOI 1 2 204

22 D! .. 00 D~D~Ai 6 4 172
22 A~D. ..0 A~D.02 1 4 172
22 A~D. .0- AoA7D4A I02 2 2 172
22 A~ .. 0 A.A~02 3 4 156
22 A~D; 00.. A;A;Af 1 8 140
22 A~D; ..00 D;A;02 1 8 140
22 A~D; .0.0 A7D,A,A,A IOI 4 2 140
22 A! ·-00 A~;02 6 4 124
22 D! **04 D:A~ 15 144 108
22 A~D4 -000- A;A,A~OI 4 12 108
22 A~D4 ..000 A;D4A~02 6 8 108
22 A! **04 A:A~02 15 16 92
22 A~ **06 A~Aio2 28 96 76
22 A~2 **0 10 Aio0 2 66 2880 60
22 Ai4 **0 22 Ai2 276 887040 44

23 DI.E. 10 AlSEs 2 480

23 A24 5 A19 A 4 2 2 400
23 Df2 12 DllAllDl 2 2 352
23 Al1E 7 60 A l1E7A, 2 2 288
23 DlOE~ 110 A.E7E.O I 2 2 288
23 DlOE; 211 D.E~02 1 4 288
23 A17E 7 31 A I4E.A 20 1 2 2 288
23 AI,D. 80 D.A~ 1 4 256
23 AI,D. 21 A 13 A.A IO I 2 2 256
23 AI,D. 42 A l1 D.A,OI 2 2 256
23 D~ 033 D.A~OI 3 4 224
23 D~ 122 D~A702 3 4 224
23 Ai2 15 Al1A7A 40 1 4 2 208
23 Ai2 32 A u,A.A 2A IO I 4 2 208
23 E! 0111 E.D;02 8 12 192
23 A l1D7E. 620 E.D.A;OI 1 4 192
23 A l1D7E. 401 D7A7D,A,OI 2 2 192
23 A l1D7E. 330 A.E.A.A20 1 2 2 192
23 A l1D7E. 111 A l oA.D,02 2 2 192
23 A l1D7E. 222 A.D.DsA I0 2 2 2 192
23 D! 2222 D~O, 1 48 160
23 A~. 501 A.AsA;OI 2 4 160
23 D! 0123 DoD,A;02 12 4 160
23 A~. 240 A7D•A sA ,AIOI 4 2 160
23 A~D. 312 A.AoD,A20 2 4 2 160
23 A~D. 121 A.A7A SA I02 4 2 160
23 A~ 036 A.A;A~OI 6 4 144
23 A~ 411 A~A4A,02 6 4 144
23 A~ 177 A7A~A~02 6 4 144
23 A;O; 4400 D;A~OI 1 16 128
23 A;O; 4022 A7D;A~02 2 8 128
23 A~D; 2031 A7A SAiA I 02 4 4 128
23 A~D; 2220 DsA;D4A~02 4 4 128
23 A;D; 1112 A~4A40, 4 4 128
23 A~D; 1303 A.DsA~A202 8 2 128
23 A! 5111 A;A4A IO, 8 6 112
23 A! 0124 AoA,A 4A,A 2A I0 2 24 2 112
23 D! 002332 D~A~O, 45 96 96
23 A~D4 00331 A;A,A;02 6 16 96
23 A~D4 02220 A,D4AjA~02 8 12 96
23 A~D4 31110 D4A!A~O, 8 12 96
23 A~D4 04111 A,A;A~AIO, 24 4 96
23 A: 011111 A4Aj0 4 12 40 80
23 A: 001234 A;A~A~Aio3 60 8 80
23 A~ 0424 A;A~O, 14 384 64
23 A~ 0'21'3 A~A~Aio4 112 48 64
23 Ail 0·1· A~Atos 264 1440 48
23 Ai4 0 1.1. A~607 759 645120 32
23 A24 min O2, 196560 84610842624000 0
418 Chapter 16

Case (4): v = Vi is a vector of the form (± 14, Ok-4) in a component

Dk . In this case the maximal Dm is the D 4 supported in the four nonzero
coordinates of v, and the reduced lattice is 20-dimensional. Again the
symbol contains O's except for a + in position i. We remark that the case
k = 4, when there is a component D 4 , is rather special because all three
nontrivial cosets of D4 in its dual contain vectors of norm 1 which can
serve as the glue vector [2). Fortunately, in both the cases N = D~ and
N = AffD 4 , Aut(N) contains elements permuting all three cosets of any D4
component, so we may suppose v = (± 14, Ok-4).
Cases (~5): v = Vi is the vector (2, Ok-I) in a Dk (k ~ 5) or any
norm 4 vector in E 6, E 7 or E g. The maximal Dm in these four cases is Dk
Ds, D6 or Dg respectively, and the symbol has a - in position i and O's
elsewhere.
Inequivalent lattices. The above remarks make it easy to determine when
two lattices constructed in this way are equivalent. Each lattice An is
specified by a Niemeier lattice
N = W\W2 "'Wo

and a symbol

which is either a glue word or a permutation of **00 - 2, +0 0 - 1 or -00 - 1•

Two lattices are equivalent if and only if they are specified by the same N
and their symbols are equivalent under Aut(N). The components VI>
V 2,,,. of An are determined by the Wi and d i as shown in Table 16.8.
The automorphism group of An. To compute the order of the
automorphism group of An we argue as follows. The lattices An, An+I>'"
of a chain whose reduced lattice is An have
1Aut(A n+m ) 1 = 1Aut(A n) 1'2 m m! ,

so that it suffices to compute 1Aut(A 23 ) I. Now a 23-dimensional lattice

A 23 is completely determined by the associated Niemeier lattice Nand
either of the norm 4 vectors ± v of An n N. Hence

1Aut(A 23 ) 1 = 21 ~~:\N) 1 , (5)

where Aut (N) is described in §3 and c(v) is the number of images of v

under Aut(N). This number is easily computed as the product

where w - d\d 2 "'do is the symbol for AN, c(w) is the number of images
of wunder Aut(N), and c(di ) is the number of choices for the component
Vi of the vector v which would lead to the digit di . For example c(*) is
the number of minimal vectors in Wi' The numbers c (di ) can be found by
Enumeration of Unimodular Lattices 419

Table 16.8. Each lattice A. is specified by the components WI, W 2.... of

a Niemeier lattice N and a certain symbol w - did 2 •• ' . This table
gives the components V" V 2.... of A. corresponding to each component W
of N and each digit d of w. as well as the numbers c(d) defined in §4.
The six parts of the table correspond to the cases W = A.. D 4 •
D.(n ~5). E 6 • E7 and Eg.

An D,

d V,V,·· . c(d) d V,V," . c(d)

0 An 0 D,
A, 8
A;_,A n_; r;l)
A,
A, 8
An-, 2(n~l)
A~ 24
+ A,,_4 + 24
6r;l)

Dn(n;.5) E. E7 E.
d V,V,·· . c(d) d V,V,·· . c(d) V,V,·· . c(d) V,V,·· . c(d)

0 Dn 0 E. 1 E7 1 E.
1 A"-l 2"-1 1 D, 27 E. 56
D n_,
2 D, 27
2 2n A, 72 D. 126 E7 240
A"_l 2n-' 270 A, 756 2160

D"-2A l 4G)

+ D n_, 16(;)
2n

elementary counting arguments and are given in Table 16.8. The values of
c(w), which are usually small. were found by careful consideration of the
group action, and are given in Table 16.7.
The orders of the automorphism groups of the lattices An in Table 16.7
were calculated from (5) and (6). The lattices with minimal norm 1 were
then found by forming direct sums An 9 1m , and the mass constants were
checked against the formulae given in Theorem 1. (This completes the
formal proof of Theorem 5,) The numbers of lattices found in each
dimension are shown in Table 2.2.
The fifth column of Table 16.7 gives the number gl (An) g2(A n )
= 1G I (An) 11G 2(A n ) I, and provides the reader with an alternative and
easier method of computing 1Aut (An) 1 via the formula (cf. Eq. (I))
k
1Aut(A n ) 1= II 1Go(V;) 1. gl (A n )g2(A n ).
;-1
420 Chapter 16

Here VI •...• V k are the components of An (given in the column headed

V), and the values of IGO(Vj) I are given in Eq. (2). For example the
penultimate line of Table 16.7 refers to the lattice A 16 0 7, for which
IAudAn) I - (20 16 '17'645120.

The values of glg2 were actually "back-computed" from (5), (6) and (7)
after the mass formula check had been applied. We note that
(VI •...• Vk ) identifies An uniquely. The final column of Table 16.7 gives
the number (t2) of norm 2 vectors in An.
The Leech lattice. It remains to consider the unique Niemeier lattice with
minimal norm greater than 2, the Leech lattice. In this lattice there is just
one orbit of norm 4 vectors (Theorem 27 of Chap. 10). Our construction
produces a certain 23-dimensional lattice with minimal norm 3, the shorter
Leech lattice 0 23 , which we have already encountered in the Appendix to
Chap. 6. It appears in the last line of Table 16.7, and is the only lattice in
the table with minimal norm 3.
Acknowledgements. The extensive arithmetical calculations needed to
prove Theorems 4 and 5 were performed using ALTRAN [Bro14] and
MACSYMA [Mat3].
 17
The 24-Dimensional
Odd Unimodular Lattices
R. E. Borcherds

This chapter completes the classification of the 24-dimensional unimodular

lattices by enumerating the odd lattices. These are (essentially) in one-to-
one correspondence with neighboring pairs of Niemeier lattices.

1. Introduction
The even unimodular lattices in 24 dimensions were classified by Niemeier
[Nie2] and the results are given in the previous chapter, together with the
enumeration of the even and odd unimodular lattices in dimensions less
than 24. There are twenty-four Niemeier lattices, and in the present
chapter they will be refered to both by their components D 24 , D I6 E g , ••• "
and also by the Greek letters a, {3, ... , the Leech lattice being denoted by
w (see Table 16.1).
The odd unimodular lattices in 24 and 25 dimensions were classified in
[Bor! J. In this chapter we list the odd 24-dimensionallattices. Only those
with minimal norm ~ 2 are given, i.e. those that are strictly
24-dimensional, since the others can easily be obtained from lower
dimensional lattices (see the summary in Table 2.2 of Chap. 2)
Neighboring lattices. Two lattices A and B are called neighbors if their
intersection A n B has index 2 in each of them ([Kne4], [Ven2D.
Assume for simplicity that A is unimodular. Take any vector x
satisfying
XEV2A\A, x· x E Z, (1)

and define
422 Chapter 17

Ax == {a EA : a·x EZ},

Then it is not difficult to show that Ax = A n B, B is unimodular, A and

B are neighbors, all neighbors of A arise in this way, and x, x I produce
the same neighbor if and only if x == x I (mod A).
An equivalent construction is the following. Define the integral part of
a lattice L to be
[L] = L n L' = {tEL: t· L !;;; Z}.

If x satisfies (I) then

B==[<A,x>]

defines the same neighbor as (3).

For example, if A = ZS and x = (~)S, we obtain B = Es.
As a second example, let A be the Niemeier lattice A [4, which we take
in the form obtain~d by applying Construction A to C€24 (see §I of the
previous chapter). Any a E Af4 has the shape 8- 1/ 2 (2c+4z) for
c E C€24, Z E Z24. There are just two inequivalent choices for x. If
x == 8- 1/ 2 (2 24) then B is the odd Leech lattice 0 24 , with minimal norm 3
(see the Appendix to Chap. 6). If x = 8- 1/ 2 (_6,2 23 ) then B is the Leech
lattice A24 itself. In fact A [4 is the unique even neighbor of A24 (see
Fig. 17.0.
Eq. (5) can be used to construct new lattices even if x does not satisfy
(0. For example many years ago Thompson [Th07] showed that the
Leech lattice can be obtained from D24 by
(6)

where

But since the intersection A24 n D24 has index 47 in each of them, A24
and D24 are not neighbors in our sense.
The enumeration of the odd 24-dimensional lattices. Figure 17.1 shows
the neighborhood graph for the Niemeier lattices, which has a node for
each lattice.
If A and B are neighboring Niemeier lattices, there are three integral
lattices containing A n B, namely A, B and an odd unimodular lattice C
(cf. [Kne4]). An edge is drawn between nodes A and B in Fig. 17.1 for
each strictly 24-dimensional odd unimodular lattice arising in this way.
Thus there is a I-I-correspondence between the strictly 24-dimensional odd
unimodular lattices and the edges of our neighborhood graph. The
The 24-Dimensional Odd Unimodular Lattices 423

n=B n= 16 n = 24

(0

Figure 17.1.
Neighborhood graphs of the
even unimodular lattices in
n - 8, 16 and 24 dimensions.
424 Chapter 17

Table 17.1a. Odd unimodular 24-dimensional lattices

Components g,g, Components g,g,
I 0,. w>/> 212 M,. 27 A,(A,A,)AtA,O. uu 8' 2
2 Afo l6 10321920·2 28 Al (A,A,)A,(A ,A ,)0, uu 4·2
3 A f4 """
>/>x 190080 29 A5A1(AIA/lA,o, uu 12·2
4 A,'·0, >/>¢ 43008 30 D.AtAtO. uu 16' 2
5 AjA IOolO xx 2880· 2 31 D.A.Afo. uu 12·2
6 Ar4 >/>T 138240 32 A; ¢~ 384
7 AtAlo, x¢ 384 33 AIA.A,AiAr03 V7r 4
8 AfA1o. xv 240 34 AIAto. V7r 16
9 AfAI'O. ¢¢ 384' 2 35 A5A.A,A,A,A,A,O. up 2
10 A,AtA10, ¢¢ 48 2 36 AtAtO. up 24
II A!O, ¢¢ 336' 2 37 AlA103 vo 12
12 A!AtO. xu 384 38 D1Ato. T7r 32
13 AfAtAtO. ¢v 16 39 A5A1A,A 3A,O. U7r 4
14 AtAlo. ¢T 384 40 A~AjArArArO, U7r 8
15 AtAtAtO. ¢u 48 41 A5D.AjA,ArA,O, U7r 4
16 AIAfA,Aro, ¢u 16 42 D.AIA.Aio. U7r 4
17 A.A,AtAtO, vv 8·2 43 A.A1Aio, pp 6·2
18 AtAiAro. vv 16·2 44 A~AfAio. pp 4·2
19 AtAtO. ¢p 24 45 A5A1Alo. pp 6·2
20 AtO. VT 240 46 AtAfo, vv 16
21 AlA tA to. vu 16 47 A~A.A,AiO, uo 4
22 A.AfA,AiArO, vu 4 48 A.A.A.A,A,A,A,03 p7r 2
23 AtA fA to, ¢7r 32 49 A~AIAro. p7r 4
24 AIAfAiAro. vp 4 50 DtAI T~ 48
25 DIAl6 TT 576·2 51 A~D.AfArO, u~ 8
26 D.AtAtO. TU 48 52 A5A5AIA,o, uv 4

156 lattices are shown in Table 17.1. Fig. 17.1 also shows the
corresponding graphs for dimensions 8 and 16.
For each lattice A in the table we give its components (in the notation
of the previous chapter) and its even neighbors (represented by two Greek
letters). The final column gives the product glg2 of the orders of the
groups G\(A), G 2 (A). We may write Aut (A) = Go (A) .N(A). T(A),
where T(A) has order 1 or 2 and interchanges the two neighbors of A if it
has order 2. Then IN(A) T(A)I = g\g2, and the last column is written
either as n . 2 if T (A) has order 2, or as n if T (A) has order 1, where n is
the order of N (A). The components are written as a union of orbits under
N (A), with parentheses around two orbits if they fuse under T (A).
The 24-Dimensional Odd Unimodular Lattices 425

Table 17.1 b. Odd unimodular 24-dimensional lattices

Components g,g, Components g[g2

53 A?D,ArAtO, uv 4 79 D?A?AtO, 7r1' 8

54 A 6A 5A,A 3A,A,O] po 2 80 A6A 6D 5A ,03 7rA 2
55 A 6D, (A ,A ,) A,O, 7r7r 2· 2 81 A7D5D,A]A]O, 7rA 2
56 D 5A,(A,A,) (A,A,)O, 7r7r 2· 2 82 D5D 5A l A tO, 7rA 4
57 A 7(Ar A r)AfO, 7r7r 8· 2 83 A 7D5A 5A]A,A,A,O, 7rA 2
58 A?DIAtO, 7r7r 8· 2 84 D6DJAi H 6·2
59 D,A 5(A 3A 3)A 3(A ,A ,)A ,0, 7r7r 2· 2 85 D6A?ArO, ~v 4
60 D5D,AtO] 7r7r 8· 2 86 A,(A 5A,)A]0] vv 2- 2
61 AtO, p~ 24 87 DoAlArO, vv 4· 2
62 AlASAio, pv 4 88 A1D,(AtA IlO, vv 4- 2
63 A 6A,A 5A,A,O, pv 2 89 A1Alo, 7rK 4
64 AtAf UI' 48 90 A,A 6A 5A,A,O, OA 2
65 A?A 5D,A]A,O, UA 4 91 A1Dlo, 7rl 8
66 AlA,A 3AtO, 7r0 4 92 A 7DlArO, 7rl 4
67 A7 A I A so ] 7r0 4 93 A,A7A,A 3O, OK 2
68 D5AlD,AtO, 7r~ 4 94 Mo, ~I' 48
69 D?AtO, 7r~ 16 95 D6D5A~0] P 4
70 A 6D 5A,A,A,O, 7rV 2 96 A,A 6D 5A,O] VA 2
71 A 7A,D,A]A,A,A,O, 7rV 2 97 A 7D6A 5A]A,O, VA 2
72 A7A 5A IA ,o] 7rV 4 98 A 9A,D,A]A,A,O, VA 2
73 A7 A l A io] 00 4·2 99 Aj0 3 01 12
74 D1 TL 48 100 A1D 5D,O, 7r8 4
75 A 6A 6A 5A,A,O, PA 2 101 A9A 6A 5A ,O, VK 2
76 A 7A 6A,A,A,O] ov 2 102 AlAro, VK 4
77 A,AIAso, ov 4 103 A1D6AtO, VI 4
78 AlA?O, PK 4 104 DlD1Af ~I 4

The first lattice in the table is the odd Leech lattice 024. The
remaining lattices have minimal norm 2, the number of norm 2 vectors
being given by the formula
8(h(A) + h (B)) 16, (8)

where h (A), h (B) are the Coxeter numbers of the even neighbors of the
lattice.
426 Chapter 17

Table 17.lc. Odd unimodular 24-dimensionallattices

Components gig, Components gIg}

105 AtD? 7r~ 8

4
131
132
D,AjO, Til
KI;
4

2
106 AgAtO, 08 A!2A 7A 40 ]
107 E6D5A~AfOl 1'1. 4 133 DsDlA[Ar ,~ 2
108 E,(A 6A 6 )A,O, AI. 2· 2 134 A 11 D 7 D 5()j A, 2
109 A 7E,D,(A 3A 3)0, AA 2· 2 135 Arl O2 4
"
110 A 9(D,D,) (A ,A ,)A ,0, AA 2· 2 136 D 9 AjO l e~ 4

III /)7(A,A,)A,A ,0, AA 2· 2 137 A 13[)t.A3A 10\ III; 2

112 A IOA 6A,A,O, AK 2 138 DJDl " 2

113 A,A,A,O, v8 2 139 E7 D oD6D 4 A I ~~ I ·2

114 A,D 6 A,A]0, vO 2 140 A9E 7A 70 , ~I; 2

115 AtiA 7O, 01; 4 141 A 12 A 11 O] 60 2
116 D 7A 7D,A]O, A, 2 142 A ll[),A,O, 0, 2 I

117 AiAj0 2 XII. 4 2 143 DIODtA ? w 2

i
118 DlD6D 4 A r h 2 144 A15 A gO ] 00 2
145 /)~ '1 6
119 A,A 7D,A,0, v~ 2
146 D~Dg ,j! 2
120 A,A7D6A ,0, vI; 2
147 A 16As7O, 1;0 2
121 A llD",D 4A 301 1.0 2
148 A I5 [)'S0j Ol! 2
122 A9D7A SA 1A 01 J AH 2
149 DsF?Ar 2
D8D t
~1
123 4·2
" ISO D\Or.'7 D 6A I ~{3 I
124 AllAgA 1 0, KO 2
lSI A15E7AsIO, I(l 2
125 El/)~o, I'~ 8
152 D 12 [),D, <(3 I
126 A,E 6 /)6 A ,O, A~ 2
153 ExDg -y(J 2
127 /)7 A 7E 6A 30 , A~ 2
154 Dr2 w 2
128 A lOE 6 A 6 O, AI; 2
155 AnD) "IX 2
129 A1I D 6A SA 101 AI; 2
156 D16 D S {3" I
130 Dt ~, 8 -- ..- - -
 18
Even Unimodular 24-Dimensional Lattices
B. B. Venkov

Niemeier's classification of even unimodular 24-dimensional lattices is

simplified. The methods involve the theory of modular forms, algebraic
coding, and root systems.

1. Introduction
We consider even unimodular Euclidean lattices A in Rn or, in more
classical terminology, classes of integral, positive definite, even quadratic
forms of determinant 1 in n variables. Such lattices exist only if n = 8k.
A complete classification of such lattices is known only for k = 1,2,3.
When k = 4 there are already more than 80 million lattices (see
Table 16.5.). When k = 1 or 2 the classification is very simple: there are
precisely one 8-dimensional and two 16-dimensional lattices of the above
type. A complete classification of the 24-dimensional lattices was obtained
in 1968 by Niemeier [Nie21. According to Niemeier, there are precisely
24 classes of even unimodular lattices. Each lattice A is uniquely
determined by its set of minimal vectors A (2) = {A E A : (A, A) = 2},
which form root systems of the following types:
121,

24Aj, 12Az, 8A 3, 6A 4, 4A 6 , 3A g, 2A 12 , A 24 ,
6D 4, 4D 6 , 3D g, 2D 12 , D 24 ,
4E 6 ,3E g,
4As+D4, 2A7+2Ds, 2A 9 +D 6, A 1S +D 9, E g +D 16 , 2E 7+D IO , E 7+A 17 ,

The lattice corresponding to the empty root system is the Leech lattice.
Niemeier obtained his classification by Kneser's method [Kne41, which
involves studying lattices that are "neighbors" of a given one (see the
428 Chapter 18

previous chapter), requires extensive calculations, and in no way explains

the appearance of the strange list (1) of root systems. Here we offer
another approach to the classification of the even unimodular 24-
dimensional lattices, based on an a priori proof of the fact that the root
system of a 24-dimensional lattice is one of the systems in (I) and that
each root system in (1) can be realized in one and only one way. In §2 we
prove that the root system A (2) of an even unimodular lattice A in R24
possesses the following properties: (i) either A (2) = 0 or rank A (2) = 24;
(ii) all irreducible components of A (2) have the same Coxeter number h;
(iii) !A (2)! = 24h, where !A (2)! is the number of elements in A (2). This
is proved by means of modular forms (using theta series with spherical
coefficients). It is easy to show that the only root systems possessing
properties (i) - (iii) are those in the list (I).
We then show that each root system in (I) can be realized as A (2) for
a unique lattice. This is actually a problem in algebraic coding theory
(see §3), and the existing literature on algebraic coding quickly enables us
to obtain the final answer (§4). The last section is devoted to the Leech
lattice.

2. Possible configurations of minimal vectors

Suppose A C R24 is an even unimodular lattice, A (2n) =
{A E A : (A, A) = 2n}, and A (2) is the root system of A.
Proposition 1. If a E R 24 , then

~ (y, a)2 = _1 (a, a)!A(2)!.

yEA(~ 12

Proof We use the classical result (see [Hec2] or [Oggl, Chap. VI]) which
says that, for an even unimodular lattice A in V = Rf, the theta series
with spherical coefficients
f ..p=~[ ~ P.(x»)qn, q=e 27riz ,Imz >0,
n xEA (2n)

is the Fourier expansion of a modular form for PSL (2, Z) of weight

1/2f + v. This form is a cusp form if v > O. Here p. is a homogeneous
polynomial of degree v which is a A-spherical function, i.e. ~A p. = 0,
where ~A is the Laplace operator in a coordinate system in V in which the
inner product (,) reduces to a sum of squares. The 2-dimensional A-
spherical polynomials have a very simple form: they are linear
combinations of polynomials

P 2,,,(x) = (x, a)2- ~ (a, a)(x, x),

where a E V (see [Hec2]). Thus, for any a E V,

Even Unimodular 24-Dimensional Lattices 429

is a cusp form of weight 1/2f + 2 for the full modular group. In particular,
for f = 24 the series (3) is a cusp form of weight 14; but such a form is
equal to zero, since any modular form of weight k is a polynomial in two
modular forms, E 2 and E 3, of weights 4 and 6, and the cusp forms in this
polynomial ring constitute the principal ideal generated by the unique
parabolic form ~ of weight 12 (see [Ser I]). Thus
1
~ (x, a)2 = "242n (a, a) IA (2n) I,
xEA(2n)

which for n = 1 yields (2).

Corollary 1. Either A (2) = 0 or rank A (2) = 24.
Proof Indeed, suppose rank A (2) < 24, i.e. the vector subspace generated
by A (2) has dimension < 24. Then R24 contains a vector v ¢ 0
orthogonal to all of A(2). Putting a = v in (2), we see that IA (2) 1 = o.
Corollary 2. All irreducible components of the root system A (2) have the
same Coxeter number h, and IA (2) 1 = 24h.
We now make use of the following property of the Coxeter number [Boul,
Chap. VI, §l.ll, Proposition 321: if in an irreducible root system R all
roots have the same length, then the number of elements of R not
orthogonal to a fixed a E R is equal to 4h - 6. We take the usual
normalization for an invariant inner product, where (a,a) = 2 for a E R.
N ow if a, a' E R, then (a, a') can assume only the values 0, ± I, ± 2, and
(a,a') = ±2 only if a' = ±a. Therefore 4h - 6 = 2 + 2{3(R), where
(3(R) is the number of roots a' E R such that (a',a) = I for a fixed a.
This does not depend on the choice of a, since the Weyl group W(R) acts
transitively on the roots. Thus (3(R) = 2h (R) - 4.
Proof of Corollary 2. If a E A (2), then, according to (2).
I
~ (y, a)2 = "6IA (2) I,
yEA(2)

It is clear that only the elements of that irreducible component Ra

containing a make a contribution to the left-hand side of the equality, and
that
~ (y, a)2 = 2 .2 2 + 2{3(Ra).
yEA (2)

Since (3(Ra) = 2h (Ra> - 4, it follows that IA (2) 1 = 24h (Ra>, as required.

Thus we have obtained the following result.
Proposition 2. If A (2) ¢ 0, then the root system A (2) in R24 possesses
the following properties:

(i) rank A (2) = 24,

(ij) all irreducible components of A (2) have the same Coxeter number h,
(iij) IA (2) 1 = 24h.
430 Chapter 18

Conditions (i)-(iii) are very strong, and it is not difficult to find all root
systems in R24 satisfying them.
Proposition 3. Suppose R is a (possibly reducible) root system in R24
whose roots all have the same length. Suppose also that R satisfies
conditions (i)-(iii) of Proposition 2. Then R is isomorphic to one of the
23 nontrivial root systems in (1).

Proof The components of R can be Ai, Dj or E k . Let

24 24 8
R = ~aiAi + ~(3jDj + ~'YkEk'
i-I j-I k-6

We take the values of the Coxeter numbers from Chap. 4:

h(At) = I + 1, h (D t ) = 21 - 2, h (E 6 ) = 12, h (E 7) = 18, h (E 8) = 30.

Since all components of R must have the same Coxeter number, it follows
that at most one a, at most one {3, and at most one 'Y can be different from
zero. Moreover, if R contains components from the A and D series, i.e.
ai ;z!' 0 and {3j ;z!' 0, then i = 2j - 3. Similarly, if ai ;z!' 0 and 'Yk ;z!' 0,
then i = 11 if k = 6, i = 17 if k = 7, and i = 29 if k = 8. Finally, if
{3j ;z!' 0 and 'Yk ;z!' 0, then j = 7 if k = 6, j = 10 if k = 7, and j = 16 if
k = 8.
Thus R has the following form:
either R = aiAi for some i, (4)
or R = {3jD j for some j, (5)

or R = 'YkEk for some k, (6)

or R = a2j-3A2j-3 + {3jDj for certain {3j ;z!' 0, a2j-3 ;z!' 0, (7)

or R = allA ll + {37D7 + 'Y6E6, 'Y6 ;z!' 0, all or {37 positive, (8)
or R = al?Al? + {3 IOD IO + 'Y7E7, 'Y7 ;z!' 0, al7 or {310 positive, (9)
or R = a29A29 + {316D16 + 'Y8E8, 'Y8 ;z!' 0, {316 positive. (10)

Condition (j) of Proposition 2 means that

~ (;ai + j{3j + k'Yk) = 24.
iJ,k

This yields the following possibilities for R. In case (4):

24AJ, 12A 2, 8A 3, 6A 4 , 4A 6 , 3A 8, 2A l2 , A 24 .

In case (5):
Even Unimodular 24-Dimensional Lattices 431

In case (6):

In case (7), condition (j) becomes j{3j + (2j - 3)a2j-3 = 24, j ~ 4, {3j
and a2j-3 ~ 1 integers. It is easy to see that the indeterminate equation
j{3 + (2j - 3)a = 24, j ~ 4, a ~ 1, {3 ~ 1, has only the following
solutions: (j, a, (3) == (4,4, 1), (S, 2, 2), (6,2, 1), (9, 1, 1). This leads to
the following possibilities for R in case (7):
4As + D 4, 2A7 + 2D s, 2A9 + D 6, A 1s + D 9.
In case (8), condition (j) becomes llall + 7{37 + 6')'6 = 24, which admits a
unique possibility for R:

In case (9) we have 17a17 + IO{31O + 7')'7 = 24, which admits two
possibilities:

Finally, in case (10) we obtain 16{316 + 8')'8 = 24, which yields

E8 + D 16·
This proves Proposition 3.
Note that we did not use property (iii) in the proof of Proposition 3.
Since all of the root systems obtained above are realized as A (2) for even
unimodular lattices A, a fact which will be proved in §4, it follows that (iij)
is a consequence of (j) and (jj), i.e. if a 24-dimensional root system
possesses properties (j) and (ij) of Proposition 2, then it also possesses
property (jij).
Corollary. Suppose A is an even unimodular lattice in R24. Then either
A (2) = flJ or A (2) is one of the 23 nontrivial Niemeier root systems listed
in (1).

3. On lattices with root systems of maximal rank

The aim of this section is to make several general observations about even
unimodular lattices having a root system of maximal rank and about the
connection between such lattices and algebraic codes. In what follows we
will consider only those mot systems in which all roots have the same
length, i.e. those root systems whose irreducible components have types
A,D,orE.
Let us consider the following construction. To each irreducible root
system R ¢ E8 we associate a triple (T(R), G(R), IR) consisting of a
finite Abelian group T(R), a finite group G (R) acting on T(R), and an
432 Chapter 18

R-valued G(R)-invariant function IR on T(R). We define this triple as

follows.
1. If R = Ai, then T(A i ) = Z/(i + Oz = {O, 1, ... , i}, and G(A 1) = 1,
G (Ai) = Z/2Z for i > 1. A nontrivial element u E G (A), i > 1, acts on
T(A i ) like multiplication by -1, i.e. u(k) = i + 1 - k. Also
IAj(k) = Hi + 1 - k)/(i + O.

2. If R = D j , j ~ 4, then

T(D.) = {d d d d} = {z/2Z EB Z/2Z if j==O (mod 2),

} 0," 2, 3 Z/4Z if j==l (mod 2).

The notation is chosen so that do is the zero element of the group T(Dj ),
and di == i (mod 4) if j == 1 (mod 2). Next, G(D4) = S3, G(Dj ) = Z/2Z
for j ~ 5. The action is the following: a nontrivial element u E Z/2Z fixes
do and d 2, and ud l =- d 3• When j = 4, G(D4) = S3 = GL(2,2) coincides
with the full automorphism group of T(D 4 ). The function IDoJ is defined as
follows: 1(do) = 0, [(d 2) = 1, [(d I) = [(d 3) = j /4. In the case where
j == 0 (mod 2), it is sometimes convenient to identify T(Dj) with the
additive group of the field F 4; G (Dj ), j > 4, can then be identified with
the Galois group GaI(F4 / F 2).

3. If R = E 6 , then T(E 6 ) = Z/3Z = {O, 1, 2}, G(E 6 ) = Z/2Z = {u},

u (I) = 2, and [(0) = 0, I (I) = [(2) = 4/3.
4. If R = E 7, then T(E 7 ) = Z/2Z = {O, I}, G(E 7 ) = 1, and
1(0) = 0, I (I) = 3/2.
The invariant meaning of this triple is the following: T(R) = P/Q,
where Q is the root lattice and P the weight lattice for R; G (R) is
isomorphic to the quotient group of the automorphism group A (R) of the
root system R by the Weyl group W(R) with the natural faithful action of
A (R) /W(R) on P /Q; I is the norm of the microweight lying in a given
coset with respect to Q. Recall that an element pEP is called a
microweight if (p, r) = 0, ± 1 for all r E R. In each coset in P /Q there is
precisely one orbit under the Weyl group W(R) consisting of
microweights, and I on a coset with respect to Q is the norm of any
microweight lying in this coset.
If a root system consists of n isomorphic components ~ E 8, i.e. has the
form nR, then we put T (nR) = T (R) n, we define G (nR) to be the
natural semidirect product

(the wreath product of G(R) by Sn) with the natural monomial action on
T(nR), and we define I by additivity:

[(x" ... , xn) = ~[(Xi).

Clearly I is G (nR) -invariant.

Even Unimodular 24-Dimensional Lattices 433

Finally, for an arbitrary root system R = ~ njRj, where R j are the

irreducible components, we define our triple as follows. We discard the
components E 8 and put
T(R) = Ell T(njR), G(R) = IIG(njR j), IR = ~/njRi'
' i i

where the summation extends over all types of irreducible root systems
other than E 8• We call a subgroup A C T(R) even and self-dual if
IAI2 = IT(R)I and the function IR assumes even integral values >2 on
A \ {O}. There may be no such subgroups, but if they exist, the set of
them is G (R) -invariant, since the function IRis G (R) -invariant.
Proposition 4. Suppose R is a root system of rank n. There exists a
natural one-to-one correspondence between the classes of even unimodular
n-dimensional lattices (to within isomorphism) with root system
isomorphic to R and the orbits of even self-dual subgroups A C T(R)
relative to the group G (R).
The proof amounts to a simple verification and is omitted. The
correspondence assigns to a lattice A the quotient A = A/ Q (A (2)) with
respect to the sublattice generated by the roots, which is viewed as a
subgroup of Q (A (2))o/Q (A (2)) = P/Q.
Example 1. Suppose R = nA I is the simplest root system of rank n. In
this case the group T (R) can be identified with the space of sequences of
length n over the field F 2, and T(R) = (Z/2z)n. The group G(R) is
isomorphic to Sn, and the function I is given by [(01) = [(011, ... ,an) = V2
wt (01), where wt (01) is the Hamming weight (§2.1 of Chap. 3). An even
self-dual subgroup is a binary self-dual Type II code (§2.2 of Chap. 3)
with minimal distance ~ 8. The classification of such codes under the
action of Sn is a famous problem of binary coding theory (see §6 of
Chap. 7). Thus this same problem is equivalent to that of classifying the
even unimodular n -dimensional lattices with root system nA I.
Example 2. Suppose R = nA 2 • In this case T(R) = (Z/3Z)n,
G(R) = (Z/2z)n'Sn, 1(01) = 1(011, ... ,an) = (2/3)' (the number of
nonzero OIj), and, exactly as above, the problem of classifying the 2n-
dimensional lattices with root system nA2 is equivalent to that of
classifying the ternary self-dual Type III codes with minimal distance > 3
(§2.2 of Chap. 3, §9 of Chap. 7).
In the general case, i.e. for any root system R, the problem of
classifying the even self-dual subgroups of T (R) relative to G (R) can also
be viewed as a problem of coding theory. This problem includes and
differs insignificantly from that of classifying the self-dual codes over all
rings Z/nZ and the field F 4 • An equivalent problem is to classify the even
unimodular lattices having a root system of maximal rank. In the sequel,
self-dual subgroups of T(R) will often be called codes.
Remarks 0) The usual concept of isomorphism in considering codes over
commutative rings is defined by monomial matrices (§2.2 of Chap. 3).
Proposition 4 shows that to classify lattices with root system of maximal
434 Chapter 18

rank of type nR (where R is an irreducible root system) in most cases it is

necessary to classify codes relative to monomial groups in which the
nonzero elements are equal to ± I. Only lattices of type nD4 lead to
complete monomial groups.
(iD The classification of lattices with root systems of maximal rank of type
nD2k depends very weakly on k: in the corresponding code only the
function I depends on k. This enables us to reduce the problem of
classifying such lattices of type nD2k to the case of small k. An analogous
remark holds for other cases of isomorphisms between the T (R j ). For
example T(D 2i + l ) = T(D s) = T(A 3 ), T(E 6 ) = T(A 2), and
T(E 7 ) = T(A 1)·
(iii) Codes corresponding to lattices of maximal rank and of type nD 2k are
not necessarily linear codes over F4; that is, self-dual subgroups of
T(nD 2k ) = T(D 2k )n = F3 need not be invariant under multiplication by
scalars of F 4. For example, this occurs in the 24-dimensional lattice with
root system 3D8 (see case XI in §4); if the corresponding self-dual
subgroup were a linear subspace, it would have dimension 312.

4. Construction of the Niemeier lattices

We are now in a position to prove the main theorem.
Theorem (Niemeier [Nie2D. To within isomorphism there exist precisely
24 even unimodular lattices in R24. Each lattice is uniquely determined
by its root system. The possible root systems are the 24 listed in (j).
Proof We already know from §2 that the root system of a given even
unimodular lattice A C R24 belongs to the list (I). Therefore it suffices to
prove the existence and uniqueness of a lattice with each of the 24 root
systems in (I). The case of the empty root system is considered in §5. All
of the other root systems have maximal rank 24; hence they come within
the scope of the theory of §3. Thus for each of the root systems in (I) it is
necessary to calculate the triple (T (R), G (R), I R) and verify the existence
and uniqueness of the corresponding code. We will number the root
systems and lattices in the order in which they appear in (j).
I. R = 24A 1. Then T(R) = (Z/2Z) 24, G(R) = S24, [(a) = 1/2wda), and
A C T(R) is a binary self-dual Type II code with minimal distance 8.
The Golay code rc24 (§2.8.2 of Chap. 3) is the unique such code (Pless
[Ple8]; see also [Rasll or §7 of Chap. 7). Its automorphism group is the
Mathieu group M24 (Chaps. 10, 11), which is 5-transitive on the
components of R. Thus there exists a unique lattice r I with root system
24A 10 and the automorphism group of r l is the extension

II. R = 12A 2 . Then T(R) = (Z/3Z)12, G(R) = (Z/2Z)12 . S12, and A

must be a ternary self-dual Type III code with minimal distance> 3. The
Golay code rc 12 (§2.8.5 of Chap. 3) is the unique such code (Pless [Ple8];
Even Unimodular 24-Dimensional Lattices 435

see also §10 of Chap. 7). Its automorphism group is isomorphic to 2 . M J2 ,

and acts 5-transitively on the components of R.
III. R = 8A 3 • Then T(R) = (Z/4Z)g and G(R) = (Z/2Z)g· Sg. If we
identify Z/4Z with the numbers 0, ± 1, 2, then 1 has the form
1(0) = 0, [(±)) = 3/4, 1(2) = 1. Thus in the space of sequences
a = (ai, ... , ag), ai = 0, ± 1, 2, we must look for subgroups A of order
44 such that [(a) = ~ [(ai) E 2Z and [(a) > 2 for all a E A \ {OJ. A
direct calculation shows that for such an A we can take the subgroup
spanned by the vectors
Xl = (0, 0, 0, 1, 1, 1, 1, 2),
X2 = (0,0,1,0,2, 1, -1, )),
X3 = (0, 1,0,0,1, -1, 2, )),
X2 = (I, 0, 0, 0,1,2, -1, -)),

and such a subgroup is unique (to within G (R), of course). It can be

shown that the permutation part of the group Aut A is isomorphic to
GA (3,2) (the full affine group of order 3 over the field F 2, having order
1344) and that the elements of order 2 of A form a binary Hamming code.
The automorphism group of the lattice r acts 3-transitively on the
components of R.
IV. R = 6A 4 • Then T(R) = (Z/5Z)6 and G(R) = (Z/2Z)6. S6' If we
identify Z/5Z with the numbers 0, ± 1, ± 2, then the function 1 has the
form 1(0) = 0, [(±)) = 4/5, [(±2) = 6/5. In the space of sequences
a = (ai, ... , (6), ai = 0, ± 1, ± 2, we must find 3-dimensional subspaces
A C T(R) such that [(a) = ~[(a) for a E A \ {oJ is even and >2.
Elementary calculations show that for A we can take the subspace spanned
by Xl = (0,0, 1, 1, -2, -2), X2 = (2, -2, 0, 0,1, )), X3 = (1, 1,
2, -2, 0, 0), and that A is unique to within the action of G (R).
V. R = 4A 6 • Then T(R) = (Z/7Z)4 and G(R) = (Z/2Z)4. S4' If we
identify Z/7Z with ±1, ±2, ±3, then 1(0) = [(±)) = 6/7, [(±2) = 10/7
and [(±3) = 12/7. We must look for 2-dimensional subspaces of T(R) on
which 1 assumes even values > 2. Again such a subspace exists and is
unique. As a representative we can take the subspace spanned by
Xl = (1,2,3,0), X2 = (0,3, -2, )).

VI. R = 3Ag. Then T(R) = (Z/9Z)3 and G(R) = (Z/2Z)3. S3;

1(0) = 0, [(±)) = 8/9, [(±2) = 14/9, [(±3) = 2 and [(±4) = 20/9.
The desired subspace of T (R) exists and is unique (relative to G (R»: it is
generated by Xl = (3,3, 3), X2 = (1, -2, -2), X3 = (-2, -2, )).
VII. R = 2A!2' Then T(R) = (Z/13Z)2 and G(R) = (Z/2Z)2. S2;
[(0) = 0, [(±)) = 12/13, [(±2) = 22/13, [(±3) = 30/13, [(±4) =
36/13, [(±5) = 40/13 and [(±6) = 42/13. The desired code is the 1-
dimensional subspace generated by the vector (I, 5).
VIII. R = A 24 • Then T(R) = Z/25Z, G(R) = Z/2Z, and the desired
code consists of the elements divisible by 5.
436 Chapter 18

IX. R = 6D 4. Then T(R) = F¥ and G(R) = (S3)6. S3. If

a = (al •...• a6) E T(R), then /(a) is the number of nonzero aj. The
desired code is a self-dual Type 4 code over F4 with minimal distance> 2.
The hexacode (§2.5.2 of Chap. 3) is the unique such code ([Mac41, §1O of
Chap. 7). Its automorphism group is 3 . A6 • It is interesting to observe
that this is one of the three known cases (the others are the two Golay
codes) where the automorphism group of an extended quadratic residue
code over Fp is larger than SL(2,Fp) (see [RasI1 and Chap. 10).
X. R = 4D 6 • Then T(R) = Ft and G (R) = (Z/2Z)4 . S4; /(O) = 0,
1(1) = 1 and /(w) = /(w 2) = 3/2, where w is a primitive element of F 4.
The desired code exists and is unique: as a representative we may take the
subgroup spanned by (1, 1, 1, I), (w, w, w, w), (0, 1, w 2, w), (0, w 2 , w, I).
XI. R = 3D 8 • Then T(R) = Fl and G(R) = (Z/2Z)3. S3; 1(0) = 0,
/(I) = 1 and /(w) = /(w 2 ) = 2. As the desired subgroup of T(R) we can
take the subgroup generated by (1, 1, w), (1, w, 1), (w, 1, I). The
uniqueness is very easy.
XII. R = 2D 12 • Then T(R) = Fl and G (R) = (Z/2Z)2 . S2; 1(0) = 0,
I(I) = 1 and /(w) = /(w 2) = 3. The code is generated by (1, w), (w, I).
The uniqueness is obvious.

°
XIII. R = D 24 . Then T(R) = F4 and G(R) = Z/2Z; 1(0) = 0, I (I) = 1
and I (x) = I (x 2) = 6. The code consists of and x. The lattice obtained
is, of course Dt (§7.3 of Chap. 4), or r 24 in the notation of [Serl,
Chap. V, Example 1.4.31-
XIV. R
1(0) = °
= 4E 6 • Then T(R) = (Z/3Z)4 and G (R) = (Z/2Z)4 . S4;
and I (± 1) = 4/3. In the space of sequences a = (aI, a2, a3, a4),
aj = 0, ± 1, we must look for 2-dimensional even self-dual subspaces. The
problem has a unique solution: the subspace spanned by (1, 1, I, 0), (0,
-1, 1, 1) (the tetracode C6'4, §2.5.1 of Chap. 3).
XV. R = 3E 8 • Then T(R) = °
and the code is trivial. The lattice is
isomorphic to E8 E9 E8 E9 E8 (or f8 E9 r8 E9 r8 in Serre's notation).
XVI. R = 4As + D 4 • Then T(R) = (Z/6Z)4 x F4 and G (R) = (Z/2Z)4 .
S4 x S3; /(0) = 0, [(±I) = 5/6, /(±2) = 4/3, 1(3) = 3/2 and /(b) = 1,
b E F4, b ¢ 0. In the space of sequences a = (aJ, a2, a3, a4, b),
aj = 0, ± 1, ±2, 3, b E F 4, we must find a subgroup A of order 72 such
that for all a E A, a ¢ 0, we have /(a) = '1;/(a) + /(b) E 2Z and
/(a) > 2. For such a subgroup we may take that generated by (0, 1, 2,
-1, w), (1, 1, 1,3,0), (3,3,0,0, 1), where w E F4 is a primitive element.
The uniqueness of such a subgroup to within the action of G (R) is easily
verified. (Details are given in [Nie2, p. 163].)
XVII. R = 2A7 + 2D s. Then T(R) = (Z/8Z)2 x (Z/4Z)2 and
G (R) = (Z/2Z) 2 . S2 x (Z/2Z) 2 . S2; the function I on the first two
summands has the form /(o) = 0, /(±I) = 7/8, /(±2) = 3/2,
/(±3) = 15/8, 1(4) = 2, and on the last two it has the form
1(0) = 0, /(± I) = 5/4, /(2) = 1. For the desired subgroup of order 32 we
Even Unimodular 24-Dimensional Lattices 437

may take that generated by (3, 1, 1, 0), (2, 0, -1, I). Uniqueness is very
easy.
XVIII. R = 2A9 + D 6 • Then T(R) = (Z/10Z)2 x F4 and
G (R) = (Z/2Z) 2. S2 x Z/2Z; the function I on the first two summands
has the form [(0) = 0, [(±I) = 9/10, [(±2) = 8/5, [(±3) = 21/10,
[(±4) = 12/5, [(5) = 5/2, and on F4 it has the form [(0) -= 0, I (I) - 1,
I (w) = I (w 2) = 3/2. The desired subspace of order 20 exists and is unique:
as generators w~ may take (5,5, I) (1,2, w).
XIX. R = A 1s + D 9 • Then T(R) = Z/16Z x Z/4Z and G(R)
= Z/2Z x Z/2Z; the function I on Z/16Z has the form
[(0) = 0, [(±I) == 15/16, [(±2) = 7/4, [(±3) = 39/16, [(±4) = 3,
[(±5) = 55/16, [(±6) = 15/4, [(±7) = 63/16, [(8) = 4, and on Z/4Z it
has the form [(0) = 0, [(± I) = 9/4, [(2) == 1. The desired subgroup of
order 8 is generated by the element (2, 1). The uniqueness is obvious.
XX. R == E8 + D 16 . Since the sublattice generated by E8 is unimodular,
the whole lattice is decomposable and is isomorphic to E 8 ED D t6.
XXI. R = 2E7 + D IO • Then T(R) = (Z/2Z)2 x F4 and G (R)
= s2 x Z/2Z; the function I on the first two summands has the form
1(0) = 0, 1(1) = 3/2, and on the last it has the form [(0) == 0, [( 1) = 1,
I (w) = I (w 2) = 5/2. For the desired subgroup of order 4 we may take that
generated by (1, 0, w), (0, 1, w2 ). The uniqueness is obvious.
XXII. R = E7 + A\7. Then T(R) = Z/2Z x Z/18Z and G(R) -= Z/2Z;
the function I on the first summand has the form [(0) = 0, [(I) == 3/2,
and on the second it has the form [(0) = 0, [(±1) =- 17/18,
[(±2) "" 16/9, [(±3) = 5/2, [(±4) = 28/9, [(±5) = 65/18, [(±6) -= 4,
[(±7) = 77/18, [(±8) = 40/9, [(9) = 2. The desired subgroup of order 6
is generated by the element (1, 3). The uniqueness is obvious.
XXIII. R = E6 + D7 + All. Then T(R) "" Z/3Z x Z/4Z x Z/12Z and
G (R) = Z/2Z x Z/2Z x Z/2Z; the function I on the first summand has
the form 1(0) = 0, [( ± 1) = 4/3, on the second the form [(0) = 0,
[(± I) = 7/4, 1(2) = 1, and on the third the form 1(0) = 0,
[(± 1) = 11/12, [(±2) = 5/3, [(±3) = 9/4, [(±4) - 8/3, [(±5)
= 35/12, [(6) = 3. The desired subgroup of order 12 is generated by (1,
I), and the existence and uniqueness are obvious.
Thus we have considered all possible nontrivial root systems, and in
each case we have obtained precisely one lattice. This proves the main
theorem in the case of lattices with a nonempty root system and provides a
complete description of the lattices. (An alternative description is given in
Table 16.1.)

5. A characterization of the Leech lattice

To complete the proof of the main theorem it remains to consider the case
of an empty root system, i.e. we must consider a lattice A is in which
(x, x) ¢ 2 for x E A. In this case there also exists a lattice and it is
438 Chapter 18

unique: it is the remarkable Leech lattice. A characterization of Leech's

lattice as the unique even unimodular lattice in R24 with (x, x) ¢ 2 was
given by Conway (see Chap. 12). Conway's proof is very pretty and can
hardly be improved. However, if we assume (as is natural in our
approach) that the classification of lattices with nontrivial root systems is
already known, then the characterization of Leech's lattice can be obtained
more simply as follows.
Suppose A C R24 is an even unimodular lattice with A (2) = 0. We
want to prove that A is isomorphic to Leech's lattice. From Theorem 17 of
Chap. 7, the condition A (2) = 0 implies that the theta series of A is the
same as the theta series of Leech's lattice, in particular, A (8) ¢ 0. Let
u E A (8) and consider the lattice r = AU = A' U (u/2 + A'), where
A' = {X E A : (X, u) == 0 (mod 2)}. Then r is an even unimodular lattice
in R24 (it is a neighbor of A in the sense of the previous chapter). The
lattice r has a non empty root system, since u/2 E r(2). We shall prove
that the root system for r is kA I. It suffices to show that nonproportional
roots of r (2) are orthogonal. This is very easy. Suppose
x,y E r(2), x ¢ ±y, so that x,y E u/2 + A', and suppose (x,y) ¢ o.
Changing the sign of y if necessary, we may assume that (x ,y) = -1, but
then x - y is a root, i.e. (x - y, x - y) = 2. But x - yEA' C A and
A (2) = 0, which is impossible. Thus r (2) = kA); but then, scanning the
list (I), we see that k = 24 and r is described in paragraph I of §4, i.e. in
R24 we can choose a basis e), ... , e24 such that (ei' e) = 0 if
i ¢ j, (ei' e;) = V2, and "1:.niei E r if and only if ni E Z and {i : ni == I
(mod 2)} is a codeword in the binary Golay code. Since the property of
being adjacent lattices is symmetric, there is a v E r such that A = r".
Suppose v = ');rniei, rni E Z; we will prove that all of the rni are odd.
Indeed, if some rna is even, then (v, 2e a ) = rna == 0 (mod 2); hence
2e a E r" = A, which is impossible, since A (2) = 0. Thus each rni == 1
(mod 2). Since v is defined mod 2r, we may assume that rni = ± 1, and
the action of the Weyl group W{r(2)) allows us to assume that rn) = 1,
i.e. as v we can choose e) + ... + e24. For such a v, r" is the Leech
lattice, as required. This completes the proof of the main theorem.

[Translated by G. A. Kendall]
 19
Enumeration of Extremal
Self-Dual Lattices
J. H. Conway, A. M. Odlyzko and N. J. A. Sloane

We saw in Chapter 7 that the minimal norm of a unimodular lattice in Rn

does not exceed [n/8] + 1. If the minimal norm is equal to this quantity
the lattice is called extremal. In this chapter we show that there are
unique extremal lattices in dimensions 1, 2, 3, 4, 5, 6, 7, 8, 12, 14, 15, 23,
24, and no other such lattices.

1. Dimensions 1-16
The theorem stated in the title is Theorem 13 of Chap. 7. We proceed at
once to the proof. The following result was proved by Kneser [Kne4], and
can also be obtained from Table 16.7.
Theorem 1. The only extremal unimodular lattices in Rn for n ~ 16 are
Zn (I ~n ~7), E g, n(2, (E 7+E 7)+, A(s, in Rl to R7, Rg, R l2 , R l4, R 1S
respectively.
Here A = (L I + ... + L k)+ indicates that A is obtained by gluing
component lattices L I, . . . ,Lk together (cf. §3 of Chap. 4).

2. Dimensions 17-47
Theorem 2. The only extremal unimodular lattices in Rn for
17 ~ n ~ 47 are the shorter Leech lattice 0 23 in R23 and the Leech
lattice A24 in R24.
Proof We could read the results for dimensions 17 to 24 straight from the
tables in Chaps. 16 and 17, but we prefer to give independent arguments.
Suppose first that A is a Type II (or even) lattice in Rn, which is also
extremal as a unimodular lattice. From Cor. 21, Chap. 7, the minimal
norm !l of A satisfies !l ~ 2[n/24] + 2. Also!l = [n/8] + 1 and 81 n,
which imply n = 8, 16 or 24, in which dimensions the even unimodular
440 Chapter 19

lattices were enumerated by Witt and Niemeier - see §2.4 of Chap. 2 and
Table 16.1. Only E 8 and A24 are extremal.
From now on we suppose that A is an odd unimodular lattice which is
extremal. The theta-series of A is therefore the extremal theta-series in
dimension n (§4 of Chap. 7). Let n = 8a + v, 0 ~ v ~ 7. This means
that

8 A(z) = i
,-0
a, 8 3(z)n-8, .:l8(Z)'

(see Eq. (36) of Chap. 7), where the integer coefficients ao . .... aa are
chosen so that

(containing no power of q with exponent between 0 and a + I).

Let Ao = {u E A Iu . u E 2Z} be the even sublattice of A. Then
det A = 1, det Ao = 4, and Ao has theta-series
8 Ao(z) = V2 (8 A (z) +8 A (z + I)J. (3)

To eliminate almost all of the remaining cases we make use of an

argument due to Ward [War2] (who applied it to the weight enumerators
of codes). It will turn out that the condition that the theta-series of the
dual lattice A; has integer coefficients implies that the coefficients a, in (I)
must be divisible by high powers of two. From the Jacobi identity
(Chap. 4, Eq. (19»

[;r [-!],
r
2
8 A (z) = 8A

8 Ao (z) = 8 A (z) + [;
2
8A [- ! + 1] .

Hence

has integer coefficients in its q -expansion. Then

[; r !+
[;r,~ r
/2
8A [- 1]

2
(- I)'r8'a,8 4 [- !r- 8
' 82 [- ;z
i (- l)'r4, a, 82(z )n-8'84(2z )8, ,
,-0
(5)
Enumeration of Extremal Self-Dual Lattices 441

which implies

2'2,-n Ia, for n < 12r . (6)

The a, are easily computed (by equating (1) and (2)) and condition (6)
eliminates all n in the range 17 ~ n ~ 47 except 22, 23 and 24. For
example, when n = 17, ao = 1, a, = - 34, a2 = - 204, and 27 ~ a2'
Finally, n = 22 is eliminated by calculating 0 Ao (z) explicitly. It is
- 11 q3/2 + ... , and the negative coefficient shows that A does not
exist.
When n = 24 the extremal theta series is equal to the theta series of
the Leech lattice, and the uniqueness of the Leech lattice (Chap. 12) shows
that A = A 24 .
In dimension 23 an extremal lattice A must have theta series

The even sublattice Ao (with theta series 1 + 931S0q4 + ... ) has dual
latticeA;with theta series (from (4), (S))
1 + 4600q3 + 931S0q4 + ... + 223 q 23/4(1 + 23q2 + ... )
+ 23·2 12 q'S14(1 + ISq2 + "')(1 - 16q2 + ... )
= 1 + 4600q3 + 94208q'S/4 + 931S0q4 + ... .

From this we can see that the four cosets of Ao in A; contain nonzero
vectors of norms
IS IS
4 + 2m, 4 + 2m, 3 + 2m, 4 + 2m,
and have coset representatives Yo = 0, y" Y2, Y3 (say), for various integers
m ;?- O. The union of the cosets Ao and Ao + Y2 is A. Now the lattice M
of even integers also has four cosets in its dual M*, whose nonzero vectors
have norms
4 + 2m, 1.4 + 2m, 1 + 2m, 1,4 + 2m ,
with coset representatives Zo = 0, z, = V2, Z2 = 1, Z3 = - V2. It follows
that there is a 24-dimensional lattice (Ao + M)+ obtained by extending
Ao + M by the glue vectors Yj + Zj (i = 0, I, 2, 3), which is extremal and
so must be the Leech lattice. A is therefore uniquely characterized as the
projection onto the 23-space orthogonal to v of the vectors of A24 that have
even inner product with a minimal vector v in A 24 . This is the shorter
Leech lattice 0 23 (see also the Appendix to Chap. 6 and Table 16.7). This
completes the proof of Theorem 2.

3. Dimensions n ;?- 48

Theorem 3. No extremal lattice exists in R n for n ;?- 48.

442 Chapter 19

Proof With the same notation as before we show that

0 Ao (z) = 0 A (z) + i (- 16)-rar02(Z)n-8r04(2z)8r

r-O

contains a negative coefficient for n ~ 48, which proves the theorem. The
q -expansion of the right-hand side is
I + (- I)"2 v- 4"a"qv/4
+ (- I)"+12 v- 4"qv/4+2((I6a-v)a,, + 4096a,,_d + ....

We will show that a" < 0 and a,,-l < 0 for a ~ 6, from which the
negative coefficient is apparent. Applying the Biirmann-Lagrange theorem
to (I) and (2) we obtain, exactly as in [Mall, Eq. (6)1,

a
s
= _ ~ -s --
1 d
, d s-l
{dOd
__
3. 08s-n-l.h s
3
} ,
(8)
5. q q q-O

for 0 ~ 5 ~ a, where, from Eq. (33) of Chap. 7,

II
00

h = ((I _q2m-l)(l_ q 4m)}-8 .

m-l

Now d0 3/dq has non-negative coefficients in its q -expansion. When 5 = a

or a-I we have
- I ~ 85 - n - I ~ - 16 .

Let I ~ k ~ 16. Then, from Eq. (35) of Chap. 4,

II II
00 00

03'k = (I _q2m)-k (I +q2m-I)-2k .

m-l m-l

The first product has non-negative coefficients. On the other hand,

II II
00 00

hS (I +q2m-l)-2k = (I _q2m-l)-(8s-2k)
m-l m-l

II II
00 00

x (I _q4m-2)-2k (I _q4m)-8s .
m-l m-l

If n ~ 48, a ~ 6, 5 ~ 5, then every product on the right-hand side has

non-negative coefficients and the first has strictly positive coefficients.
Therefore the expression in braces in (8) has strictly positive coefficients,
showing that a" < 0 and a,,-l < O. This completes the proof of
Theorem 3.
Some of the theta series were calculated on the
Acknowledgement.
MACSYMA system [Mat3].
 20
Finding the Closest Lattice Point
J. H. Conway and N. J. A. Sloane

This chapter describes algorithms which, given an arbitrary point of Rn ,

find the closest point of some given lattice. The lattices discussed include
the root lattices An, Dn , E 6, E 7, E8 and their duals. These algorithms can
be used for vector quantizing or for decoding lattice codes for a
bandlimited channel.

1. Introduction
As discussed in §3 of Chap. 2 and §1 of Chap. 3, lattices are practically
important because they can be used as vector quantizers and as codes for a
bandlimited channel. For these applications it is essential that there exist
fast "decoding" algorithms which, given an arbitrary point of the space,
find the closest lattice point.
In this chapter we describe such algorithms for the lattices An (n ~ 1),
Dn(n ~ 3), E6, E7, E8 and their duals. This chapter is based on [Con29]
and [Con401. The latter reference also gives an algorithm for decoding the
Leech lattice (omitted here because of its length), algorithms for finding
the closest codeword to various binary codes including the Golay code
("soft" decoding algorithms - see §4 beloW), and an extensive
bibliography. See also [Ado 11, [Budl], [Che2], [Con26], [Con38], [For2],
[GorO], [WoI2].
Practical applications of these lattices also require algorithms for
labeling lattice points. Suppose our code or quantizer uses a set
C = lei ..... cm } of lattice points. Efficient "encoding" algorithms are
needed which map i (1 ~ i ~ M) into Ci, and vice versa. The first problem
can be solved using the algorithms described in this chapter, and the
second by using a dual basis for the lattice - see [Con35].
The problem discussed in the following sections is this: for a lattice A
in Rn , find a decoding algorithm which, given an arbitrary point x E Rn ,
finds a lattice point u E A that minimizes the distance from x to u. We
444 Chapter 20

shall use the descriptions of the lattices and their duals given in Chap. 4.
As usual it easier to work with the squared distance or norm
N (x - u) = (x - u) . (x - u) than with the distance itself.

2. The lattices zn, Dn and An

The algorithm for finding the closest point of zn to an arbitrary point
x E R n is particularly simple. For a real number x, let
f (x) = closest integer to x .

In case of a tie, choose the integer with the smallest absolute value. For
x = (Xl .... • x n ) ERn, let
f(x) = (j(Xl) .... ,j(xn ».
For future use we also define g (x), which is the same as f (x) except that
the worst coordinate of x - that furthest from an integer - is rounded
the wrong way. In case of a tie, the coordinate with the lowest subscript is
rounded the wrong way.
More formally, for x E R we define f (x) and the function w (x) which
rounds the wrong way as follows. (Here m is an integer.)
If x = 0 then f (x) = 0, w (x) = I .
If 0 < m ~ x ~ m + V2 then f (x) = m , w (x) = m + I.
If 0 < m + V2 < X < m + 1 then f (x) = m + 1, w (x) = m . (1)
If - m - V2 ~ X ~ - m < 0 then f (x) = - m , w (x) = - m - I.
If - m - I < x < - m - V2 then f (x) = - m - I, w (x) = - m.

(Ties are handled so as to give preference to points of smaller norm.) We

also write

so that 100x) I ~ V2 is the distance from x to the nearest integer.

Given x = (Xl ....• x n ) E Rn, let k (I ~ k ~ n) be such that

10(Xk)1 ~ 10 (x) I for aliI ~ i ~ n

and

Then g (x) is defined by

g(x) = (j(XI),j(X2), ...• W(Xk).··· ,j(xn ».
Algorithm 1. To find the closest point of zn to x.
Given x E Rn, the closest point of Zn is f (x). (If x is equidistant from
two or more points of Zn, this procedure finds the one with the smallest
norm.)
Finding the Closest Lattice Point 445

To see that the procedure works, let U = (u I • . . . • un) be any point of zn.
Then
n
N(u -x) ... ~ (Ui-Xi)2,
i-I

which is minimized by choosing Ui - j (Xi) for i = 1 •...• n. Because of

(1) ties are broken correctly, favoring the point with the smallest norm.
Next we consider Dn (§7.1 of Chap. 4).
Algorithm 2. To find the closest point of Dn to x
Given x E Rn, the closest point of Dn is whichever of j (x) and g (x) has
an even sum of coordinates (one will have an even sum, the other an odd
sum). If x is equidistant from two or more points of Dn this procedure
produces a nearest point having the smallest norm.
This procedure works because j (x) is the closest point of zn to x and
g (x) is the next closest. j (x) and g (x) differ by 1 in exactly one
coordinate, and so precisely one of 'i:,j(Xi) and 'i:,g(Xi) is even and the
other is odd.
Example: Find the closest point of D4 to
x = (0.6,-1.1, l.7,0.n.

We compute j(x) == (1,-1,2,0) and g(x) = (0,-1,2,0), since the first

coordinate of x is the furthest from an integer. The sum of the coordinates
of j(x) is 1 - 1 + 2 + 0 == 2, which is even, while that of g(x) is
o - 1 + 2 + 0 ,. 1, odd. Therefore j (x) is the point of D 4 closest to x.
To illustrate how ties are handled, suppose
x - (~,~,~,~).

In fact x is now equidistant from eight points of D 4, namely (0,0,0,0), any

permutation of (1,1,0,0), and (1,1,1, n. The algorithm computes
j(x) == (0,0,0,0), sum = 0, even,
g(x) ,. (1,0,0,0), sum == 1, odd,

and selects j (x). Indeed j (x) does have the smallest norm of the eight
neighboring points. The algorithm takes about 4n steps to decode Dn.
The lattice An (§6.1 of Chap. 4) consists of the points
U = (UO,UI • ...• un) E Zn+1 with 'i:,Ui - O.
Algorithm 3. To find the closest point of An to x
Step 1. Given x E Rn +1, compute s - 'i:,Xi and replace x by

x' = x - ~1 (1,1 ..... n.

n+
Step 2. Calculate j(x') - (j(x'o), ... ,f(x'n)) and the deficiency
~ = 'i:,j(X'i).
446 Chapter 20

Step 3. Sort the X'i in order of increasing value of 0 (x') (defined in

(2». We obtain a rearrangement of the numbers 0, 1, ... ,n, say i o,
ii, ... , in, such that

Step 4. If .:l = 0, f (x') is the closest point of An to x.

If .:l > 0, the closest point is obtained by subtracting 1 from the
coordinates f(X'io) ' ... ,f(X'itJ.-l)·
If .:l < 0, the closest point is obtained by adding 1 to the coordinates
f(X'i), f(X'in_l) ' ... ,f(X'in-frl)'

Remarks. Step 1 projects x onto a point x' in the hyperplane ~Xi = 0

containing An. Then f (x') is the closest point of Zn + 1 to x', and Steps 3
and 4 make the smallest changes to the norm of f (x') needed to make
~f(X'i) vanish.

The most time-consuming step is the sorting operation, which takes

O(nlogn) steps [Knu1], [ReiS1 As pointed out to us by A. M. Odlyzko,
this can be reduced to a constant times n, by using the Rivest-Tarjan
algorithm [Knu1, p. 216] to find the l.:ll largest (if .:l > 0) or smallest (if
.:l < 0) of the numbers O(X'i).
For n = 2 and 3 there are better algorithms. A2 is the hexagonal
lattice (Fig. 1.3a), and is best decoded using the fact that it is the union of
a rectangular lattice and a translate, as suggested by Gersho ([Ger3,
p. 165], [Con38, p. 299]). A3 is equivalent to D 3, and is best decoded by
the D3 algorithm.

3. Decoding unions of cosets.

A procedure «J> for finding the closest point of a lattice A to a given point x
can be easily converted to a procedure for finding the closest point of a
coset r + A to x. For if «J> (x) is the closest point of A to x, then
«J> (x - r) + r is the closest point of r + A to x.

Suppose further that L is a lattice (or in fact any set of points) which
is a union of cosets of A:
/-1
L = U (rj+A).
i-O

(If L is a lattice, t = [L : A] = (det A / det L )1/2 is the index of A in

L.) Then «J> can be used as the basis for the following procedure for
finding the closest point of L.
Algorithm 4. To find the closest point of L (a union of cosets of a lattice)
to a given point x.
Finding the Closest Lattice Point 447

Given x, compute
(5)

for i = 0, 1 ..... t - 1. Compare each of Yo • ...• Yt-1 with x and

choose the closest.
Examples (a) E6 and E~ (in the complex versions defined by (120), (126)
of Chap. 4) both contain a sublattice equivalent to Ai, spanned by the
vectors w' (8,0,0), w'(0,8, 0), w' (0,0,8). This has index 3 in E 6 and
index 9 in E~. Explicit coset representatives ri are given in [Con381
(b) The versions of E 7 and E; defined in §2.5 of Chap. 5 contain a
sublattice 2Z7, to index 8 and 16 respectively. The coset representatives
form the [7,3,41 and [7,4,31 Hamming codes respectively [Con381.
(c) E g contains Dg as a sublattice of index 2, leading to the fast decoding
algorithm for Eg described in [Con261, [Con291. This decodes Eg in about
104 steps. An even faster algorithm is given in §6.
(d) K 12 has several possible sublattices, but the best seems to be the one
equivalent to A~, of index 64, with coset representatives forming the
hexacode [Con381.
(e) A24 has a sublattice of index 8192 that is equivalent to D 24 , which
leads to a relatively slow decoding algorithm. Faster algorithms are
described in [Con401, [For21
(f) Dn has index 4 in D;, and An has index n + 1 in A;, so Algorithm 4
can be applied to D; and A;. Coset representatives (the glue vectors for
Dn and An) are given in Chap. 4.

4. "Soft decision" decoding for binary codes

Let C be an [n, k I binary code. By replacing the O's and l's of the
codewords by the real numbers + 1 and -1 we may regard C as embedded
in Rn. A- "soft decision" (or "maximum likelihood") decoder for C is an
algorithm which, when presented with an arbitrary point x E Rn, finds a
codeword c E C that minimizes the Euclidean distance from x to c. Note
that
N(x-c) = (x-c)· (x-c) = X' x - 2x . c + n, (6)

since the components of care ± 1, so it is enough to find acE C that

maximizes the inner product x . c. Many examples (including the Golay
codes CC23 and CC 24 ) and an extensive bibliography are given in [Con401
Here we just give one example, for use in §6. A first-order Reed-
Muller code (§2.7 of Chap. 3) has parameters [n,kl = [2m,m+ll, and
may be decoded in about m 2 m steps using the fast Hadamard transform.
(The so-called "Green machine" described in [Gre21, [Gre31, [Har81,
[Mac6, Chap. 141, [Posll. Litsin and Shekhovtsov [Lit21 have recently
shown that even this algorithm can be speeded up.)
448 Chapter 20

5. Decoding lattices obtained from Construction A

If C is an [n, k] binary code, written in ± 1 notation, Construction A (§2.1
of Chap. 5, §2 of Chap. 7) produces a lattice A(C) consisting of the
vectors of the form
c + 4z, c E C, z E zn.
(The set of points (7) is strictly speaking not a lattice, but the translate of
a lattice by the vector (I,!, ... , 1).)
The following lemma makes it possible to use a decoding algorithm for
C to decode A (C).
Lemma. Suppose x = (XI, . . • ,xn) lies in the cube -I ~ Xi ~ 1
(j = 1, ... , n). Then no point of A (C) is closer to X than the closest
codeword of C.
Proof Suppose the contrary, and let u = (u I, . . . , un) be a closest lattice
point to x. By hypothesis some Xi'S are neither + 1 nor - 1. By
subtracting a suitable vector 4z, we may change these coordinates to + 1 or
-1 (depending on their parity) to produce a point of A (C) that is in C,
and is at least as close to X as u is, a contradiction.
Algorithm 5. To find the closest point of A (C) to a given point x.
(j) Given x = (x I, . . . , x n ), we first reduce all Xi to the range
- 1 ~ Xi < 3 by subtracting a vector 4z.
(ij) Let S denote the set of i for which 1 < Xi < 3. For i E S, replace
Xi by 2 - Xi'
(iii) Since X is now in the cube -1 ~ Xi ~ 1 (j = 1, ... ,n), by the
Lemma we are justified in applying the decoder for C to x, obtaining an
output c = (c I, . . . , cn) say.
(iv) For i E S change Ci to 2 - Ci' Then C + 4z is a closest point of
A (C) to the original vector x.
The total number of steps required is roughly 5n plus the number to
decode C.

6. Decoding E8
Since E8 = A (,Yfs), and the Hamming code ,Yfs (§2.4.2 of Chap. 3) is a
first order Reed-Muller code, E8 can also be decoded by this algorithm.
,Yfs can be decoded in about 3·8 + 8 = 32 steps by the fast Hadamard
transform (§4), and so E 8 can be decoded in about 72 steps. It is worth
noting that this is faster than the more obvious algorithm given in
Example (c) of the previous section.
 21
Voronoi Cells of Lattices
and Quantization Errors
J. H. Conway and N. J. A. Sloane

In n-dimensional space, what is the average squared distance of a random

point from the closest point of the lattice An (or Dn, En, A; or D;)? If a
point is picked at random inside a regular simplex, octahedron, 600-cell or
other polytope, what is its average squared distance from the centroid?
The answers are given here, together with a description of the Voronoi cells
of the above lattices. The results have applications to quantization and to
the design of codes for a bandlimited channel. For example, a quantizer
based on the eight-dimensional lattice E 8 has a mean squared error per
symbol of 0.0717 ... when applied to uniformly distributed data, compared
with 0.08333 ... for the best one-dimensional quantizer.

1. Introduction
Let A be an n-dimensional lattice in Rn. The Voronoi cells (§1.2 of
Chap. 2) around the lattice points are congruent polytopes, and we denote
by V(O) the Voronoi cell around the origin. Then, as we saw in Eq. (87)
of Chap. 2, the normalized, dimensionless, second moment of inertia of
V(O),
n+2
G (A) = -1 (det A) - 2n
- f X .X dx ,
n v(o)

is equal to the mean squared error per symbol when the lattice is used as a
vector quantizer for uniformly distributed inputs. If instead the lattice is
used as a code for a bandlimited channel, the integral of exp( - x . X /2( 2)
over V(O) gives the probability of correct decoding (§1.3 of Chap. 3).
Furthermore the vertices of V(O) furthest from A are the deep holes in A,
and their distance from A is the covering radius R of A (§1.2 of Chap. 2).
Spheres of radius R around the lattice points just cover Rn, so R
determines the thickness of A (§l.l of Chap. 2).
450 Chapter 21

For all these applications it is important to understand the structure of

the polytope V(O). In this chapter we give fairly explicit descriptions of
V(O) for the root lattices An (n ~ 1), Dn (n ~ 3), £6, £7, £8 (§2 of
Chap. 4) and their duals (except for £~ and £;). Furthermore it will turn
out that for these lattices it is possible to evaluate (1) exactly.
In §2 we study the second moment of an arbitrary polytope P, not
necessarily the Voronoi cell of a lattice. Of course Voronoi cells V(O) have
the special property of being space-filling polytopes: the whole space Rn
can be tiled by translated copies of V(O). In three dimensions polytopes
with the latter property are called parallelohedra. (See [McMll, [VenOl.)
If P is a polytope with centroid i, then its volume, unnormalized
second moment and normalized second moment are
Vol (p) = J dx,
p
U(p) = J Ilx -i1l
p
2 dx,

[(p) = U(p)
Vol (p) ,

respectively, and its dimensionless second moment is

G (p) = .1 U(p) I(P)
(4)
n Vol (p)1+2/n n Vol (p)2/n

For applications to quantizing we are interested in the minimal value of

G(p) taken over all polytopes P that tile by translation (cf. [Bar16]).
Summary of results. In §§2 and 3 we compute G (p) for a number of
important polytopes, including all the regular polytopes (see Theorem 4).

Table 21.1. Comparison of dimensionless second moments G(p) for the

sphere and various 3-dimensional polyhedra P (* indicates a space-filling
polyhedron) .

P G(P)

tetrahedron .1040042 ...

cube· .0833333 ...
octahedron
. .0825482 ...
hexagonal prism
. .0812227 ...
rhombic dodecahedron
truncated octahedron
. .0787451 ...
.0785433 ...
dodecahedron .0781285 ...
icosahedron .0778185 ...
sphere .0769670 ...
Voronoi Cells of Lattices and Quantization Errors 451

Table 21.2. Comparison of dimensionless second moments G(P) for the

sphere and various 4-dimensional polytopes P (* indicates a space-filling
polytope).

P G(p)

simplex {3,3,3} .1092048 ...

hypercube * {4,3,3} .0833333 ...
16-cell* {3,3,4} .0816497 ...
24-cell* {3 ,4,3} .0766032 ...
120-cell {5,3,3} .0751470 ...
600-cell {3,3,5} .0750839 ...
sphere .0750264 ...

The three- and four-dimensional polytopes are compared in Tables 21.1

and 21.2. The chief tools used are Dirichlet's integral (Theorem 1), an
explicit formula for the second moment of an n -simplex (Theorem 2), and
a recursion formula giving the second moment of a polytope in terms of its
cells (Theorem 3).
In §3 we consider the lattices An, Dn, En and their duals, determine
their Voronoi cells V(O), the second moments G (A) - G (V(O», and their
covering radii R. The second moments of these and other lattices are
summarized in Table 2.3 and Fig. 2.9 of Chap. 2. It is worth emphasizing
the remarkably low value of the mean squared error for the lattice E8 (see
(28) and Fig. 2.9). Fast algorithms for quantizing with this lattice were
described in the preceding chapter.

2. Second moments of polytopes

A polytope in this chapter means a closed and bounded convex region of
Rn that is the intersection of a finite number of half-spaces (cf. [Cox20,
§7.4]). If we use a minimal collection of such half-spaces then the part of
the polytope that lies in one of the bounding hyperplanes is called a face.
If all edge-lengths are equal this length will usually be denoted by U. The
main source for information about polytopes is Coxeter [Cox20], but there
is an extensive literature, particularly for low-dimensional figures. See for
example [AIel], [Coxl], [Cox5]-[Cox7], [CoxlO], [CoxI4], [CoxI8],
[Cox22], [Cox23], [Cox25], [Cox27], [Cox28], [Cunl], lEltl], [Fej9],
[Fejl0], [Griil], [Hil2], [Holl], [Loel], [McM2], [Pugl], [Wenl],
[Wen2]. Second moments about an axis are tabulated for many simple
polyhedra in standard engineering handbooks (see also [Satl]), but these
references do not give the values for G (p).
2.A. Dirichlet's integral. A few special figures can be handled using
Dirichlet's integral.
452 Chapter 21

where the integral on the left is taken over the region bounded by
XI ~ O, ... ,Xn ~ 0 and XI +···+Xn ~ l.
2.B. Generalized octahedron or crosspolytope. Consider for example the
n -dimensional generalized octahedron or crosspolytope f3n [Cox20, §7.2] of
edge-length U. Takingf=l, (XI=I (to get the volume) or (XI=3 (to get the
second moment) and (Xi = 1 for i ~ 2 in Theorem 1 we find
Vo/(f3n) 2n/2 J(f3n) n
(u)n = 7' (U)2 = (n+ 1) (n+2) ,
G _ (n!)2/n
(6)
(f3n) - 2(n + 1) (n +2)
1
- -2 = 0.0676676 ... as n - 00
2e

2.C. The n-sphere. As a second application, for the n-dimensional solid

sphere (or balI) Sn of radius p it is easy to deduce from Theorem 1 that
Vo/(Sn) 7r n/2

r(Vzn + 1)
(which establishes (16), (17) of Chap. 1), I(Sn) = np2/(n+2), and
r(Vm+ 1)2/n
G(Sn) = (n+2h (8)

1
- -2- = 0.0585498 ... as n - 00
7re

It is clear from the definition that the sphere has the smallest value of
G (p) of any figure, so (8) establishes Zador's lower bound, the sphere
bound (the left side of Eq. (82), Chap. 2). Unfortunately quantizers
cannot be built with either spheres (unless n = 1) or generalized octahedra
(unless n = I, 2 or 4) as Voronoi cells, since these objects do not fill space.
2.D. n-Dimensional simplices. The next result makes it possible to find
the second moment of any figure, provided it can be decomposed into
simplices (cf. [Hohl]).
Theorem 2. Let P be an arbitrary simplex in Rn with vertices
Vi =(ViI, ... ,Vin) for 0 ~ i ~ n. Then (a) the centroid of P is the
barycenter

(9)
Voronoi Cells of Lattices and Quantization Errors 453

of the vertices,
I VOl ••• VOn

I I Vll"'Vln
(b) VoHp) = , det , and (10)
n.
I Vn I··· vnn

(c) the normalized second moment about the origin 0 is

+1 I n 2
10 = _n_ 11,,11 2 + ( )( ) ~ Ilv;1I .
n+2 n+1 n+2 ;-0

In other words 10 is equal to the second moment of a system of n + I

particles each of mass 1/( (n + 1) (n + 2» placed at the vertices and one
particle of mass (n + 1) /(n + 2) placed at the barycenter.
Proof (a) is elementary, (b) is well-known (cf. [Alll], [Gool, p. 349]),
and (c) follows from [Goo I , Eq. (24), the case n = 21.
2.E. Regular simplex. For example if P is a regular n-simplex of edge-
length 2f then
Vol (p) = _.J_n_+_1 I (p) n
(.J2f)n n! , (.J2f)2 (n+ 1) (n+2) ,
G (p) = _----'(_n!..;...)2_/n_ _
[+1..
(n+1) n (n+2)
- e- 2 = 0.135335 ... as n _ 0 0 .

For n=I,2 and 3 the values of G(p) are 1/12, 1/(6.Ji) = 0.0962250 ... ,
and 32/ 3/20 = 0.104004 ... , and G (p) increases monotonically with n.
2.F. Volume and second moment of a polytope in terms of its faces.
Instead of decomposing a figure into simplices one may proceed by
induction, expressing the volume and second moment of a polytope in terms
of the volume and second moment of its faces, then in terms of its (n - 2)-
dimensional faces, and so on. Theorem 3 is the basis for this procedure.
Suppose P is an n-dimensional polytope with NI congruent faces Fb
F I ', F I ", ••. , N2 congruent faces F 2, F'2, F 2", • •• , and so on. Suppose
also that P contains a point 0 such that all of the generalized pyramids
OF b OF I' , ... are congruent, all of OF 2, OF 2' , ... are congruent, .... Let
a; E F; be the foot of the perpendicular from 0 to F;, let h; = II0a;ll, and
let V n - I (j) be the volume of F; and Un - I (j) the unnormalized second
moment of F; about a;.
Theorem 3. The volume and unnormalized second moment about 0 of P
are given by
Nh·
Vol (p) = ~ - '- ' Vn - I (j) ,
; n
454 Chapter 21

Proof Follows from elementary calculus by dividing each generalized

pyramid OF 1 into slabs parallel to the face F;.
2.G. Truncated octahedron. For example let P be the truncated
octahedron with vertices consisting of all permutations of .Jie (0, ± 1, ± 2).
P has N 1 = 6 square faces and N 2 = 8 faces which are regular hexagons,
all with edge-length U. The second moments of these faces can be
calculated directly, or else found in §2.1. Then from the theorem we find
that

Vol (p) = 6· e-/8 4e2 + 8· e../6 6.J3e 2 = 64.Jie 3

3 3 '

U(P) = 6· :-/8 [8f2 . 4e2 + 8t ] + 8· :../6 [6e 2 . 6.J3e 2 + 10.J3e4]

= 304.Jie 5 ,

hence
1 ](p) 19
G (p) = 3 VoHp)2/3 192 3.Ji = 0.0785433 .... (13)

2.U. Second moment of regular polytopes. The next theorem gives an

explicit formula for the second moment of any regular polytope. Suppose
P is an n-dimensional regular polytope [Cox20). For 0 ~ j ~ n choose a
j -dimensional face Fj of P so that F 0 t;; FIt;; ... t;; Fn = P, and let OJ be the
center of F j , R j = I10nOjll, and for j ~ 1 let rj = I10j_IOjll. Thus rj is the
inradius of Fj measured from OJ, and r] = R]-l - R}. Let N jJ - 1 be the
number of (j -1) -dimensional faces of Fj . Then it is known that the
symmetry group of P has order
g = Nn,n-l N n- 1,n-2'" N 2,IN 1,o

[Cox20, §7.6], and the volume of P [Cox20, §7.9) is

rlr2"' rn
VoHp) = Nn,n-l'" N 2,IN 1,o . n!

Theorem 4. The second moment of any n -dimensional regular polytope P

about its center On is given by

](p) = 2 (R6 + 2Rr + 3Ri + ... + nR;_I) , (15)

(n+1) (n+2)

or equivalently
2 2 2 2 n (n + 1) 2
J(p) - (n+1) (n+2) (rl + 3r2 + 6r3 + ... + 2 rn)· (16)

Proof The proof is by induction, the I-dimensional case being immediate.

From Theorem 3 we have
Voronoi Cells of Lattices and Quantization Errors 455

where (from (14) and the induction hypothesis)

Then
I(P) = U(p) = _n_{,2 + 2 (,,2 + ... + n (n-1) 2 )}
Vo/{p) n+2 n n (n+ 1) 2 'n-' ,
which simplifies to (16).
The values of g, Vo/{p) and R j are tabulated for all regular polytopes
in [Cox20, Table I, pp. 292-2951. We have already dealt with the simplex
and generalized octahedron. For an n-dimensional cube G (p) = 1/12 for
all n (since the cube is a direct product of line segments). We now treat
the remaining regular polytopes.
2.1. Regular polygons. If P is a regular p-gon of edge-length U, then
from Theorem 4 we find
1r' £2 1r'
Vo/{p) = p£2 cot - , [(p) = -6 (1 + 3 cot 2_) ,
p p
G (p) = _1_( cosec 21r' + cot ~) .
6p P P

For p = 3,4 and 6, G(p) = 1/(6,/3), 1/12 and 5/(36,/3).

2.J. Icosahedron and dodecahedron. For the icosahedron
20£3 2 r 1
Vol = - 3 - ' G = 20"5
2/3 [6r)
= 0.0778185 ... , (18)

where r = (J5 + 1)/2, and for the dodecahedron

r3s
2/3
[ )
= 0.0781285 ....

2.K. The exceptional 4-dimensional polytopes. There are three

"exceptional" regular polytopes in four dimensions, the 24-cell, the 120-cell
and the 600-cell. For the 24-cell
13 (20)
Vol = 3U4 , G = ----;;:: - 0.0766032 ... ;
120,,2

for the 120-cell

~ 43r + 13
Vol = 120,,5£4r 8, G = 300.J6 51/4 = 0.0751470 ... ;
456 Chapter 21

and for the 600-cell

Or + 4) rl12
Vol = 100e4r 3 , G = 150 = 0.0750839... . (22)

The three- and four-dimensional polytopes that have been considered are
compared in Tables 21.1 and 21.2.

3. Voronoi cells and the mean squared error of lattice

quantizers
3.A. The Voronoi cell of a root lattice. The Voronoi cells of the root
lattices An(n~1), Dn(n~3), £6, £7 and £8 may be obtained in a
uniform manner. The method is based on finding a fundamental simplex
for the affine Weyl group of the lattice (§2 of Chap. 4). We denote the
(finite) Weyl group of A by W(A), and the (infinite) affine Weyl group by
Wa(A).

The affine Weyl group Wa (A) is described by the extended Coxeter-

Dynkin diagram shown in Figs. 21.1-21.3 (and in the last column of
Table 4.1). This diagram can be read in several different ways ([Bou 11,
[Cox20], [Gr031, [Hazl]). First, it provides a presentation for Wa (A),
defining the group in terms of generators and relations (see §2 of Chap. 4);
however we shall not make use of this interpretation here. Second, it can
be used to specify a fundamental simplex S for Wa (A). This is an n-
dimensional closed simplex whose images under the action of Wa (A) are
distinct and tile Rn. In other words we can write
Rn = U g(S), (23)
gE W.(A)

where (except for the boundaries of g(S), a set of measure zero) each
point x E Rn belongs to a unique image g (S) . In this interpretation the
nodes of the diagram represent the hyperplanes that are the walls of the
fundamental simplex [Cox20, §11.3]. The angle between two walls or
hyperplanes is indicated by the branch of the diagram joining the
corresponding nodes. If the hyperplanes are at an angle of 7r /3 the nodes
are joined by a single branch, if the angle is 7r /4 they are joined by a

Figure 21.1. Extended Coxeter-Dynkin diagram for W.(An) . The

extending node is indicated by a solid circle. The n +I nodes are labeled
with the equations to the hyperplanes that are the walls of the
fundamental simplex. The labelings in Figs. 21.1-21.3 are based on
[Boul, pp. 250-270) and [Con25J.
Voronoi Cells of Lattices and Quantization Errors 457

In- I= Xn

Wo (O n )

Figure 21.2. Extended Coxeter-Dynkin diagram for Wa(D.) . There are

n+l nodes.

double branch (see Fig. 21.4), if the angle is 7r /p with P > 4 they are
joined by a branch labeled p, and finally if the hyperplanes are
perpendicular the nodes are not joined by a branch. The nodes in
Figs. 21.1-21.3 have been labeled with the equations to the corresponding
hyperplanes.
In the third interpretation the nodes in the extended Coxeter-Dynkin
diagram are taken to represent the vertices of a fundamental simplex,
rather than the bounding hyperplanes. Each node represents the vertex
opposite to the corresponding hyperplane (some examples are shown in
Figs. 21.6-21.8) - cf. [Cox20, §11.6].
One of the nodes in the diagram is indicated by a solid circle. This is
an extending node; removing it leaves an (ordinary) Coxeter-Dynkin
diagram for the Weyl group W(A}. Of the n+ 1 hyperplanes represented
by the extended diagram, all except that corresponding to the extending
node pass through the origin. It is helpful to think of the latter hyperplane
as forming the roof of the fundamental simplex. The vertex of the
fundamental simplex opposite the roof is the origin.
For later use we remark that the finite Weyl group W(A) also has a
fundamental domain, consisting of an infinite cone centered at the origin.
A fundamental simplex for Wa (A) is obtained by taking the finite part of
the cone beneath the roof. The intersection of this cone with the roof, or
more precisely with a unit sphere centered at the origin, is a spherical
simplex. The ordinary Coxeter-Dynkin diagram for W(A) describes this
spherical simplex in the same way as the extended diagram describes the
fundamental simplex for Wa (A) . These spherical simplices and (ordinary)
Coxeter-Dynkin diagrams can be used to define the Weyl groups of all the
root systems (and not just the root lattices An, Dn and En). In §§3.E, 3.F
we shall require the spherical simplices corresponding to W(A n ) and
W(Cn ), shown in Fig. 21.4. The Weyl group W(A n ) is usually written as
[3 n - I ] and is isomorphic to the symmetric group on n+l letters. W(Cn ) is
written [3n-2,4] and has order 2nn!. (See §2 of Chap. 4.)
Lemma. The origin is the closest lattice point to any interior point of the
fundamental simplex.
Proof Let u be the closest lattice point to xES . Suppose u ;:co. Then
u ¢ S, and u and x are on opposite sides of a reflecting hyperplane of
458 Chapter 21

XI+XS ' xZ + ' .. + x7

x2 : x 3

(a)

XI + xs ' x2 + ". + x7

(b)

XI + xS ' Xz + '.. + X7

'2I ('I + ' +'5 - '6 - '7 + 'S) , I

(c)
Figure 21.3 . Extended Coxeter-Dynkin diagrams for (a) W. (£8),
(b) W.(£,) and (c) W.(£6) .

Wa (A). Let u' E A be the image of u in this hyperplane, and let y be the
foot of the perpendicular from x to the line uu' (see Fig. 21.5). Then
Ilxu'112 = IIxyl12 + Ilyu'112 < IIxyl12 + Ilyull 2 = Ilxul1 2,

and x is closer to u' than to u, a contradiction. Therefore u = O.

(a)

(b) ~
n n·1 3 2 I
Figure 21.4. Coxeter-Dynkin diagrams for the spherical simplices of (a)
W(A.) and (b) W(C.). (The labeling of the nodes is for convenience
only and has no geometrical significance.)
Voronoi Cells of Lattices and Quantization Errors 459

u
REFLECTING
HYPERPLANE

u'

Figure 21.5. Why 0 is closer to x than u is.

The connection between the fundamental simplex and the Voronoi cell
is given by the following theorem.
Theorem 5. For any root lattice A, the Voronoi cell around the origin is
the union of the images of the fundamental simplex under the Weyl group
W(A).

Proof Let x be any point of the Voronoi cell around the origin. From
(23), x E g (S) for some g E Wa (A). Suppose x is an interior point of
g (S) . By the lemma, the closest lattice point to x is g (0). Therefore
g(O) = 0, g E W(A), and

x E U g(S) .
g E w(A)

We omit the discussion of the case when x is a boundary point of g(S).

The converse statement, that x E U g(S) implies x is in the Voronoi cell,
follows by reversing the steps.
It follows from Theorem 5 that the Voronoi cell is the union of I W(A) I
copies of the fundamental simplex S. Furthermore the faces of the
Voronoi region are the images of the roof of the fundamental simplex
under W(A) . Thus the Voronoi cell is bounded by hyperplanes which are
the perpendicular bisectors of the lines joining 0 to its nearest neighbors in
the lattice.
Corollary. The number of (n - 1) -dimensional faces of the Voronoi cell of
a root lattice is equal to the kissing number of the lattice.
This is not true for all lattices, as we shall see in §3.G. The second
moment of the Voronoi region can now be obtained from that of the
fundamental simplex.
3.B. Voronoi cell for An. We first find the vertices Yo, Vlo .. . , Vn of the
fundamental simplex S. These are found by omitting each of the
hyperplanes of Fig. 21.1 in turn and calculating the point of intersection of
the remaining n hyperplanes. The results are shown in Fig. 21.6, where
each node is labeled with the coordinates of the vertex opposite the
corresponding hyperplane. The ith vertex is
v. = ( (-=L)i (_i_)j )
I n+l' n+l
460 Chapter 21

where i + j = n + I, for 0 ~ i ~ n, and is the same as the glue vector til

for An in A~ (see Eq. (55) of Chap. 4). Also

Ilv.11 2
I
= .J.L
n+1
. (24)

The barycenter of S (Eq. (9» is

A -n -n+2 n-2 n
v = (2n+2' 2n+2 ,... , 2n+2' 2n+2)

and from (J I) the normalized second moment about the origin is

() n+1 2 I n 2 n n
+ ( )( ) + ( )
A

I S = -- Ilvll ~ IIVil1 = -
n+2 n+1 n+2 i-O 12 6 n+1

Now I w(An)1 = (n+l)! and det An = n + I (§6.1 of Chap. 4). By

Theorem 5 the Voronoi cell around the origin, yeO), is the union of
I WeAn) I copies of S, so
[(yeO»~ = u(V(O» U(S) (25)
Vo/(V(O» Vol (S) = I (S) ,

G(A)=.l [(yeo»~ I [I I) (26)

n n Vol(V(o»2/n = (n+l)l /n 12 + 6(n+1)

I
- - as n -+ 00
12

(in agreement with [Dav2]) . Once the Voronoi cell has been found we can
also determine the points in R n at maximal distance from the lattice, since
these are necessarily vertices of the Voronoi cells . From (24) it follows
that the covering radius of An is R = (ab I(n + J)) 1/2 = p(2ab I(n + I) 1/2,
where p is the packing radius and a = [(n + 1) 121 and b = n + I-a.
Typical points at this distance from An are the vertex Va of yeO) and its
images under WeAn)'
The lattice A I consists of equally spaced points on the real line, and
G (A I) = 1/12. The lattice A 2 is the hexagonal lattice, the fundamental

n_I)2 ( 2 )n-I)
. .. Vn = ((- nh
~
r. (n ~ IY)
v2 = ((-0+1 . ii+I

Figure 21.6. Vertices of fundamental simplex for An.

Voronoi Cells of Lattices and Quantization Errors 461

Figure 21.7. Vertices of fundamental simplex for Dn .

((~)
(a)

(b)

(08 )

(c)
Figure 21.8 . Vertices of fundamental simplices for (a) £8. (b) £7 and
(c) £6.
462 Chapter 21

simplex is an equilateral triangle, the Voronoi cell is a hexagon, and

G(A 2 ) = 5/(36./3) (compare §2.I). The lattice A3 is the face-centered
cubic lattice, the Voronoi cell is a rhombic dodecahedron (Fig. 2.2a) and
G (A 3) - r ll/3 = 0.0787451 ....
For n = 1 and 2 it is known that An is the optimal quantizer, but for
n = 3 the dual lattice A;
is better (see §3.3 of Chap. 2). The values of
G (An) for n ~ 9 are plotted in Fig. 2.9. G (An) decreases to its minimal
value 0.0773907 ... at n = 8 and then slowly increases to 1112 as n -+ 00.
3.C. Voronoi ceU for Dn(n ~ 4). Dn (for n ~ 4), E 6 , E7 and Eg are
handled in the same way as An and our treatment will be brief. The
vertices vo, ... , Vn of a fundamental simplex for Dn, n ~ 4, are shown in
Fig. 21.7. Their barycenter is
1
v= 2(n+I) (0,2,3, ... ,n - 2,n -1,n + 1),

I W(Dn) I = 2n - 1 . n! and det Dn = 4, from which we obtain

G (Dn) = 2~/n (112 + 2n (~+ 1) ) (27)

1
-+ 12 as n -+ 00 •

The covering radius of Dn (for n ~ 4) is R = .In/4 = p.Jnfi, as

illustrated by the vertex Vo = (\~n). For n = 4 the Voronoi cell is a 24-cell
(see for example [Cox20, §8.7D, and G (D 4 ) = 13/(I20.Ji) = 0.0766032 ... ,
in agreement with (20). The values of G (Dn) for n ~ 9 are plotted in
Fig. 2.9. G (Dn) takes its minimal value 0.0755905 ... at n = 6 and then
slowly increases to 1112 as n -+ 00.
3.0. Voronoi cells for E 6• E 7• Eg. The vertices of fundamental simplices
for E 6 , E7 and Eg are shown in Fig. 21.8. For Eg
IW(Eg)1 = 2 14 .3 5 .5 2 .7 - 696729600, det = 1,

v= 1;80 (5,35,55,79,109,149,209,751),
929 (28)
G(E g) = 12960 = 0.0716821 ...

(Watson [Wat2] found a different value.) The Voronoi cell V(O) is an 8-

dimensional polytope which is the reciprocal(J) to the Gosset polytope 421
described on page 204 of [Cox201
Let Ni denote the number of i-dimensional faces of V(O). Then from
[Cox20, §11.8] we have No=19440, N 1 =207360, N2 = 483840,
N3 = 483840, N4 = 241920, N5 = 60480, N6 = 6720 and N7 = 240. The

WIt is not difficult to give a direct proof of this statement; it also follows
from Theorem 8 below.
Voronoi Cells of Lattices and Quantization Errors 463

19440 vertices consist of 2160 at distance 1 from 0 and 17280 at distance

2.Ji/3. The former are the images of the vertex (07,}) of the fundamental
simplex under W(E g), and are at the maximal possible distance from E g,
while the latter are the images of the vertex «1/6)7,5/6) under W(E g).
Thus R = 1 = p.Ji. The other seven vertices of the fundamental simplex
are not vertices of the Voronoi cell.
For E7,1 W(E 7) 1 = 210 .3 4 .5.7 = 2903040, det = 2,

v= - ;6 (l,5,8, 12, 18, 30,-42,42),

163 -In
G(
E7) - 2016 . 2 = 0.0732306 .... (29)

The covering radius of E 7 is R - .J372 = p.J3, as illustrated by the vertex

(0 5, -1, Ih, -Ih).
For E6,1 W(E 6) 1 = 27 . 34 • 5 = 51840, det "" 3,

v'" 412 (0,3,5,8,14,-14,-14,14),

5 (30)
G (E 6 ) = 1/6 = 0.0743467 ....
56·3

The Voronoi cells for E7 and E6 are the reciprocals of the Gosset polytopes
231 and 122 described in [Cox20, §11.8]. The covering radius of E6 is
R = 2/.J3 = pM, as illustrated by the vertices (05, -2/3, -2/3, 2/3) and
(04, 1, -1/3, -1/3,1/3).
For E~ and E; see [Con38], [Worl1, [Wor2].(2)
3.E. Voronoi cell for D;. In order to determine the Voronoi cells for the
dual lattices A; and D; we shall use Wythoffs construction, as described
in [Cox5] and [Cox20, §11.6]. The idea is to construct new polytopes out
of the spherical simplices described in §3.A, the vertices of the new
polytope being indicated by drawing rings around certain nodes in the
ordinary Coxeter-Dynkin diagram. More precisely, let Vlo ••• , Vn be the
vertices of a spherical simplex for a Weyl group W(A). If a single node of
the diagram is ringed, say that corresponding to Vi, the vertices of the new
polytope are the images of Vi under the Weyl group. If two or more nodes
are ringed, say those corresponding to Vi, Vj, ••• , the symbol represents a
polytope whose vertices are the images under W(A) of some interior point
of the spherical subsimplex with vertices Vi' Vj,.... We can adjust the
metrical properties of the polytope (for example, equalize its edge lengths)
by choosing this interior point suitably. Some 1-, 2- and 3-dimensional
examples are shown in Fig. 21.9; others may be found in [Cox5], [Cox20].

(2)Worley [Worl1, [Wor2] finds the following exact values

G(E;) = 12619·31/6/204120 = 0.0742437,
G(En = 21361·2 In/322560 = 0.0731165.
464 Chapte r 21

(a) ® o (k)~

(b)@----<> o (I) @---@--:>

(c)@--€) ( m) o----@------o

(d)@== @ (n) @--B----@

(e)@=:::€) (j) CC(l=~@

Figure 21.9. Examples of polytopes obtained by Wythoff's construction :

(a) edge, (b) triangle, (c) hexagon, (d) square, (e) octagon, (f) cube, (g)
truncated cube, (h) cuboctahedron, (j) truncated octahedron, (j)
octahedron, (k) tetrahedron, (]) truncated tetrahedron, (m) octahedron
(again), (n) truncated octahedron (again>.

We now use this construction to find the Voronoi cell for n ~ 3. D:,
We change the scale so that D: becomes the union of (2z)n and
(I n) + (2z)n . The closest points to the origin in the first set consist of 2n
points of the form (± 2,on-I), and the closest in the second set consist of
2 n points of the form (±In) . The Voronoi cell V(O) is the intersection of
the Voronoi cells determined by these two sets. The first of these, P say, is
a cube centered at 0 with vertices (± 1n). The second, Q say, is a
generalized octahedron with vertices (±nI2,On-') . Furthermore Q can be
obtained by reciprocating P in a sphere of radius p = Jnfi centered at the
origin.
Thus the Voronoi cell V(O) is the intersection of P and a reciprocal
polytope Q=p. . In other words V(O) is obtained by truncating P in the
manner described on page 147 of [Cox201, and is therefore specified by
ringing one or two nodes of the Coxeter-Dynkin diagram (Fig. 21.4b) for
the spherical simplex of P ([Cox51, [Cox20, §§8 .1, 11.7]) .
The radii R j (defined in §2.H) for the cube P are given by R j = .In-j
[Cox20, Table I, p. 2951. If n is even the radius p of the sphere of
reciprocation is equal to Rn/2' and we must ring the node labeled nl2 in

o~==@

oj" 0 @

04 =---@---<>
o;~

06 =---<>-----@-<>

•••
Figure 21.10. Voronoi cells for the lattices D~ .
Voronoi Cells of Lattices and Quantization Errors 465

Fig. 21Ab. If n is odd p lies between R (n-t)/2 and R (n+t)/2 and both nodes
(n - 1)/2 and (n + 1)/2 must be ringed. We have therefore established the
following theorem.
Theorem 6. The Voronoi cell around the origin of the lattice D: is the
polytope defined by the diagrams in Fig. 21.10.
The coordinates for (Hn,k) and O(n,k) given below show that the
edge-lengths of the Voronoi regions are all equal. In Coxeter's notation
[Cox20, §8.I] the Voronoi cell for D;,
is

{ 3 3 ... 3 }
33 ... 34

with t-I 3's in each row, and for D;,+I it is

{ 3 33 ... 3 }
33 ... 34

with (-I 3's in each row.

We shall determine the second moments of these Voronoi cells
recursively, using Theorem 3. In order to do this it will be necessary to
find the second moments of all the polytopes a(n,k), (Hn,k), -y(n,k) and
O(n, k) defined as in Fig. 21.11. In this notation the Voronoi cell of is D:
(up to a scale factor) equal to (Hn,n/2) if n is even and to O(n,(n-1)/2)
if n is odd. Let Ra(n,k), Va(n,k) and Ua(n,k) denote respectively the
circumradius, volume and un normalized second moment about the center
of a(n,k), with a similar notation for (3(n,k), -y(n , k) and O(n,k).

a ( 6 , 2) ~

(3(6,2) ~

r (6,2) o--o--o----@---
8 (6 , 2) =----->-----@--
Figure 21.11. The polytopes a(n,k), {3(n,k), ,,(n,k) and Q(n,k). In
general a(n,k) has n nodes with the k-th node from the right ringed,
and ,,(n,k) has n nodes with the k-th and (k+I)st ringed (except that
,,(n,O) = a(n,I) and ,,(n,n) - a(n,n». {3(n,k) and o(n,k) are the
same as a(n,k) and ,,(n,k) respectively, except that the left branch is a
double bond. By convention a(O,O) and ,,(0,0) represent a point.

For the vertices of the polytope a(n, k) it is convenient to take the

points in Rn + 1 whose coordinates are all the permutations of (on-k+I, 1k)
[Cox20, §8.7). The centroid of a(n,k) is the point
_k_ (In)
n+I '
466 Chapter 21

and so the circumradius is

Similarly
rHn,k) has vertices (on-k+l,±l k ), Rp(n,k) =.Jr,
.
'Y (n, k) has vertices
12k) R -y (n, k) --
(on-k '" -J4k (nn +
kHn
I'
Hn,k) has vertices (on-k,±I,±2k ), R~(n,k) = .J4k+1 .

(These polytopes appear with different names in [Cox221.) Each of these

polytopes has two kinds of faces, obtained by deleting either the left or the
right node of its diagram [Cox20, §§ 7.6, 11.6, 11.71. For example deleting
the left node of the a(n,k) diagram produces an a(n -I, k - I), while
deleting the right node produces an a(n-I,k). Thus in general a(n,k)
has faces of type a(n-I,k-I) and a(n-I,k). (If k=O the first type is
absent, while if k = n the second type is absent.) The number of faces of
each type is given by the ratio of the orders of the underlying Weyl groups
(obtained by ignoring the r.ings on the diagram). Thus the number of
a(n-I,k-I)-type cells of an a(n,k) is
I[3 n - I JI _ (n+l)! _
I[3 n - 2 JI n! -n+l.

This is also the number of a(n-I,k)-type faces. We represent this process

of finding the faces by the graph shown in Fig. 21.12.

a(n -I,k-I) a(n-I.k)

Figure 21.12. The polytope a(n,k) has n+1 faces which are
a(n-l,k-l)'s and n+l faces which are a(n-l,k)'s.

p(n,k)
/ I
/ I
2n/ 12n
/ I
/ I
/ I
p(n-I,k -I) a(n-I,k)
Figure 21.13. fJ(n,k) has 2n faces which are ,8(n-l,k-l)'s and 2" faces
which are a(n-l,k)'s. (We use broken lines here to make the structure
of Figs. 21.14, 21.15 more visible.)
Voronoi Cells of Lattices and Quantization Errors 467

We can now apply Theorem 3 to a(n,k), obtaining

(n+I)h L ( ) (n+I)h
R
Va(n,k) = Va n-I,k-I + --....;;.;;... Va(n-I,k),
n n

where

the subscripts on h standing for left and right. If we write

.In+l
Va(n,k) - va(n,k) --,-,
n.
.In+l (32)
Ua(n,k) = ua(n,k) (n+2)!

then Va and Ua are integers satisfying the recurrences

for n ~ 2 and 1 ~ k ~ n, with va(n,O)

vaG,I) = 1, and
= va(n,n+I) = ° for n ~ 1,

ua(n,k) - (n-k+I)3 va(n-I,k-I) + k 3 va(n-I,k)

+ (n-k+I) ua(n-I,k-I) + k ua(n-I,k), (34)

for n ~ 2 and 1 ~ k ~ n, with ua(n,O) - ua(n,n+I) - for n ~ 1,

u/I,I) = 1. The first few values of Va and U a are shown in Table 21.3.
°
With the help of [S102J the Va may be identified as Eulerian numbers
[RioI, p. 2151, and are given by
k . n+I
va(n,k) = ~ (-1)1 ( . ) (k-j)" . (35)
j-o J

There is a more complicated formula for ua(n,k) which we omit.

Similarly for the polytope fJ(n,k) we have the graph shown in
Fig. 21.13, and writing
2"
Vp(n,k) = vp(n,k) - , ' (36)
n.
468 Chapter 21

we obtain the recurrences

(38)

for n ~ 2 and 1 ~ k ~ n, with vp(n,O) = 0 for n ~ I, vpO,I) = 1, and

up(n,k) = k 3(n+l)v,,(n-l,k) + ku a(n-l,k)
+ n 2 (n+l)vp(n-l,k-I) + nup(n-I,k-I), (39)

for n ~ 2 and I ~ k ~ n, with up (n,O) = 0 for n ~ I, up(I,l) = 2 (see

Table 21.3). Furthermore one can show by induction that
k
vp (n,k) = ~ v,,(n,i), (40)
i-I

which implies vp(n,n) = n!. Since the v,,(n,k) satisfy va(n,k) =

v,,(n,n-k+l) it follows that

For 'Y(n,k) and O(n,k) we have a pair of graphs similar to Figs.

21.12, 21.13 (simply replace a by 'Y and (3 by 0). As before we set
.In+1 .In+1
V'/n,k) = v./n,k) - - ,- , U../n,k) = u./n,k) ( )"
n. n+2.
2n 2n
Va(n,k) = va(n,k) ~' Ua(n,k) = ua(n,k) (n+2)! '

and obtain the recurrences

v./n,k) = (2n-2k+I)v./n-l,k-l) + (2k+I)v./n-I,k), (42)

for n ~ I and 0 ~ k ~ n, with v./n,-I) = v./n,n+l) = 0 for n ~ 0,

v,,(O,O) = I;

u,,(n,k) = + (2k + 1)3 v./n-l,k)

(2n-2k+l)3 v ../n-l,k-I)
+ (2n-2k+l) u,,(n-l,k-l) + (2k+I) u,,(n-I,k), (43)

for n ~ 1 and 0 ~ k ~ n, with u,./n,-I) = u,./n,n+l) = 0 for n ~ 0,

u./O,O) = 0;

for n ~ 1 and 0 ~ k ~ n, with va(n,-I) = 0 for n ~ 0, va (0,0) = I;

ua(n,k) = (2k+l)3(n+I)v'/n-l,k) + (2k+I)u,,(n-l,k)
+ 8n 2 (n+l)va(n-l,k-I) + 2n ua(n-I,k-l), (45)
Voronoi Cells of Lattices and Quantization Errors 469

Table 21.3. The first few values of va(n,k), ua(n,k), v~(n,k) and
u~(n,k). The diagonals correspond to k = 1,2, ....

1 57 302 302 57 1
1 26 66 26 1
11 11 1

6 1158 8916 8916 1158

400 1290 400 5
116 116
24

1 58 360 662 719 720

1 27 93 119 120
4 12 23 24
3 6

12 2568 28848 69624 80388 80640

10 950 5490 8250 8400
312 880 960
84 120
16

for n ~ 1 and 0 ~ k ~ n, with ub(n,-l) = 0 for n ~ 0, Ub(O,O) = 0

(see Table 21.4). Also

(46)

The j-dimensional faces of a(n,k), ... ,O(n,k) for any j can be found from
Figs. 21.14, 21.15.
470 Chapter 21

Table 21.4. v-y(n,k), u-y(n,k), v.(n,k) and u.(n,k). The diagonals

correspond to k = 0,1, ....

1 237 1682 1682 237 1

76 230 76
23 23

10065 124338 124330 10065

2416 10520 2416
477 477
60

238 1920 3602 3839 3840

77 307 383 384
24 47 48
V6 (n ,k)

10 20840 369740 954120 1074490 l075200

8 5224 41240 61048 61440
6 1140 3654 3840
188 256
16

The special cases we are most interested in are {3 (2t ,t) and {) (2t + 1, t),
the Voronoi regions for n;1 and n;I+1 respectively. For n;1 we have
221
Vol (V(O» = vfl(2t,t) - - = 221 - 1
(2t) !
221
U(V(O» = ui 2t ,t) (2t+2)!'
• ui2t ,t)
G (D21) = 2 1/1 ( )' p = 1, (47)
2- t 2t+2 !

and covering radius

Voronoi Cells of Lattices and Quantization Errors 471

"'! /'
0(4,1) ,8(4,1)

0(3,1)
0(4,2)

/~ i / '
,8(3,1)
,8(4,2)

0(3,2)
0(4,3)

/~! /'
,8(3,2)
,8(4,3)

0(3,3)
0(4,4)

~i/Xi/0(2,1) ,8(2,1) 0(2,2)

~!/ 0(1,1)
Figure 21.14. Interconnections between the a(n,k) and (:J<n,k).

y(3,O) 8(3,0) y(3,1) 8(3,1)' y(3,2) 8(3,2) y(3,3)

" ' ! / /~!/

y(2,O) 8(2,0) y(2,1)
/>(i/
8(2,1) y(2,2)

~l/ y(I,O) 8(1,0)

Xl/ y(I,I)

~l/ y(O,O)

Figure 21.15. Interconnections between the 'Y(n,k) and o(n,k).

For this version of D;I+1 (which differs by a scale factor from the
definitions given in §7.4 of Chap. 4) we have
221 +1
Vo/(V(O» = v6(21+1,t) ( ) = 241 +1 ,
21+1 !
221 + 1
U(V(O» = u6(21+1,t) (21+3)!'

u6(2t+ 1, t)
( . ) =-:---....:....-----::-;-:-
GD (48)
21+1 (21+1) (21 + 3)! 2/ (1)'
where
f(t) = 2(21 2+51+1)
21+1 '
p = .J3 Of 1 = 1), p = 2 Of 1 > 1),
R = R 6 (21+1,t) - .J41+1
= p$ Of 1 = 1), p.Jl + 1/4 Of 1 > 1) .
For example D; ;;: A; is the body-centered cubic lattice, the Voronoi cell
is a truncated octahedron (Fig.2.2b), and G(D;) = 19/(192 3..Ji) On
agreement with (13». Also
GW;) = 13/(120..Ji) = G(D 4 ) ,

2641
G W;) = 3/5 = 0.0756254 ... , (49)
23040·2
601 . 21/3
G(D~) = 10080 = 0.0751203 .... (50)
472 Chapter 21

The values of G (D;) are plotted in Fig. 2.9. The minimal value is
0.0746931... at n = 9.
3.F. Voronoi cell for A;.
Theorem 7. The Voronoi cell for the lattice A; is the polytope Pn defined
in Fig. 21.16. If we rescale A; by multiplying it by n + 1, the vertices of
the Voronoi cell consist of the (n + 1)! points obtained by permuting the
coordinates of

= [-n -n +2 -n +4 n -2 !!...]
(J 2' 2 ' 2 , ... , 2 ' 2 .

Pn is sometimes called a permutohedron ([BowIJ, [CoxlS, pp.72-73],

[Cox23, p. 574]), while (J itself is the Weyl vector for An [Cox25].
Proof It is easy to check that (J is equidistant from the walls of the
fundamental simplex S for An; i.e. that (J is the in center of S. Let P be
the convex hull of the images of (J under W(A n ). Since the walls of S are
reflecting hyperplanes for Wa (An), P and its images under Wa (An) tile
Rn. Thus P is the Voronoi cell for some lattice A ~ A;. But A must
contain all the glue vectors [iJ for An in A; (§6.1 of Chap. 4), since these
are the images of 0 in the walls of P. Since these points span A;, A = A;.
The second moment of P n may be found as follows. First, the covering
radius of A;, R (n) say, is the circumradius of Pn , which is

R(n) =..ru:; = {~ (nj2) rz = Rn (say),

= p{ (n +2)/3}1IZ,

since now p = .In(n+I)/4, and the volume of Pn is Vn = (n+I)n-1IZ. Let

In = [(Pn) be the normalized second moment. A typical face of Pn is
obtained by deleting say the rth node from the left in Fig. 21.16, and is a
prism P r x P s with r + s = n-l (see Fig. 21.17). The number of such
faces is

(52)

Furthermore I (Pr x pJ = Ir + Is.

Let hrs be the height of the perpendicular from the center of P n to a
typical face Pr x P s . Then (see Fig. 21.1S)

h/s = R(r+s+I)z - R(r)Z - R(s)Z

(r + 1) (s + 1) (n + 1) . (51 )
4 ' usmg .
Voronoi Cells of Lattices and Quantization Errors 473

@----@-@--@--@ Pn

Figure 2l.16. Voronoi cell p. = V(O) for A;. There are n nodes, all
ringed.

6 6

Figure 2l.17. The three types of faces of P s.

CE TER
OF Pn

Figure 2l.18. Calculation of height hrs of perpendicular from center of

p. to face Pr x Ps •

We now apply Theorem 3, obtaining

1 n-I (n+l)
In Vn = n+2 ~ r+l
,-0

(with r+s = n-I), which, if we write I n = In-dn, becomes

J = I n-I
~ (
n ) r' S S (!!. J +
+ _' J
_ S ) .
n 2(n+I)nn-I,~ r 4 s r

where r+ s =n-2. Using Abel's identity [Rio2, §1.51 to simplify the first
term, this becomes

(53)
474 Chapter 21

for n ~ 2, with J I = o. The first few values are as follows:

n: 2 3 4 5 6 7
5 19 389 1045 78077
I n: 0
12 18 32 375 648 33614

Finally, the dimensionless second moment of the Voronoi cell of A; is

(54)

The values for n ~ 9 are plotted in Fig. 2.9. The curve is extremely fiat,
the minimal value 0.0754913 ... occurring at n = 16.
G. The walls of the Voronoi cell. As explained in §1.2 of Chap. 2, the
walls of the Voronoi cell V(O) for a lattice A are defined by the relevant
vectors of A. For a root lattice we saw in the Corollary to Theorem 5 that
the relevant vectors in a root lattice are precisely the minimal vectors. In
this case there is a simple description of V(O).
Theorem 8. If the relevant vectors are precisely the minimal vectors, then
V(O) is the reciprocal of the vertex figure of A at the origin.
Equivalently. V(O) is (on a suitable scale) the reciprocal of the polytope
whose vertices are the minimal vectors of A.
Proof This follows immediately from the definitions of vertex figure and
reciprocal polytope - see [Cox20).
The following is a useful sufficient condition for a lattice to have this
property. Let us write A - Ao U Al U ... , where the ith shell Ai
consists of all v E A with v . v = Xi (say), and 0 = Xo < XI < ....
Theorem 9. Suppose that (i) Ar C Al + Al + ... + Al (r times). and
00 rX I ~ Xr • for all r = 1, 2, ... Then the relevant vectors are precisely
the minimal vectors.
Condition (i) states that Al spans A, and moreover does it economically in
the sense that any vector in Ar is the sum of not more than r vectors of AI.
In practice this condition is very easily checked by induction. An
important class of lattices satisfying (i) are those obtained by applying
Construction A to a linear binary code with minimal distance ~ 4 that is
spanned by the codewords of minimal weight, and if d < 4 has the
additional property that no coordinate of the code is always zero.
Proof of Theorem 9. Suppose the contrary, so that there is a point u E Ar
with r > 1 and a point x E Rn such that x· u > Y2X r , but
x·v ~ lhX I for all v E AI. From 0), u = ~nivi with Vi E AJ, ni > 0
and ~ni ~ r. Then x ·u = ~ni(x ·Vi) ~ lhX I ~ni ~ lhrXI ~ lhX r ,a
contradiction.
Voronoi has given a simple characterization of the relevant vectors in
any lattice (see [Vorl, Vol. 134 (1908), p. 277] and [Ven5]).
Voronoi Cells of Lattices and Quantization Errors 475

Proof Note that each nonzero vector v E A determines a half-space

Hv = {x ERn: x·v ~ 112V·V}, and V(O) is the intersection of these half-
spaces. (Only if) Suppose v, w satisfy v == w (mod 2A), v ¢ ± w,
N (w) ~ N (v). Then I = Y2 (v + w ), U = Y2 (v - w) are nonzero vectors in
A. If x E HI n Hu then x·1 ~ IIzI·I, x·u ~ 1/2U·U, implying
x·v ~ 1/2 v·v, and so Hv is not needed to define V(O), i.e. v is not relevant.
(If) Suppose v is not relevant. In particular, the point Y2 v must lie on or
outside some H w for w ¢ 0, w ¢ v, i.e. Y2 V·W ~ 1/2 w·w . Therefore
N (v - 2w) ~ N (v). Furthermore v - 2w ¢ ± v and is also in the coset
v + 2A.
For example, for the Leech lattice we see from Theorem 2 of Chap. 12
that the Voronoi cell is bounded by the 196560 + 16773120 = 16969680
hyperplanes corresponding to the vectors of norm 4 and 6.
Acknowledgements. During the early stages of this work we were greatly
helped by several discussions with Allen Gersho. Some of the calculations
were performed on the MACSYMA system [Mat3]. We should also like
to thank E. S. Barnes and H. S. M. Coxeter for their comments.
 22
A Bound for the Covering Radius
of the Leech Lattice
S. P. Norton

This chapter describes a method for bounding the covering radius of a

lattice. When applied to the Leech lattice it gives an answer very close to
the true value.
The problem of determining the covering radius of the Leech lattice is
solved in the following chapter by Conway, Parker and Sloane. Before the
discovery of their line of approach, the following close approximation to the
true answer was obtained. This is considered worth recording not only for
historical reasons, and because the proof is very short, but also because the
method is not dependent on the properties of the Leech lattice and is
therefore applicable to other lattices.
The only property of the 24-dimensional Leech lattice that we use is
that the midpoint of the line joining any pair of lattice points satisfies one
of the following conditions.
(a) It is itself a lattice point.
(b) It lies halfway between two lattice points of type 2 or 3 relative to
each other (i.e. distance 2 or .J6, assuming, as we do from now on, that the
lattice is unimodular), with all other lattice points being further away.
(c) It is the midpoint of 24 mutually orthogonal lines joining pairs of
lattice points of type 4 (distance 2../2), so that we can choose a Cartesian
coordinate system with it at the origin, where all points whose coordinates
are a permutation of (±../2, 0 23 ) are lattice points.
From this property (Theorem 28 of Chap. 10), we derive the following
result.
Theorem. Every point in 24-space is within a distance (2k) 1/2
= 1.4518442... of some lattice point. where k = (37 - J73) /27 =
48/(37 + J73) = 1.0539258 ....
A Bound for the Covering Radius of the Leech Lattice 477

The distance (2k) 1/2 is therefore a bound for the covering radius, since
spheres with this radius, centered on lattice points, will cover the entire
space. (In the following chapter it is proved that the actual value of the
covering radius is..fi.) The packing radius of this lattice is 1.
Proof We start by putting a normed space structure on our 24-space,
with the zero point as one of the lattice points, and the norm of a vector as
the distance of the corresponding point from the zero point.
Lemma. If (the point corresponding to) 2v satisfies the result of the
theorem. then so does v.
Proof By hypothesis there is a vector u, with 2u in the lattice, such that
112v-2u II ~ 2k, so that II v-u II ~ Ihk. Now since u is the midpoint of
the line joining 0 to 2u, it satisfies one of (a) -(c) above. In case (a), u is
a lattice point, and v is obviously within the required distance of it. In
case (b), let x and y be the two lattice points closest to u, so that
2u =x+y. The last equation, together with lIu-xll ... 1f.Illx-yll = lor
3/2, implies that Ilv-xll + Ilv-yll =2(lIv-ull + Ilu-xll) ~ k+3 < 4k,
from which the result of the theorem follows, since one of Ilv-x II, Ilv-y II
will be less than 2k. In case (c) we apply a translation that takes u to the
origin (and v to say v' = v-u), and use the coordinate system defined in
the condition. Let x' be one of the 48 lattice points named in this
condition with the greatest inner product with v', and let this inner product
be a. Then Ilv'll ~ 12a 2, and Ilv'-x'll = Ilv'll + 2-2a, as v'· x' = a
and II x' II = 2. From this we can show by computation that the hypothesis
IIv'll ~ Ihk implies Ilv'-x'll ~ 2k, so that v is within the required
distance of a lattice point. This proves the lemma. It turns out that the
peculiar-looking value of k is the least that allows the final deduction.
To complete the proof of the theorem, we note that the lemma implies
immediately that every dyadic rational point satisfies the result of the
theorem, which is automatically true for lattice points. But it is obvious
that the distance from the nearest lattice point is continuous on 24-space,
and the inverse image of [0, 2k 1, being the set of points satisfying the
result of the theorem, will be both closed and dense (as the dyadic
rationals are dense in R). The theorem now follows directly.
Postscript. The method described here has proved even more successful
when applied to other lattices. Illustrations may be found in §5 of Chap. 6
and in [Con361, [Con371, [Con43].
 23
The Covering Radius of the Leech Lattice
J. H. Conway, R. A. Parker and N. J. A. Sloane

We investigate the points in 24-dimensional space that are at maximal

distance from the Leech lattice, i.e. the "deep holes" in that lattice. The
distance of such a point from the Leech lattice is l/.Ji times the minimal
distance between the lattice points. Furthermore there are 23 inequivalent
types of deep hole, one for each of the 23 even unimodular 24-dimensional
lattices found by Niemeier.

1. Introduction
The main result of this chapter is the following, which confirms a
conjecture made by Leech soon after his discovery of the lattice.
Theorem 1. The covering radius of the Leech lattice is .Ji times the
packing radius.
The covering radius R is defined in Eq. (3) of Chap. 2. An upper bound
of R ~ 1.452 ... p (where p is the packing radius) was obtained by Norton
in [Nor4] (= the previous chapter). In the course of proving Theorem 1
we shall classify all the "deep holes" (defined in §1.2 of Chap. 2) in the
Leech lattice. There is a remarkable connection between these holes and
the Niemeier lattices. The Leech lattice is the unique 24-dimensional even
unimodular lattice with minimal norm 4 (Chap. 12), and Niemeier
([Nie2]; §1 of Chap. 16) found that there are 23 other 24-dimensional
even unimodular lattices, all with minimal norm 2.
Theorem 2. There are 23 inequivalent deep holes in the Leech lattice
under congruences of that lattice, and they are in one-one correspondence
with the 23 Niemeier lattices (see Table 23. J).
An outline of the remainder of this chapter is as follows. In §2 we
associate a Coxeter-Dynkin diagram (§2 of Chap. 4) to any hole in the
Leech lattice. Two kinds of diagrams occur; those for which all
components are ordinary Coxeter-Dynkin diagrams, and those that contain
The Covering Radius of the Leech Lattice 479

Table 23.1. The 23 types of "deep hole" in the Leech lattice. Here h -
Coxeter number, V - number of vertices of hole. Note that the
component of highest dimension identifies the hole uniquely.
Components h V Fig. Components h V Fig.

D24 46 25 23.22 A§ D6 10 27 23.15

DJ6 E 8 30 26 23.29 D2 10 28 23.25
Ei 30 27 23.30 Ai 9 27 23.14
A24 25 25 23.11 ArD~ 8 28 23.13
Df2 22 26 23.28 A2 7 28 23.12
AJ7 E 7 18 26 23.19 At D4 6 29 23.24
DIOEr 18 27 23.27 D3 6 30 23.23
AJ5 D 9 16 26 23.18 A3 5 30 23.7
Di 14 27 23.26 A~ 4 32 Th. 11
Af2 13 26 23.17 AF 3 36 Th.l0
A 11 D7 E 6 12 27 23.16 A f4 2 48 Th. 8
E2 12 28 23.31

an extended diagram as a subgraph. The holes of the first kind have

radius less than .J2 times the packing radius, and can therefore be ignored
(but see Chap. 25), while the second kind have radius greater than or
equal to .J2 times the packing radius. In §§3-5 we shall determine all
holes of the second kind, without any further assumption about the radius
of the hole. We do this by systematically classifying all holes whose
diagram contains a subgraph of type AJ, A 2, ... , A 24 , DJ, D 2, ... , D 24 , E 6,
E 7 or E 8. At the conclusion of this process we find that all these holes
have radius equal to .J2 times the packing radius, and so the proof of
Theorem 1 is complete.
Furthermore, precisely 23 inequivalent holes of this kind occur, and
their Coxeter-Dynkin diagrams exactly describe the Witt lattice
components of the 23 Niemeier lattices, completing the proof of
Theorem 2.
Certain sublattices of the Leech lattice (the Y-Iattices, defined in §3)
were classified by Curtis [Cur2!. In the course of proving Theorems 1 and
2 we also find all sublattices of the Leech lattice that contain a set of
points whose distances correspond to an extended Coxeter-Dynkin diagram.
The sublattices corresponding to A 5 , A 6 , ••• have an especially simple
structure that is described by the An-tree of Fig. 23.9, while those
corresponding to D 5, D 6, ... are described by the Dn-tree of Fig. 23.20.
Further information about these sublattices is given in §§3-5.
We have tried to keep this chapter as short as possible. But for future
applications it has seemed desirable to give explicit coordinates for the
vertices and center of a hole of each type.
480 Chapter 23

2. The Coxeter-Dynkin diagram of a hole

The Leech lattice will be denoted by A. The norm N (x) = x . x of a
vector x is its squared length, while its type is lj2 N (x). We choose the
scale so that the first four shells in A, the so-called short vectors (see
Chap. 12), consist of 1 vector of type 0, 196560 vectors of type 2,
16773120 vectors of type 3, and 398034000 vectors of type 4. The type 2
vectors can be taken to have the shapes
{j «±4)2,022),
where the ±2's occupy an octad and have
positive product,
{j (=t=3,(± 1)23), where the entries congruent to 1 modulo 4
occupy a ~-set,
and {j = 1/-/8; while the type 3 vectors are

where the ±2's occupy an octad and have

negative product,
where the ±2's occupy a dodecad and have
positive product,

where the entries congruent to 1 modulo 4

occupy a ~ -set,
where the entries congruent to 1 modulo 4
occupy a ~ -set.
The packing radius of A is now 1. As in the previous chapter, Theorem 28
of Chapter 10 plays a central role. We restate it as follows.
Theorem 3. Every vector in A is congruent modulo 2A to a short vector.
Furthermore the only congruences among the short vectors are that each
vector of type 2 or 3 is congruent to its negative, while the vectors of type
4 fall into congruence classes of size 48, each class consisting of a
coordinate frame (i.e. a set of 24 mutually orthogonal pairs).
Let (c, p) be a hole in A of radius R, and let the vertices of P be
The Vi are in A, so N(Vi-Vj) = 4, 6, 8, ... for i ¢ j. On the other
Vlo ••• , Vp.
hand N(Vi-Vj) ::::;; 8, for if N(Vi-Vj) ~ 10 then Theorem 3 implies that
the midpoint 'h(Vi+Vj) is also the midpoint of a pair x, x' E A with x-x' a
short vector, and then at least one of x or x' is closer to c than Vi and Vj
are. Thus N(Vi-Vj) = 4, 6 or 8.
The diagram for this hole is constructed as follows. There is one node
(also labeled Vi) corresponding to each vector Vi. Two nodes Vi, Vj are
(i) not joined (or disjoined) if N(Vi-V) = 4,
(ij) joined by an edge if N(Vi-Vj) = 6, or
(iii) joined by two edges if N(Vi-Vj) = 8.
The Covering Radius of the Leech Lattice 481

The resulting graph with these nodes and edges is the hole diagram. By
the end of the chapter we shall have proved that this diagram is actually a
Coxeter-Dynkin diagram, or more precisely that the following theorem
holds.
Theorem 4. The diagram for a hole in the Leech lattice is a graph whose
connected components are taken from the list an (n ~ 1), d n (n ~ 4), e6,
e7, eg, An(n ~ 1), Dn(n ~ 4), E 6, E 7, Eg (see Fig. 23 .1) .
The components denoted by lower case letters are called ordinary
Coxeter-Dynkin diagrams, the others, extended. (In this chapter the terms
ordinary and extended diagram will always refer to the connected graphs
shown in Fig. 23.1.)
The following construction is helpful when studying a hole (e, p). We
embed the space in which A lies in R 25 , with A contained in a hyperplane
H, and choose a point c' on the line through e perpendicular to H as
follows. If the radius R ~ J2 we take c' = e, but if R < J2 we choose e'
so that its distance from the vertices of P is J2 (Fig. 23.2) . If R < J2
two nodes Vi, Vj in the hole diagram are not joined, or are joined by one or

an a 0 0 0--.. -0----0 (n NODES)

d. ~._~ (.NODES)

e~~ e7~
ea o~~--~--~o--~l__~o--~o
,
A 1\ 1 A... -
2,L...1 10'
, 1

, 1

On ~~ (n+' NODES>
2 2 2 2 2
,
'~
ES 1 2 3 2 1
E7 ,
0 0
2
0
3
r
4
0
3
0
2 ,
0

Ea
'2345642
Figure 23.1. Ordinary Coxeter-Dynkin diagrams an (n ~ 1),
dn(n ~ 4), e6. e7. e8 and extended Coxeter-Dynkin diagrams An(n ~ n,
Dn (n ~ 4). E 6. E 7, E 8. The latter are labeled with the integers Cj .
482 Chapte r 23

Figure 23.2. Construction of the point c'.

two edges according to whether the inner product (Vi-C', Vj-c') is 0, -lor
-2 respectively (see Fig. 23 .3).
The first step is to dispose of the case when the diagram contains no
extended diagram as a subgraph.
Theorem 5. A hole diagram with no extended diagram embedded in it
contains only ordinary diagrams as components.
Proof This is proved by an elegant and elementary combinatorial
argument (given on page 195 of [Cox20, §11.5]).
Theorem 6. The hole corresponding to a diagram in which all components
are ordinary has radius R < .Ji.
Proof Such a diagram also describes a root system in which all the roots
have the same length [Boul], [Humn Let us choose a fundamental set
(or base) of roots, i.e. a set of linearly independent vectors V" .. ., Vv in an

II,
o

...---./s - -- ...
oo--~---oo--~--~o
IIi ./2 c' ./2 II,

(a) (b) (e)

Figure 23.3 . The conditions for a pair of nodes in the diagram for a hole
of radius R < .J2 to be (a) disjoined, (b) joined by an edge, or (c)
joined by two edges.
The Covering Radius of the Leech Lattice 483

appropriate Euclidean space S, of constant length and having the same

mutual distances as VI> ••• , VV' Therefore the set of points {Vi} is isometric
to the set {Vi}. Since the origin in S is not a linear combination of the Vi,
it follows that the corresponding point c' is not in the hyperplane
containing the Vi' Thus R < J2, by definition of c'. This completes the
proof.
On the other hand, if VI> ••• , V I' are a set of fundamental roots
corresponding to an extended diagram, there are positive integers CI> ••• , C I'
such that}; CiVi = 0 ([BouI], [Cox20, p. 194], [Huml, p. 194]). These
integers are shown in Fig. 23.1. If this diagram occurs as a subgraph of a
hole diagram, VI> ••. , VI' are the corresponding vertices and c is the center of
the hole, then}; Ci (Vi-C) = O. Thus the center can be found from
I' / I'
C = ~ CiVi ~ Ci'
i-I i-I

V
Suppose we can find points v[1l, ... , ~I/ in the Leech lattice A whose
mutual distances define an extended diagram Ill> and such that their
circumcenter c is distant J2 from all of them. The next theorem, which is
essential in all that follows, asserts that under certain conditions c is the
center of a unique hole in A of radius J2.
Theorem 7. Suppose c E R24 is distant J2 from Leech vectors v[1l •....
(I). (2) (2).. (s) (s)
VI'I ' VI • . . • • V1'2' .•• , VI • . . . • vI's suc
h th at th e vec tors Vj(0 - c span
R24. and. for i = 1•... • s. the mutual distances of v?). '" .v~: define an
extended diagram Il i . Then there is a unique hole in A with center c and
radius J2. whose vertices are precisely the vJo and whose hole diagram
has components Ill. 1l 2• . . • • Ils. In this case we say that the hole is of
type 1l 11l 2 · .. ll s .
Proof We first show that there is a hole of radius J2 centered at c, and
the vJo are all its vertices. Suppose z EA is distinct from the vJO and
satisfies N (z-c) ~ 2. For any vJO we have

4 ~ N(z - vJO) (since both are in A)

= N (z - c - (vJO - c»
~ 2 - 2(z - c) . (vJO - c) +2,
so
(z - c) . (vJO - c) ~ 0 .

Let Zi be the projection of z onto the subspace of R24 spanned by {vJo - c:

). = 1•...• I-li' } S'mce t he Vj(0 - c span R24 we can wnte . z- c-
(Zl • . . . • zs), and from (2) Zi . (vJO - c) = (z - c) . (v!O - c) ~ O. As
mentioned at the end of §2 there are positive integers cl) satisfying}; c}O
(vJO - c) = o. It follows that Zi = 0 for all i, and z = c, whic~ is a
contradiction since c ~ A.
484 Chapter 23

A similar argument shows that if b is any point of R24 at a constant

distance of JR from all the vJO then b = c. Thus the center of the hole,
and therefore the hole itself, is unique.

3. Holes whose diagram contains an An sub graph

In this section we use Theorem 7 to classify all holes whose diagram
contains an An diagram as a subgraph, without any further assumption
about the radius of the hole.
Theorem 8. Any hole that has diagram containing a subgraph A 1 is of
type A f4. Furthermore there is a unique hole of this type. We express
these two statements by the formula
Al ~ Ar4 .

Proof Since Co=, the automorphism group of A, is transitive on pairs of

vectors u, v E A with N(u-v) = 8, there is essentially only one way to

of the diagram we may assume these vertices to be u = (.J2, °,. .,

choose the vertices of the first A 1 diagram. Taking the origin at the center
0), v =
(-.J2, 0, ... , 0). The origin is also at the center of 23 other A 1 diagrams,
with vertices (0, ... ,0, ± .J2, 0, ... ,0). Since these 48 vectors span R 24 , the
desired conclusions follow from Theorem 7. When represented in the usual
coordinates a center of a hole of this type has the form c = 0 (4,0, ... ,0),
where 0 = 1/J8.
Following Curtis [Cur2] we define an Y-Iattice to be a sublattice L of
A such that every vector of L is congruent modulo 2A to a vector in L of
type 0, 2 or 3. An Y-lattice of type 2i 3j contains 2i vectors of type 2 and
2j of type 3.
Theorem 9. ([Cur2]). There are 12 types of Y-Iattice. and each occurs
uniquely (up to isomorphism) inside A.
Theorem 10.

Proof Consider a hole diagram containing an A2 subgraph with vertices

u, v, w. Then the vectors v-u, w-u span an Y-lattice of type 33 . By the
previous theorem we may take u = 0, v = 0(5,1 23 ), w = 0(5,1 11 , _1 12),
with center c = 0(10/3, (2/3)11,0 12). Then c is the center of 11 other A2'S
having vertices of the form
u' = 0 (4, 4, 0 10 , 0 6 , 0 6) ,

v' = 0(3, -1,1 10 ,1 6 , -1 6) ,

W' = 0(3, -1,1 10 , -1 6 ,1 6) ,

etc. These are permuted by an element of order 11 fixing u, v and w.

Again Theorem 7 completes the proof.
Theorem 11.
The Covering Radius of the Leech Lattice 485

Proof We take one vertex of A3 as the origin and label the others u. v. w.
and calculate the following inner products:
u v w u+v-w
u 6 2 4 4
v 2 4 2 4
w 4 2 6 0
u+v-w 4 4 0 8

Since u+v-w is of type 4. by Theorem 3 it belongs to a coordinate frame.

and we may assume u+v-w" 5<8. 023 ) and u" M4 •.. .). From the list of
type 3 vectors given at the beginning of §2. u = 15(4. (±2)8. 0 15). where the
2's occupy an octad and there are an odd number of minus signs. It will
be convenient from here on to write vectors in R24 in MOG format (see
Chap. 11). and to omit the factor 15.
We may now assume without loss of generality that
-22 40 00
22 00 00
u-
22 00 00
22 00 00

The subgroup of Co"" fixing the origin is the group Coo. and the subgroup
of Coo fixing u is 24A7 (An denotes an alternating group and Sn a
symmetric group). To find v we use the facts that v is a type 2 vector. its
inner product with u is 2. and that the group 24. A7 is transitive on the last
IS coordinates of u to write

00 40 40
00 00 00
v'" 00 00 00
00 00 00

and then
-22 00 40
22 00 00
w-
22 00 00
22 00 00

We conclude that there is essentially a unique A3 diagram. which we

can assume to have the vertices O. u. v. w given above. The center is

-1 1 20 20
c .. 1 1 00 00
1 1 00 00
1 1 00 00
486 Chapter 23

For the second part of the proof we observe that c is also the center of
seven other A3's such as that shown in Fig. 23.4, and which are based on
octads meeting the left-hand octad of the MOG in four places. (The
others are the images of the one shown in Fig. 23.4 under powers of the
permutation ex given in Fig. 23.8 below.} Again Theorem 7 completes the
proof.
Theorem 12.

Proof. We label the vertices O,u,v,w,x. By examining inner products (as

in the previous proof) we find that u, v, W, x generate an Y-lattice of type
2 53 10, which is unique by Theorem 9. Therefore we may assume that the
vertices are labeled as in Fig. 23.Sa, with center

4 4 2 6 6 6
8 4 2 2 2 2
C = (I/S)
4 4 6 2 2 2
4 4 6 2 2 2

This is also the center of five other A/s, namely the one shown in
Fig. 23.Sb and its images under the following element of order S in Coo

bo Co Cz C3
ao do b4 b l
a3 a4 C4 d3 d4 bz
az al Cl dz b3 d l

(This has four fixed points indicated by dots and four cycles (ao • ...• a4),
(b o• ...• b 4 ) • . . . • (do . .... d 4 ).)
The remaining An's can be handled uniformly. Consider a hole
diagram containing an An' where n is large, labeled as in Fig. 23.6 and

0 2 2 2 2 2 2 0 2 2 -2 2
2 0 0 0 0 0 0 2 0 0 0 0
2 0 0 0 0 0 0 2 0 0 0 0
2 0 0 0 0 0 0 2 0 0 0 0

-2 0 2 2 2 -2 0 2 2 2 -2 -2
0 2 0 0 0 0 2 0 0 0 0 0
0 2 0 0 0 0 2 0 0 0 0 0
0 2 0 0 0 0 2 0 0 0 0 0

Figure 23.4. One of seven other A /s used in the proof of Theorem 11.
The Covering Radius of the Leech Lattice 487

(a) 1 1 1 1 1 1

/
5 1 1 1 1 1
1 1 1 1 1 1
1 1 1 1 1 1

00 o 0 00 2 2 o 2 2 2
00 o 0 00 00 2 0 o 0
00 00 00 o 0 2 0 00
00 o 0 00 00 2 0 00
,/'
00 o 2 2 2 1 1 1 1 1 1
2 2 2 0 o0 1 1 -3 1 1 1
2 2 2 0 o0 1 1 1 1 1 1
2 2 2 0 o0 1 1 1 1 1 1

(b) o0 20 2 2
2 2 02 o 0

/
00 2 0 o 0
00 2 0 o 0

1 1 -3 1 1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1 1 -3 1 1
1 1 1 1 1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1 1 1 1 1

/
02 2 2 2 2 20 o 2 00
2 0 o 0 00 2 0 o 2 00
20 o 0 00 o2 2 0 o0
20 o 0 o 0 o2 2 0 o0
Figure 23.5. Some vertices of the A: hole.

with Vo - O. As in Theorem 11 we examine the inner products and

discover that V\ + V2 - V4 is of type 4, and so can be taken to be /l{8, 0 23 ).
Proceeding as before we find that VI ••••• V5 can be taken to be those
shown in Fig. 23.7. The subgroup of Coo fixing VI ••••• V5 is a PSL 2 (7)
generated by the permutations a, {3, 'Y shown in Fig. 23.8.

Figure 23.6. Labels for the nodes of an An diagram.

488 Chapter 23

00 00 00 5 1 1 1 1 1 40 -4 0 00
00 o 0 o 0 1 1 1 1 1 1 000000
00 00 00 1 1 1 1 1 1
o000000
0 00 0 0 1 \
00 00 o 0 1 1 1 1 1 1
112

40 o 0 4 0
00 00 o 0
00 o 0 o 0
1
1
1
1
1
1
-3
1
1
1
1
1
1 1
1 1
1 1
3 -1 1 1
-1 -1 1 1
-1 -1 1 1
1
1 11/
1 1
00 00 o 0 1 1 1 1 1 1 -1 -1 1 1 1 1

Figure 23.7. The beginning of the An-tree.

ex
111111
fJ
Figure 23.8. Generators for PSL 2 (7).

From this it follows that each Vm in Fig. 23.6 for m ~ 6 is of the form
3 -1
- [ijk] (say) ,

where i, j, k are the locations of the -1's in the left octad with the
following labeling:
000
3 2
5 1
6 4

Furthermore N(vs - [ijk)) - 6 if and only if [ijk] is one of [124], [235],

[346], [450], [5611, [602] or [013] (corresponding to the lines in the
projective plane of order 2 with the usual labeling). Also we have
N([ijk] - [rst)) - 6 if and only if {i,j,k} n {r,s,t} - flJ.
These remarks make it easy to determine the subsequent 'm. Without
loss of generality we may take
3 1 -1 -1 -1 -1
1 -1 1 1 1 1
'6 - [124] ... 1 -1
1 1 1 1
1 -1 1 1 1 1
The Covering Radius of the Leech Lattice 489

3 1 -1 1 1 1
-1 1 -1 1 1 1
" - [356] - -1 1 -1 1 1 1
-1 1 -1 1 1 1

'8 - [012] and [345]. Now only the identity element of Coo fixes
'9 -
'I, ... ,'9. and the remaining 'm
are not unique (see Fig. 23.9). We
summarize these results in a tlteorem.
Theorem 13. All possibilities for the vertices '0,"" 'n-I of an An
diagram are displayed in the An -tree of Fig. 23.9.

THE An - TREE

UtO
Utt

UI2 [125] [256]

Ut 3 (046) [134] [236]
Ut4 [135] [025] [145] [146]
uI 5 [2~ ~46] [023] [023]
uI6 [015J [0231 [146J [145]
uI7 [236] [145J [025] (236)
Ute [145] [236J [1341 [015]
u19 [023J [015J [256) [246)
u20 [146] [246] [034J [135]
u 21 [025J [I35J [125J [046]
u22 [136J [046] [046] [125]
u23 [245J [123J [123] [034]
u24 U~) =':2 U14(2) • Z2 u~(31 -z2 [256]
Figure 23.9. The A.-tree.

So far we have mentioned only n of the n+l nodes in the An diagram.

We must now consider how to choose the last node For As, ... ,A9 v:.
there is a unique choice for ,:. as follows:
5 -1 -1 1 1 1
• 1 -1 -1 1 1 1
's - 1 -1 -1 1 1 1
1 -1 -1 1 1 1
490 Chapter 23

o0
.
'6 =
4 0
o0
o0
-2
2
2
2
2
2
o0
o0
o0 2 2 o0

4 0 -2 2 2 2
o0 2 2 2 2
o0 o0 o0
o0 o0 o0

4 0 -2 0 2 0
o0 2 0 2 0
o0 2 0 2 0
o0 2 0 2 0

4 0 -20 2 0
o0 2 0 20
o0 o2 o2
o0 o2 o2

For An (n ~ 10) there are at most two choices for v;, namely

4 0 -20 20
o0 o2 o2
o0 2 0 2 0
o0 o2 o2

4 0 -2 0 2 0
o0 o2 o2
o0 o2 o2
o0 20 20

and these are also shown in Fig. 23.9. Inspection of the tree reveals that
we have proved the following theorem (the assertions about the
AI, ... ,A4 holes having been established earlier).

Theorem 14. There is a unique An diagram in A for n = I, ... ,9, 11, 12.
There are at most two AI5'S, three A\7's and five A24's, and none of type
A n forn=lO, 13, 14, 16, 18, ... ,23.
It remains to consider the holes containing these diagrams. We
postpone consideration of A5 until Theorem 18.
The Covering Radius of the Leech Lattice 491

Theorem 15.
A6 .. At (Fig. 23.10) ,

A7 .. A1 D? (Fig. 23.11) ,

As .. A~ (Fig. 23.12) ,

A9 .. A§ D6 (Fig. 23.l3) ,

All .. AllD7 E 6 (Fig. 23.14) ,

Al2 .. Af2 (Fig. 23.15) .

Proof (A l2 .. A f2)' The unique A 12 diagram obtained above is shown in

Fig. 23.15a and has center
39 3 -13 5 13 5
3 1 5 9 7 9
c = (l/l3)
1 1 7 9 11 7
3 1 7 7 7 9

To find an A 12 disjoined from this diagram we need only consider vectors

[;jk] that are not lines and that intersect the triples [i 'j k '] occurring in
I

the first A 12 in one or two places. The result is shown in Fig. 23.15b.
Theorem 6 then completes the proof. A similar argument applies to the
other cases, the corresponding centers being
21 1 -7 5 7 3
1 1 5 5 3 3
1 1 5 5 3 3
1 1 5 5 3 3

12 1 -4 2 4 2
1 0 3 3 3 3
1 0 2 2 2 2
1 0 2 2 2 2

27 3 -9 3 9 3
1 1 5 5 7 5
A~ : (l/9)
1 1 5 5 7 5
1 1 5 5 7 5

15 1 -5 2 5 2
1 0 3 3 3 3
1 0 2 4 3 3
1 1 2 3 3 4
492 Chapter 23

A~

(b) [ [012]- [345]- [016]- [234] - [056] - [123] - [456] ]

[ [014]- [236]- [145]- [036]- [125] -[034] - [256] ]

[ [025]- [136] - [024] - [135] - [246] -[035] - [146] ]

Figure 23.10. Vertices of the At hole.

-[124]- u~ ]

Figure 23.11. The Ai Dl hole.

~)
L Uo - U, - U2 - U3 - U4 - U5 -[124]-[356]-u: J

(b) , - - - - - - - - - - - - - - - _ - - ,
L[015] [236] [145] [023] [456] [123] [046] [135] [246]J

[ [016] [245] [136] [025] [134] [256] [034] [126] [345]J

Figure 23.12. The Al hole.
The Covering Radius of the Leech Lattice 493

(b) [[015] [234] [016] [245] [136] [025] [146] [023] [145] [236]J

~26] ~2~
" [034]-[125 ]-[046]/
~5~ / "~3~
Figure 23.13. The A~ D6 hole, the unique hole containing Dg,

A11
(a)
Lvo v1 v2 v3 v4 v5 [124][356][012][345][126] Z1 J

(b) [023]
I
[146] Es
I
[245 ]-[136 ]-[025]-[134 ]-[256]
[234] [123]
" [015 ]-[246]-[135 ]-[046]/ 07
[236] / "[125]

(a)

(b)

[[015] [236] [145] [023] [146] [025] [134] [256] [034] [125] [046] [135] [246]J
Figure 23.15. The A f2 hole.
494 Chapter 23

18 2 -6 2 6 3
1 0 3 4 3 5
1 0 3 4 4 4
1 1 3 3 4 3

Theorem 16. Every A 15 is disjoined from a unique D 9• and every A 17 is

disjoined from a unique E7 (examples are given in Figs. 23.16 and 23.17).
Proof The assertion was verified for both A 15 diagrams and all three A 17
diagrams found from the An-tree. (It will be seen later that there is a
unique hole of each type.) The centers of the two diagrams illustrated are
24 2 -8 3 8 3
1 1 4 7 4 5
1 0 4 6 5 5
2 1 4 4 5 5

27 2 -9 3 9 4
1 1 4 7 4 6
2 0 5 6 7 5
2 1 4 6 6 5

Figure 23.16. A hole of type A 15 D 9•

[136] AfT E7
(b) I
[145]-p23]-[146]-[025]-[ 134]-[256]-[034]
Figure 23.17. A hole of type A I7E 7.
The Covering Radius of the Leech Lattice 495

4. Holes whose diagram contain a Dn subgraph

The diagrams containing a Dn subgraph can be handled uniformly, just as
the An's were. For this purpose we label the positions in the MOG as
shown in Fig. 23.18. There are three distinguished positions labeled I, II,
III, the remaining 21 being identified with the points of a projective plane
of order 4, as in §11 of Chap. 11. We take an oval C in that plane
consisting of the points {oo, 0, I, 2, 3, 4}, and label the remaining fifteen
points with synthemes (to use Sylvester's term [Syll]) such as 000.14.23 .
Of the 21 lines in the plane, 15 are secants which meet the oval in two
points, and any such line is labeled with the duad of those two points. For
example the duad 000 meets the oval in 00 and 0, and also contains the
points
000.12.34, 000.13.24, 000.14.23

The remaining six lines are the axes of C, and do not meet the oval (see
Fig. 23.19). They are labeled with totals, a total albcdef being an
abbreviation for the set of five synthemes ad.ce.bf, ae.bc.df, af. be. cd ,
ab.cf.de , ac.bd.ef, which are the points on the axis. The vertices in the Dn.
diagrams will involve the following vectors:
0, the zero vector;
[p], where P is a point in the plane or I, II or III, has -3 at
position P and l's elsewhere;
[J>] has 5 at position P and l's elsewhere;
[L], where L is a line in the plane, has 2's at the five points of the
line and at I, II, III, and O's elsewhere;

IX) 0 IX) 0 IX) 4 IX) 2 IX) 3 1

IX)

1 4 1 3 0 3 0 1 o 2 o 4
2 3 2 4 1 2 3 4 1 4 2 3

IX) 0 IX) 1 3
IX) IX) 2 4
IX)

I 1 2 0 2 o 4 0 3 o 1
3 4 3 4 1 2 1 4 2 3

IX) 2 CD 4
II CO 0 4 0 2 1 3
1 3 1 3

IX) 3 CD 1
ill 0 0 1 0 3 4 2
2 4 2 4

Figure 23.18. Labels for positions in the MOG, based on an oval in the
plane of order 4.
496 Chapter 23

·: ru3j.:
atE
00101234

..
.. .
...
00101324

HE·
..
00101423

....

00101243 00101342 00101432

~~~
[I[IJ [IIJJ [IT]]
Figure 23.19. The six axes of C.

[n] (0:>
[c'] (05)
(O~> [0] [00101234] (O~)
[001] (07)
[000.12.34] (Oa)
[23] (Og)

[003.02.14] (010)

[001.02.34]
[02]
[004.02.13]
The Dn - tree
[13]

[003.04.12] (024)
Figure 23.20. The D.-tree.
The Covering Radius of the Leech Lattice 497

o0 o0 o0
o0 o0 o0
2 2 o0 2 2
[e] -
2 2 o0 2 2

o0 o0 2 2
[e'] -
o0 o0 2 2
2 2 o0 o0
2 2 o0 o0

2 0 2 0 o0
[.0]-
2 0 2 0 o0
2 0 o2 2 2
2 0 o2 2 2

The edges in the diagrams can be found easily from the incidence relations
in the plane. For example
N([Pl,[L)) - 6 if and only if P is on L,
N ([p l,[L)) == 6 if and only if P is not on L,
N([p],[e)) == 6 if and only if P is on e,
N([Ll,[e)) - 6 if and only if L is an axis of e,

Arguments similar to those used in the previous section now lead to the
following theorems.

Theorem 17. All possibilities for a Dn diagram are displayed in the Dn-
tree of Fig. 23.20. There are at most two distinct D/s and D6'S, the
diagrams of type D 5, D 7 , D s, D 9 , D IO , D l2 , DI6 and D24 are unique, and
there is no diagram of type Dn for n - II, 13, 14, IS, 17, ... , 23.

Theorem 18.
D4 ,.. D3 (Fig. 23.21) or At D4 (Fig. 23.22) ,
D5 ,.. Ar D~ ,
D6 ,.. Dt (Fig. 23.23) or A~ D6 ,

D7 .. AII D 7E 6'
Ds .. Di (Fig. 23.24) ,
D9 ,.. A 15 D 9 ,
DIO ,.. DIOEr (Fig. 23.25) ,
DI2 ,.. D'l2 (Fig. 23.26) ,
DI6 ,.. DI6 E S (Fig. 23.27) ,
D24 ,.. D 24 ·
498 Chapter 23

(a) ( b)
[ill] [0] [1]

~l
[a]
[iJ
[CD 0.12.34]
,
[ CD 0.14.23]

[ CD 0]
[CD 1.02.34]
,
[CD 104.23]

[CD 1]

[CD 0.13.24] [CD 1.03.24]

[2] [3] [4]

[ CD 2.04.13] [CD 3.04.12] [CD 4.03.12]

[CD 2.01.34] [CD 3.01.24] \ [CD 3] [CD 4.01.23] \ [CD 4]

[CD 2]

[CD 2.03.14] [CD 3.02.14] [CD 4.0213]

Figure 23.21. The D¥ hole, the unique hole containing D 4·

(a)
[m]

[fi]
[n]
(b) I rrnI rwnLll rl':O.l r?Lll I
[CD 2 .04.13] [CDO.1324] [CD 301.24]

[CD 2 0134] [CDO.12.34] [CD3.0412]

[CD 1] [23] [004]

[02] [14] [03]

Figure 23.22. The AtD4 hole, the unique hole containing Dr
The Covering Radius of the Leech Lattice 499

(b) [0:>1.04.23] [0:>203.14]

[r] [0:>014.23]

[0:>401.23]

[0:>2.04.13] [0:>304.12] [0:>1.0234]

[0:>0.12.34]

[0:>40213] [0:>4.0312] [0:>2.0134]

Figure 23.23. The Dt hole, the unique hole containing Dg.

[fi]
D~ [<00.14.23]

[r] [<0]
(a) ~ [<00]
[c]

[<00.12.34]
[:iIr ]

[<04.01.23] [<02.03.14]

[<03.01.24] [<01.03.24]
[24]

[<02.01.34] [<04.03.12]

[<01.02.34] [<01.04.23]

[<04.02.13] [<02.04.13]

[<03.02.14] [<03.04.12]

Figure 23.24. The DJ hole.

500 Chapter 23

[fr ] [00 203.14]

DlOE~
[r] [00] [000.1423]
(a) III [14]
[c] [000]

[fu]
[0040123]

(b) [001.03.24] [00301.24] [002.01.34]

[24] [01]
[ 001.0423]

[004.03.12] [003.04.12] [002.0413] [004.0213]

[12 ] [04] [13]

Figure 23.25. The DIOE? hole.

[001.03.24]

[c] [000]
(a)
[1J [00] [000.14.23] [002.03.14]

[m] [004.03.12]

[004.01.23] [001.04.23]

(bl [01] [34] [02] [13] [04]

[002.01.34] [001.02.34] [004.02.13] [002.04.13]

[003.01.24] [003.04.12]

Figure 23.26. The Df2 hole.

The Covering Radius of the Leech Lattice 501

D16
[fr] [<D2.01.34]

(a)
)--m_J~:OI241
[m] [<D4.01.23]

(b) Ea [<D1.04.23]

[<D1.02.34] [<D4.02.13] [<D2.04.13] [<D3.04.12]

[02] [13] [04] [12]
Figure 23.27. The D16ES hole. (a) D16 (from the D.-tree), (b) Es.

There is no need to give diagrams for the holes of types A1 D~, A~ D 6,

AllD7E6 or A\sD9 since we have already seen these holes from a different
point of view in §3. The D24 diagram is the main trunk of the D.-tree in
Fig. 23.20. The centers of these holes are:
2 2 2 2 2 2
4 2 2 2 2 2
Dj: (I/3)
4 4 2 2 2 2
4 2 2 2 2 2

5 3 5 3 3 3
9 3 5 3 3 3
9 3 3 5 5 5
9 3 3 5 5 5

3 3 3 3 3 3
7 3 3 3 3 3
Dt: (l/5)
7 3 3 3 5 5
7 3 3 3 5 5

4 6 4 4 4 4
10 4 4 4 4 4
10 4 4 4 6 6
10 8 4 4 6 6
502 Chapter 23

5 7 5 5 5 5
l37 5 5 5 5
13 5 5 5 9 7
13 9 5 5 9 7

6 8 6 6 8 6
16 8 6 6 6 6
16 6 6 6 1010
16 12 6 6 10 8

8 12 108 10 8
22 10 8 8 8 8
22 8 8 8 14 12
22 16 8 8 14 12

12 18 14 12 16 12
34 16 12 12 12 14
D24 : (1/23)
34 12 12 12 20 20
34 24 12 12 22 18

Theorem 19. There is a unique A IS diagram and a unique hole of type

A IS D 9•
Proof. D9 is unique (Theorem 17) and is disjoined from A IS
(Theorem 18), so by Theorem 7 the hole is unique. From Theorem 16
each A IS is disjoined from a D 9 , so A IS is unique.

5. Holes whose diagram contains an En subgraph

Theorem 20. There are two possible E 8 diagrams in A.
Proof. From the An-tree the backbone of an E8 diagram can be assumed
to be vo, VI,.' . ,Vs, [124], [356]. The remaining node is an [ijk] which
since it is joined to Vs must be a line. The subgroup of Coo fixing vo•...•
[356] is an S3 generated by {3 and 'Y (see Fig. 23.8), and has three orbits on
lines, with representatives [045], [346] and [124]. Since [124] has already
appeared in the diagram only two possibilities remain.
The next theorem, whose proof follows the standard pattern, shows that
both possibilities occur and are distinct.
Theorem 21.
E8 .. Ei (Fig. 23.28) or DI6E8 .

We have already encountered the latter hole in §4. The center of the Ei
hole is
The Covering Radius of the Leech Lattice 503

45 6 -15 6 15 6
1 2 8 7 10 7
c = (1/15)
4 2 8 10 10 10
1 -1 8 10 10 10

Similar arguments lead to Theorems 22 and 24.

(a)

(b)

[134] [256][034] [126] [345] [016] [245][136]

[146]

[246] [135][046] [123] [456] [023][145] [236]

Figure 23.28. The El hole.

(al [124] (b) [123 ]

[456]

[134 ]
[256]

Figure 23.29. The Et hole.

504 Chapter 23

Theorem 22. There are two possible E 7 diagrams in A. and

E7 .. DIOEr or A17E7 .

Theorem 23. There is a unique A 17 diagram and a unique hole of type

A17 E 7'

Proof Every A 17 is disjoined from an E 7 (Theorem 16) and every D 10 is

disjoined from two E7'S (Theorem 18). From Theorem 22 this accounts
for all E/s, so there is a unique diagram of type A 17 .
Theorem 24. There are two possible E6 diagrams in A. and
E6 .. E: (Fig. 23.29) or A ll D7 E 6 .

The vector Z3 and the center of the E: hole are

2 2 o2 2 2
o0 2 0 o0
o0 2 0 o0
o0 2 0 o0

18 3 -6 4 4 4
1 0 5 3 3 3
c- (1/6)
1 0 5 3 3 3
1 0 5 3 3 3

Now only the A24 holes are left.

Theorem 25. There is a unique hole of type A 24.
Proof After Theorem 14, it is known that A contains at most five
inequivalent labeled A24 diagrams. We check in the following way that
these diagrams all arise from the same type of hole. Let fo == '0,
f l - 'I • . . . • f21 - [146], f22 == [023], f23 = [145], f24 == ZI be the vertices
of the fourth A24 diagram in the An-tree in Fig. 23.9, and define
1 24. .
g, - 2s .~ 1(25-,) (f,+c-c), r = 0, 1•...• 24 ,
.-0

where the subscripts are to be read modulo 25 throughout this proof, and
75 7 -25 9 25 11
3 1 13 17 13 17
c = (1/25)
5 1 13 15 23 13
5 3 11 15 15 15

is the center of the fi'S. We then verify that g, E A-c for r = 0, 5, 10, 15,
20, but not for the other values of r. (In fact A is spanned by the vectors
(fi-fo, gs,-fo}. This is one of the 23 "holy constructions" of the Leech
The Covering Radius of the Leech Lattice 505

lattice - see the next chapter.) Therefore no element of Coo can send fo
to fl' f1 to f l +1• . . . • f24 to f 24+1 for t - 1,2, 3, or 4, and so there are five
labeled A24 diagrams which are inequivalent under Coo. This accounts for
all the diagrams, and completes the proof.
We have found a unique hole for everyone of the Niemeier lattices
(see Table 23.1), proving Theorems 2 and 4. Each of these holes has
radius J2, so the proof of Theorem 1 is also complete.
 24
Twenty-Three Constructions
for the Leech Lattice
J. H. Conway and N. 1. A. Sloane

In the previous chapter we classified the points at maximal distance from

the Leech lattice (the "deep holes" in that lattice), and showed that there
are 23 classes of such holes, the classes being in one-to-one correspondence
with the 23 Niemeier lattices in 24 dimensions. We now present 23
constructions for the Leech lattice, one for each class of hole or Niemeier
lattice. Two of these are the usual constructions of the Leech lattice from
the Golay codes over F2 and F 3 .

1. The "holy constructions"

For each of the 23 Niemeier lattices or classes of deep hole there is a "holy
construction" of the Leech lattice. In each case we shall define a set of
fundamental vectors V) and a set of glue vectors (gw)' It then turns out
that the Niemeier lattice is the set of all integer combinations

while the set of all integer combinations

~ mJi + ~ nwgw with ~ mi + ~ nw = 0

is a copy of the Leech lattice.

We begin by defining certain fundamental vectors fi and glue vectors
gx for each root lattice, as shown in Fig. 24.1. The fi form an extended
fundamental set of roots (or basis) for the lattice (compare Figs. 4.5-4.9 of
Chap. 4), and are described abstractly by specifying that their inner
products should satisfy
Vi, f) = 2,
Vi, fj) = - (the number of lines joining fi to fj in
the Coxeter-Dynkin diagram), for i ~ j.
 An' h = n + I On' h=2n-2

h~,(+ (n -I), -en - 2), ... , -1,0) g,= h-' (0,1 ,2, ... , n f:2'1-(~ >-3
(go=h-' (-~n, -~n +. 1, .. " ~n -- I, ~n) (= ~
It>
;l
' - - - - + 010... 0- - + 00... ...00 - + «'
.:,
;l"
+ 0...
o - I, 000 -14+ 0... ~n_'
. .. 0 - + 0 ....
It>
~--- It>
00 - + 0 ... (')
[3 o
;l
-+0 ... 0-+0 ... 00-+0 ..... 0--+0 .. 00 -- + ++oo... ...OO--~ ~
....
(
~
I, I, 13 In _ I In C h ~ .. ) ;:?,
'"
g,=h-I(1n,-1n,-1n+I, . . ,-\:n-I). g3=h- I (-(n -I), -en -2), ... , -1,0) go=h-'(0,1,2, ... ,n -2, +(n -I» o·
;l

E7 , h = 18
'"
E., h = 12 0'
....
H++++----) S-
It>
go = h-I (If, -1. -~. -t LH, -¥)
e -000000 +
?b'
It>
+ 000000 00 - + 0000 0000 - + 00 000000 - + ) (')
;l"
C- o- + 00000 000 - + 000 00000 - + 0 t""
gl =11- 1 ('It, -Ij, -Ii, -lj, -i, -i, L~) go=h-' (-t -a, i, to lj, Ii, 11, -4(,) ~
;:;.
It>
!(++++----)

00 - + 0000 0000 - + 00 Eg , h = 30 j(++++----)

co - + 00000 000 - + 000 00000 - + 0 \
1(+-------) 0-+00000 000-+000 00000-+0
g,=h-'H,If, ~,-L-1.-!'H) ; ,,)
( _ -I - + 000000 00 - + 0000 0000 - + 00 000000 - +

g,= h- I H, -1,-\:,i, L t, -If, t) go-h (5,6,7,8,9,10,11,12)

Figure 24,1. Fundamental vectors ii. glue vectors gx. and Coxeter
(J1
number h for the root lattices An (n ~ 1). Dn (n ~ 4). E 6. E 7 and E 8 o
(plus and minus signs stand for + 1 and -I respectively), "
508 Chapter 24

Each glue vector gx points to a certain Ii, and satisfies

(gx'/i) = h1 - 1 for this Ii ,

(gx,/j) = h1 for the other Ij ,

where h is the Coxeter number (§2 of Chap. 4). The Ii and gx for the
components are then combined to form the fundamental and glue vectors
for the corresponding holy construction, in a way that we now illustrate.
Example 1. For the case A [2 the fundamental vectors are
V) Id,/t.···. it2,fg,ff . .... If2

(fundamental sets of roots for two A\2 lattices in orthogonal 12-spaces A

and B) while the glue vectors are
(gw) gd + gg, gt + gg, g1 + gfo, ... , gt2 + gf .
The subscripts
{(O,O), (1,5), (2,10), (3,2), (4,7), (5,12), (6,4),
(7,9), (8,1), (9,6), 00,11), (11,3),02,8)}

are a subgroup of C \3 x C \3 (the direct product of two cyclic groups) and

form the glue code. There is a glue vector gw = g1 + gff for each word
w = xy in the glue code.

We now assert that the Niemeier lattice of type A [2 consists of all

integer combinations of the fundamental and glue vectors for which the
sum of the coefficients of the glue vectors is zero, while the combinations
for which the sum of all coefficients is zero is a copy of the Leech lattice.
This construction is described by the hole diagram in Fig. 24.2a (defined
below).
Example 2. For the case AllD7E6 there are components of three distinct
types, and the fundamental vectors are
Vi) Id,ft ..... It2,fg,ff ..... If,f~,ff ..... If ,
the superscripts picking out fundamental sets of roots for All, D 7, E 6 in
mutually orthogonal spaces A, B, C. The glue vectors in this case are
(gw) gd +gg +g~, gt +gf +gf, g1 +d +gf, gt +gf +g~, ... ,

corresponding to the words w = (x,y, z) of the glue code

{(O,O,O), (1,1,1), (2,2,2), (3,3,0), (4,0,1), (5,1,2),
(6,2,0), (7,3,1), (8,0,2), (9,1,0), 00,2,1), (1l,3,2)}.
Again the glue code is a group, addition being modulo 12, 4, 3 in
coordinates x,y,z respectively. The hole diagram for this construction is
shown in Fig. 24.2b.
Twenty-Three Constructions for the Leech Lattice 509

3
9

(b)
o
II

KI. I . I

9
~J 2
3

7
r · ~g6.2.0 ~'I'l
l ,
.

1 5
6
Figure 24.2. Hole diagrams for Af2 and A 11 D 7E 6 •

The other constructions are entirely similar. In each case our

fundamental vectors fi are the union of extended fundamental sets of roots
ft, ff, ... for the component root lattices, in orthogonal spaces, and our
glue vectors are the vectors
g., = g1 + gff + g; + ...
for which the word w - xyz ... lies in a certain glue code. In all cases the
glue code is abstractly an additive group whose generators are given in
Table 16.1. The symbols of the glue code corresponding to an An
component are read modulo n + 1, those for D 2n + h E 6 , E7 or Eg
510 Chapter 24

components are read modulo 4,3,2 or I respectively, while those for a D zn

component belong to the four-group with elements {0,1 ,2,3} satisfying 1+ 1
= 2+2 = 3+3 = 0, 1+2 = 3, 2+3 = 1, HI = 2.
As to the proofs that these constructions work, we remark first that our
construction (I) of the typical Niemeier lattice is identical to that given in
Chap. 16. The fact that (2) always gives the Leech lattice still quite
astonishes us, and we have only been able to give a case-by-case
verification, as follows. For each of the 23 Niemeier lattices, the previous
chapter gives an explicit set of Leech lattice vectors having the same
mutual distances as the fi' The glue vectors gx for each component are
easily found. The vector hgo is the Weyl vector (p in Bourbaki's notation
[Bou 1]), and is half the sum of the positive roots. We were able to verify
computationally that, with a suitable labeling, the vectors corresponding to
our gw were indeed Leech lattice vectors, establishing the desired result.
We would like to see a more uniform proof. (J)
There is a "holy construction" analogous to (2) associated with any
even unimodular lattice (I) that can be obtained by gluing root lattices,
but it will only give an even unimodular lattice when the component root
lattices have the same Coxeter number h, and 24h divides (h + l)(n - 24).
It is remarkable that the Niemeier lattices are precisely the 24-dimensional
combinations permitted by this rule (see Chap. 18).
The Coxeter number h has the following property. For each root
lattice there are positive integers Ci with 'f;Cdi = 0 (see Fig. 23.1). In
Fig. 24.1 each gx points to an fi with Ci = 1 (a special node - see §2 of
Chap. 4). The sum of all the Ci for a component is the Coxeter number h.
Furthermore for any integer k the set of vectors 'f;ndi with 'f;ni = k is
obviously the same as the set of vectors 'f;ndi with 'f;ni == k (modulo h).
This remark shows that the equalities in (I) and (2) may be replaced by
congruences modulo h, and hence the intersection of the Niemeier lattice
with the Leech lattice defined by (I) and (2) has index h in each of them.

2. The environs of a deep hole

The results given here and in the previous chapter enable us to form a
clear idea of the environs of any deep hole in the Leech lattice. If we take
the origin of coordinates at the hole, the Leech lattice vectors become
~mdi + ~nwgw with ~mi + ~nw = 1

(or equivalently with 'f;mi + 'f;nw == 1 modulo h), and the nearest lattice
points to the hole are just the fi' The radius of the hole is
Vi, fi) I/Z =..fi. The gw are lattice points only slightly more distant,
having (gw, gw) = 2 + 2h- l .
It is easy to check from our description that fi is distant .J4 or .J6
from fi according to whether Vi, f) = 0 or -1, i.e. according to whether

())R. E. Borcherds has recently found such a proof [Bor2l.

Twenty-Three Constructions for the Leech Lattice 511

(Q) A , . h=2 (b) A:'

gooo ... g 010.

g, = \ (1. - I) - - '
o
I ,' ( - .+)

gill ..

Figure 24.3. (a) Fundamental and glue vectors for A I . (b) Hole diagram
for A r4.
There is a glue vector gxyz '" for each codeword in the binary
Golay code.

Ii and hare unjoined or singly joined in the diagram. Again gxyz ... is
distant .J6 from 11./:./;, ... and distant .J4 from all other Ii' We
therefore indicate distances of .J6 between Ii and /j or between Ii and gw
by joins in the hole diagram. which has a node for each Ii and gw ' Some
examples appear in Fig. 24.2. (Distances between one gw and another are
not indicated in the diagram.)
The case A [4 is slightly exceptional. The diagram for A I is given in
Fig. 24.3a, where the pair of joins between 10 and I I indicates that their
inner product (fo, II) is -2, showing that II - -/0, and the distance
between 10 and I I is.J8. For the holy construction of type Af4 we change
to a different coordinate system and rescale, obtaining
I~ =- (-4,0 23 ) , Ig ~ (0 , -4, 022 ), ... ,

It - (4,0 23 ) , If = (0,4,0 22 ) , ... ,

and there is a glue vector

gxyz ... = «- Ox, (-1)1 , (-1)', ... )
for each word xyz ... in the binary Golay [24,12] code C€24 (§2.8.2 of
Chap. 3). After differencing we obtain Leech's original construction for
his lattice (see §4.4 of Chap. 5) . The hole diagram is sketched in
Fig. 24.3b.
For AF it is best to use 12 complex coordinates, and to take
It - (-3w k , 011), If - (0, -3w k , 010), .. .

for k = 0, 1, 2, where w - e 21fi / 3• There is a glue vector

for each word xyx ... in the ternary Golay [12,6] code C€12 (§2.8.5 of
Chap. 3). This is the complex version of the Leech lattice (Example 12 of
Chap. 7, §3.6 of Chap. 10). It seems likely that some of our other
512 Chapter 24

constructions correspond to the various quaternionic Leech lattices [Tit6],

[Tit7].

The symmetry group of a deep hole. For each entry in Table 16.1, the
group of automorphisms of the Leech lattice that fix the corresponding
deep hole is a compound of groups Goo, Gj, G 2, where Goo is a group
isomorphic to the glue code (or group). A typical element 11"/ E Goo takes
gw to gw+/. The orders of Goo, G 1 and G 2 are shown in the table.
 25
The Cellular Structure of the Leech Lattice
R. E. Borcherds, J. H. Conway and L. Queen

We complete the classification of the holes in the Leech lattice, and of the
associated Delaunay cells, by showing that there are precisely 284 types of
shallow hole.

1. Introduction
In Chap. 23 it was shown that there are 23 types of deep hole in the Leech
lattice A24 , and that these holes are in one-to-one correspondence with the
Niemeier lattices. The existence of this correspondence, and the recently
discovered correspondence between the conjugacy classes in the Monster
group and certain modular functions [Con17], suggested that it might be
worth enumerating the shallow holes in the Leech lattice and completing
the classification of its Delaunay cells, in case any "deep structure" became
apparent. Although this has not yet happened, the complete list of deep
and shallow holes has already found several uses, and it seems worth while
to put it on record. The main result is the following.
Theorem 1. There are 307 types of hole in the Leech lattice, consisting of
23 types of deep hole and 284 types of shallow hole. They are listed in
Table 25.l.

2. Names for the holes

We use the notation of Chap. 23, and describe sets of Leech lattice points
by graphs, with a node for each lattice point, and where two nodes x and y
are
not joined if N (x - y) - 4 ,
joined by a single edge if N (x - y) - 6, or
joined by two edges if N (x - y) - 8 .

Larger numbers of joins will not arise here.

514 Chapter 25

It was shown in Chap. 23 that the vertices of a deep hole in A24 are
described by a graph that is a disjoint union of extended Coxeter-Dynkin
diagrams which have total subscript (or dimension) 24 and constant
Coxeter number h. There are just 23 possible combinations, which can be
seen in the first 23 lines of Table 25.1. Using the same graphical notation
we prove the following result.
Lemma 2. The vertices of a shallow hole in the Leech lattice are sets of
25 points of A24 for which the corresponding graph is a union of ordinary
Coxeter-Dynkin diagrams.
Proof Theorems 5 and 6 of Chap. 23 show that the graph is a union of
ordinary Coxeter-Dynkin diagrams, and by dimension considerations the
hole must contain at least 25 vertices. But the fundamental roots
corresponding to such a diagram (whether or not it is connected) are
linearly independent in the vector space they lie in, and so are affinely
independent. This shows that the graph cannot contain more than 25
nodes. It is just as easy to verify that any such set of 25 points is the
vertex set of a shallow hole. The argument is similar to Theorem 7 of
Chap. 23, which is the corresponding result for deep holes. Thus all the
shallow holes in the Leech lattice are simplices.
3. The volume formula
Let P"P 2, ""PN be a system of representatives for all the holes in A24
under the full automorphism group Coco of A24 . Let voHPi ) denote the
volume of Pi and g (Pi) the order of its automorphism group (i.e. the
subgroup of Co co fixing Pi)' Then we have the volume formula:
1 = volume of a fundamental domain of A24
N
... ~ voHPi ) x no. of images of Pi under Coo
i-I

N ICool
- ~ ----u>T vol (Pi) ,
i-I g I

where I Cool denotes the order of Coo.

------------------------------------------~~
Table 25.1. A list of all 307 holes in the Leech lattice. The first 23
entries are the deep holes. The entries give the name of a hole P j , the
order g(Pj ) of its automorphism group, its scaled volume
svol(Pj) - voJ(pj )· 24!/ICo oI ,

the norm s (Pj ) of its Weyl vector, and the determinant d (p j ) of the
Cartan matrix. The volume formula then becomes
~svoJ(Pj)/g(pl) - 24!/ICo ol-74613.
I

The name of a hole indicates the orbits of its automorphism group on the
components of the diagram. Thus the hole aia/a3ar has two components
of type Q7 that are equivalent under the automorphism group, also two
equivalent components of type ah and three components of type a3, only
two of which are equivalent.
 Table 25.la Table 25.1 b ...,
~
rt>
Name g svol d Name g svol d ("l
~
D,. 2 92 4 d'5 140 9800/2 4 C-
A,. 10 125 25 a'5 195 2925/2 26 ..,~
A17E 7 12 1944 36 d,.a, 1 186 864912 8 ~
..,
D'6E , 2 1800 4 a24 Q l 2 255 260112 50 C
16 2048 64 <"l
A'5 D 9 Q23 Q 2 2 288 230412 72
Dr, 8 1936 16 d"a,a, 282 662712 24 ..,2"
rt>
Ar, 52 2197 169 d 21 a. 240 576012 20 0
24 20736 144 ....,
A"D7E 6 Q21Q)Q, 396 178212 176
D,oE? 8 23328 16 d20d 5 200 500012 16 S-
rt>
A§D6 80 20000 400 Q20Q5 315 157512 126 t""
EJ 6 27000 1 rt>
rt>
<"l
DJ 48 21952 64 d'g e 6 162 437412 12 ~
AJ 324 19683 729 a'9 d 6 1 240 144012 80 t""
A?Dl 256 131072 1024 Q19 Q 4Q r 2 520 135212 400 ~
Et 432 186624 81 d 18e7 126 396912 8 r:;'
rt>
Dt 384 160000 256 GISe7 171 1539/2 38
At 1176 117649 2401 al8 Q 6 Q 1 399 119712 266
A5D• 3456 559872 5184 d 17e8 1 92 423212 4
D: 138240 2985984 4096 GI7 e S 2 141 220912 18
A: 30000 1953125 15625 Q17G 8 2 297 108912 162
A~ 688128 16777216 48 °17 Q 8 2 297 108912 162
Al' 138568320 387420489 312
Af' 1002795171840 68719476736 224 Q17 7Q I
e 2 222 1369/2 72
a17d 7a , 1 288 115212 144
a 17d7a, 2 288 115212 144
a'7d 6a , 2 342 108312 216
Q17 Q 6 Q 2 2 441 1029/2 378
QI7 Q sQ) 2 468 101412 432
°17a4Q)Q, 2 600 100012 720
d'6 d 9 152 2888/2 16
(Jt
d'6 a 9 230 264512 40 .....
(Jt
d'6 e 8a , 122 372112 8
 Table 25.lc Table 25.1d Table 25.le 01
...
0\
Name g svol s d Name g svol s d Name g svol s d

d'6Q gQ , 306 260112 72 d n dl2 136 231212 16 Q 12d 7e 6 I 234 70212 156
d'6e 7Q 2 186 288312 24 d n QI2 208 166412 52 d "ege6 I 114 2166/2 12
d '6e6Q3 252 2646/2 48 dnQ"Q, 276 158712 96 d"e? 2 112 156812 16
d'6 Q 6Q 2Q , 462 2541/2 168 dI3Q<J02Q, 420 147012 240 Q"d IOQ 3Q , 408 86712 384
d'6 d SQ 4 320 256012 80 d13e gQ4 160 256012 20 Q"dlf)IJ~ 2 432 864/2 432
d '6Q SQ 4 390 253512 120 d13Q gQ 4 360 144012 180 Q"d9Q s 2 324 72912 288
Q'~9 204 122412 68 d l3e7QS 204 173412 48 Q"Q~4Q, 480 480/2 960
QI6 e SQ I I 187 2057/2 34 d13Q7d s 304 144412 128 QIl e Se6 174 168212 36
Q,sd lO 2 200 1250/2 64 d ne6Q6 252 151212 84 Q"egdsQ, I 276 158712 96
Q,sd9Q, 2 264 108912 128 an D I2 273 81912 182 Qllesala? 2 576 1536/2 432

al5e8a~ 2 248 192212 64 Q 13d,002 294 102912 168 Q"dge 6 2 228 72212 144
a lSe7Q3 2 264 108912 128 Q13Q9Q 3 420 63012 560 Q"e7d 7 I 204 86712 96
Q,sd7Q 2Q , 2 408 86712 384 Qne sD4 245 171512 70 Q"e7Q SQ r 2 456 72212 576
Q'Se6d 4 2 288 86412 192 a neSo2a)0 I 378 1701/2 168 Qlld7e 6Q , 2 300 62512 288
Q,sd6Q4 2 360 81012 320 Q13 e 7 Q 4 Q I 350 875/2 280 Qlld7Q6Q , I 420 52512 672
Q,sdsd s 2 320 80012 256 Q13d7Q s 336 67212 336 Qlld7Q sQ2 2 468 50712 864
d'4d ,oO, 188 220912 32 °13 Q 7Q 4Q I 560 560/2 1120 Qlld7Q3Q~ 2 648 486/2 1728
d'4 Q ,oO, 286 185912 88 Ql3e 6d SQ , 336 67212 336 Q"d7Q3Q~ 2 648 486/2 1728
d'4Q9Q ,Q, 380 180512 160 a )3Q606 I 441 56712 686 QII Q 7d SQ ,Q, I 576 43212 1536
d'4e gQ2Q , 186 2883/2 24 dr2Q , 2 180 202512 32 Q"e6e 6Q r 2 360 600/2 432

d'4e 7Q 3Q , 256 204812 64 d 12 d IOQ2Q , 276 158712 96 Qll e 6d SQ 3 2 384 51212 576
d '4 Q 7Q 2Q ,Q, 576 172812 384 d 12d 9Q4 240 144012 80 Qll e 6d SQ2Q , 2 468 507/2 864
d'4e 6Q 4Q , 330 181512 120 d '2e gd s 136 231212 16 Qll e 6d 4Q4 2 420 49012 720
d'4 Q 6Q 4Q , 490 171512 280 dl2d gd s 208 135212 64 Q"d6Q sQ3 2 504 44112 1152
d'4d sQ sQ, I 408 173412 192 d 12 e 7d 6 I 156 1521/2 32 QIIQ6d4Q~ 2 756 37812 3024
(j
a 14D902 2 405 72912 450 d '2d7e6 I 180 135012 48 Qlld sQSQ3Q , 2 672 39212 2304 :r
QI4 e SQ2QI 285 180512 90 ar2Q) 4 351 729/2 338 QlldsQ4Q~Q, 2 900 37512 4320 Dl
'tl
Q'4d 7Q 2Q 2 450 75012 540 Q'2e gd s 208 166412 52 Q"QsQ Sd 4 2 576 384/2 1728 (1)
.....,
QI4 Q 6Q 302 I 630 63012 1260 Q12 e SQ4QI 325 162512 130 dro03 Q r 2 384 115212 256 ~
QI4 a j a 2Q I 2 825 60512 2250 QI2 e 7Q 6 273 81912 182 d,oO,oOs 330 82512 264 01
 Table 25.1f Table 25.lg Table 25.lh >-l
::r
('!)

svol d
Name g svol d Name g svol d ('J
Name g
~
101412 192 d 9al 2 324 64812 324 egeja 3 2 128 204812 16 C-
d iOd 9a 5a , 312
d 9a?ar 4 544 57812 1024 ege7a 7G 2a l 288 172812 96 1ii"
...,
d iOa9d6 260 84512 160
a9a ge7 2 270 72912 200 ege7e 6Q 4 165 1815/2 30 u;
d iOa9aJa2a, 600 75012 960 ....
220912 8 a~d, 4 320 51212 400 ege7a 6a 4 245 1715/2 70 ...,
d iOeSe, 94 ~
a~d6a, 4 420 44112 800 48 ()
diOd Sd 6a , 248 96112 128 ese,d 5a 5 204 173412
72 a§d 5ar 4 560 392/2 1600 ega7e 6a 4 I 300 150012 120 2'
...,
diOase, 198 108912 ('!)
a~d.a2a, 4 660 36312 2400 esa,dl; 2 304 144412 128
d iOa saSa2 I 486 72912 648 0
....,
a~a.aJ 4 600 36012 2000 egelas 2 207 158712 54
diOefa, 2 148 136912 32 go
a§a4a J 2 600 36012 2000 eSe6a 6d S I 252 151212 84
d iOe,d,a, 192 115212 64 ('!)
age j 2 115 264512 10 dla, 6 232 84112 128
d iOe,d6a2 228 1083/2 96 t"'
('!)
('!)
190 1805/2 40 ()
agege 7a l dle,a la I 2 248 96112 128
d iOe,a6a2 294 102912 168 ::r
1014/2 192 agegd Sa 2a l 420 147012 240 dle6a J 2 264 726/2 192
diOe,asaJ 312 t"'
agega l 2 425 144512 250 2 360 67512 384 III
d iOe,a.a Ja I 400 100012 320 dld 6a 2a I ::+
a9d Se ,a, I 260 84512 160 dldsd. 2 288 64812 256 ri'
d iOa,d6a 2 384 768/2 384 ('!)
a9a j 2 405 405/2 810 dldsa. 2 320 64012 320
diOa,a}ar 2 832 67612 2048
age ,d,a2 I 300 75012 240 d sefa2a I 2 228 108312 96
dH#taJ 2 320 80012 256
age 7a 6 a ) I 420 630/2 560 d se,d6aJ a i 320 80012 256
d iOd 6a sa. 420 73512 480
age 7aia) 2 550 60512 1000 d Se,a,a2a l 384 76812 384
d H#6a SaJa I 528 72612 768
a9a ,d saJ a i I 640 32012 2560 dse,asa.a I 420 73512 480
d H,u6a Sa Ja l I 672 67212 1344
a9 d 5a a 2a l
i 2 900 27012 6000 d ge 7a Sa )a,al 528 72612 768
diOa~ 2 540 67512 864
a9a 1a1 4 875 24512 6250 d Sa,e6a • 360 54012 480
a iOa9d6 330 49512 440
22 eiat 6 61 372112 2 dsa,dl 2 352 484/2 512
a iOege7 143 185912
ejagQl 2 153 260112 18 d Se 6a l;al 2 468 50712 864
alOege6o) 231 161712 66
eje7Q 2 2 93 2883/2 6 dsdld.a, 2 368 52912 512
a iOega4a2a 1 495 148512 330
ele6a 3 2 126 264612 12 dsdtaJar 2 512 51212 1024
a iOe7a7al 352 704/2 352
eja 6a 2Q l 2 231 254112 42 dsd6dsa5a, 432 48612 768
a iOd,a6a2 I 462 46212 924
eldsa. 2 160 256012 20 aJal 12 513 36112 1458
d§a, 2 240 90012 128
ejasa4 2 195 2535/2 30 asa,dl; 2 432 32412 1152
d 9a 9a 6a l 420 63012 560
egage6 Q 2Q I 351 152112 162 (J1
d gel 2 76 288812 4 agela?a 1 2 567 44112 1458
d 9dl 2 176 96812 64 esasdsa. 360 144012 180 aSe6d Sa.a2 540 36012 1620 ...,....
 Table 25.li Table 25.lj Table 25.lk

Name g svol d Name g svol d Name g svol s d

ejd4 6 140 122512 32 Q7 ag 3 588 25212 2744 d3a, 2160 832 169/2 8192
ejd,a4 2 200 100012 80 G7 a ?ar 6 1152 19212 13824 di a 2a r 360 1344 14712 24576
e?e6 d S 2 156 101412 48 Q7 af 24 1536 14412 32768 dta 3a f 144 2048 128/2 65536
dd6d s 2 176 968/2 64 e~al 48 225 62512 162 dtai 432 2816 12112 131072
e7 d 7 Q 7Q 3Q I 416 676/2 512 ega6 Q , 6 315 52512 378 d 4aj 336 1792 9812 65536
e,d,d 6a, 264 726/2 192 egaSQ2 12 351 50712 486 d 4ar' 120960 14336 49/2 8388608
e,d,a6a 5 I 336 67212 336 eia 3a } 12 486 486/2 972 aSa! 240 1375 12112 31250
e7G a 3Q 1
f 2 544 57812 1024 etala3 8 504 39212 1296 Q4 aj 168 1920 9012 81920
e7Q7 Q 6Q 4Q 1 560 56012 1120 eta sQ4a i 4 675 37512 2430 afo! 2688 2304 8112 131072
e7Q7 Q SQ4Q]Q] 720 540/2 1920 etasai 891 36312 4374 a3a~1 7920 4374 5412 708588

e7 e g 6 153 867/2 54 e6d ,alar 4 756 29412 3888 Q3 Q f2 887040 16384 32/2 16777216
e,dg 3 216 72912 128 e6a gd4 12 612 28912 2592 aFa 1 190080 5103 4912 1062882
e7a g 3 441 56712 686 dta, 24 336 44112 512 a2a [3 10200960 18432 2712 25165824
e7a ?ar 6 936 507/2 3456 dgd,ar 6 448 39212 1024 ar 4a 1 244823040 20480 2512 33554432
d 7afa3Q l 4 608 36112 2048 dgd 4a 2a , 6 528 36312 1536
d,a,a6a 3a 2 672 336/2 2688 dia4a 3 6 480 36012 1280
d7Q7QrQ3Q2 2 960 30012 6144 did 4a r 6 608 36112 2048
d,dg 6 256 512/2 256 dla5d 4a 3a , 672 294/2 3072
d,ai 490 350/2 1372 dgdla3 a f 4 768 28812 4096
d,a¥ 24 1408 242/2 16384 d 6agd 4 6 648 24312 3456

a1a~ 24 1024 256/2 8192 d 6dtar 24 960 22512 8192

a?e6 d 5 2 384 38412 768 ata 1 24 735 22512 4802
ajd 6d, 4 416 33812 1024 a6a gd 4 6 756 189/2 6048
ajd5a, 8 544 28912 2048 d,at 16 720 20012 5184
a?d 5Q40 f 4 800 25012 5120 d,di 120 640 200/2 4096
ajd SQ 3Q 2Qj 4 864 243/2 6144 d,al 20 1000 16012 12500
aja SQ3 Q rQ 1 4 1152 216/2 12288 atd 4a, 48 936 16912 10368
ajaja 3a r 8 1280 20012 16384 ata2 r Q 48 1512 14712 31104
Q7e a SQ1
t 2 432 432/2 864 Gsa; 20 1125 13512 18750
a,dla? 4 576 32412 2048 QsajO 720 3645 75/2 354294
The Cellular Structure of the Leech Lattice 519

The volume of a hole P can be expressed in terms of familiar concepts

in the Lie theory (cf. [Boul]). For a deep hole,

voHP) = 2~! h.Jd ,

and for a shallow hole

vol (p) = 2~! .Jid ,

where h is the Coxeter number, d is the determinant of the Cartan matrix
of the ordinary Coxeter-Dynkin diagram, and s is the norm of the Weyl
vector. For a connected component the values of these quantities are
shown in Table 25.2 (note that s = h(h+l)n/12). For a disconnected
graph, hand d are the products of the values for the components, while s
is their sum. One can also show that the radius of a shallow hole is
(2-1/s) 1/2. .

Table 25.2. The parameters h, d, s for connected diagrams.

an (or An) dn(or Dn) e6(or E 6) e7(or E 7) e8(Or E 8)

h: n+l 2n-2 12 18 30

d: n+l 4 3 2

n (n+ l)(n+2) (n-l)n (2n-l) 399

s: 78 620
12 6 2

4. The enumeration of the shallow holes

We shall only give a brief description of how the shallow holes were
enumerated and Theorem 1 proved. Two methods of classification were
used.
Method 1. This method was used to find all the shallow holes that contain
a particular ordinary Coxeter-Dynkin diagram X as a component, by
finding all occurrences of X as a set of points in the Leech lattice. This
method was used only when X has at least seven points, since for others
the classification becomes too complicated. We drew the graph of all
points in A24 not joined to any point of X, and then deleted enough points
from this graph to make the union of X with the remaining graph be a
collection of ordinary diagrams with a total of 25 points. The group orders
of such holes, found by inspection, are usually 1 or 2.
As an example we consider the diagram X We find from Fig.
= aiS'
23.9 that there are just nine types of ordered diagrams in the Leech
alS
lattice. When we identify reversals these reduce to five distinct types,
520 Chapter 25

Figure 25.1. The enumeration of

shallow holes whose diagram
contains a component al5 (indicated
by heavy lines). For types a-d the
set of points disjoined from a 15
{indicated by shaded nodes and
broken lines} has no lO-point
subgraph of mutually disjoined
ordinary diagrams, so only case e
leads to holes.

which are drawn in Figs. 25.1a-e as heavy lines. (The background of these
figures is taken from Fig. 27.3, and shows all edges between pairs of the 35
points nearest to the center of a deep hole of type A 24 • This is a
convenient portion of the Leech lattice to work with, and in particular
contains all the points of A24 not joined to the a 15 diagrams.)
In four of figures a-e one finds that in the subgraph not joined to the
al5 diagram (indicated by shaded nodes and broken lines) it is impossible
to find a union of disjoint ordinary Coxeter-Dynkin diagrams with a total
of 10 (= 25-15) nodes, and so the corresponding type of a 15 diagram
cannot be part of a shallow hole. However, for the fifth type (figure e), the
disjoined subgraph contains 11 points, and by omitting each point in turn
we can break it into Coxeter-Dynkin diagrams, as shown in the figure .
Three of the eleven possibilities fail (those marked with an x), since they
leave a graph containing an extended diagram. The remaining eight
succeed.
The Cellular Structure of the Leech Lattice 521

Thus there are eight types of shallow hole with a component of type al5
in their diagram: al5egar, al5e7a3, al5e6d4, al5d~, al5d6a4, al5d 7a 2a ),
al5d9a), and al5dlO'
Method 2. We take a given deep hole and find all shallow holes having a
face in common with it. The environs of a deep hole are described in
sufficient detail in the previous chapter to make this quite easy. The group
orders of these holes can be found by considering the subgroup of those
automorphisms of the corresponding deep hole that fix the face in question
and then making due allowance for the fact that the shallow hole may
touch several deep holes in the same manner.
The complete enumeration was carried out by combining the two
methods and using the volume formula as a check. A second check was
supplied by the fact that the quantity sd for each hole must be a perfect
square (since the volume of a hole is a rational number). Most of the
enumeration was performed twice: first by L.Q. and J.H.C., then
independently by R.E.B who completed the list. The group orders were
computed by both teams.
Remarks. (i) It turns out that (in contrast to the deep holes) the diagram
of a shallow hole does not determine it uniquely. There are two holes of
each of the types al7ag, al7d7a), alld7a3af and a§a4a3'
(ii) There is a unique hole of type ar4al' This has the shape of a regular
simplex with 25 vertices, although its group is not transitive on the vertices.
The group is the Mathieu group M 24 acting with orbits of size 24 and 1.
This hole has the smallest radius, namely (2 - 2/25) 1/2, while the shallow
hole of largest radius is d 25 , with radius (2-1/4900)1/2. The deep holes
have radius ..fi.
(iii) If the closed polytopes corresponding to the deep holes of type Dt are
deleted from R24, the resulting space is disconnected.
Acknowledgement. We should like to thank R. A. Parker for several
helpful discussions.
 26
Lorentzian Forms for the Leech Lattice
1. H. Conway and N. 1. A. Sloane

The "holy constructions" of Chapter 24 lead to some very simple

definitions of the Leech lattice using Lorentzian coordinates.

1. The unimodular Lorentzian lattices

We begin by defining the lattices In,l and II n ,l' Let Rn ,l denote the real
(n+ 1) -dimensional vector space of vectors x = (xo,,,,,xn-II xn) with inner
product and norm defined by
X'Y = XoYo + ... + Xn-IYn-1 - XnYn, N(x) = x·x .

XO"",Xn-1 are sometimes called the space-like coordinates and Xn the

time-like coordinate. If x E Rn,l, then xJ.. is the set of Y ERn-I with
X'Y = O.

Just as for Euclidean lattices (i.e. lattices in Rn, see §2.4 of Chap. 2),
a lattice A in Rn,l is called integral if X· Y E Z for all X,Y E A, and
unimodular if it has a basis v(O), ... ,v(n) such that the determinant of the
Gram matrix (v(O . v V» is ± 1. An integral lattice is even, or of type II,
if N (x) E 2Z for all x E A, and otherwise is odd or of type I.
In contrast to the Euclidean case, the classification of integral
unimodular lattices in Rn,1 is easy: there is a unique odd unimodular
lattice in Rn,l for all n, and a unique even unimodular lattice when n == 1
(modulo 8) (see [MiI7], [Neu5], [Serl] and Chap. 15). These lattices will
be denoted by In,l and IIn,1 respectively. On Chap. 15 we used In,l and
lIn I to denote genera of forms, but there is no confusion in using the same
sy~bols for the unique forms in the genera.)
In I can be taken to be the set of vectors x = (xo,,,,,xn-II xn) with all
XiE Z. For n == 1 (modulo 8), IIn,1 can be taken to be the set of x for
which all Xi are all in Z or all in Z + lh and which satisfy
Xo + ... + Xn-I - Xn E 2Z.
Lorentzian Forms for the Leech Lattice 523

As we shall see here and in the following chapters, the lattice 11 25 ,\ is

especially interesting.
Background references on Lorentzian or hyperbolic space are [Cox21a],
[Mi16], [Neu5].
If A is an integral unimodular lattice in Rn,l, and tEA has norm -1,
then t.l.. n A is an integral unimodular n -dimensional Euclidean lattice.
(It is Euclidean by Sylvester's law of inertia (§6.2 of Chap. 15), because
there can not exist two mutually orthogonal vectors x,y in Rn ,\ with
x . x ~ 0 and y . y ~ 0.) For example it has been known for a long time
([Coxl0, p. 419], [Neu3], [Neu5]) that for
t = (I,1,1,1,1,1,1,113),

a vector of norm -1 in 18 ,J, the lattice t.l.. n 18 ,\ is a copy of E 8.

Similarly, let u E A be an isotropic vector, i.e. have norm O. Then
u E A.l.., and (u.l.. n A)/<u> - which we abbreviate to (u.l.. n A)/u -
may be regarded as an integral unimodular (n - 1) -dimensional Euclidean
lattice. (Since (x + AU) . (y + ILU) = X' Y for x,y E u.l.. and A , IL E Z,
the inner product is well-defined on u.l.. /u.) For example if
u = (I,1,1,1,1,1,1,1,113),

an isotropic vector in 19 ,J, then (u.l.. n 19 ,\) /u is again a copy of the E 8

lattice ([CoxlOl, [Neu3], [Neu5]). The 240 minimal vectors of E8 are
represen ted by 9· 8 vectors of shape (I ,- 1, 0 7 10) and 2 (j) vectors of
shape ±Cl 3 , 06 1I).

2. Lorentzian constructions for the Leech lattice

Some years ago J. H. Conway and R. T. Curtis found the following
construction for the Leech lattice A24 . Their proof was long (and
unpublished). Using one of the "holy constructions" of Chap. 24 we can
now give a short proof.
Theorem 1. If t = 0,5,7, ... ,45,47,511145), a vector of norm -1 in 124 ,\,
then t.l.. n 124 ,\ is a copy of the Leech lattice.
Proof From the above remarks we know that t.l.. n 124 ,\ is an integral
unimodular lattice. To show that it is A24 we work instead in the affine
hyperplane
H = {v E R24,\ : v' t = -2} ,

and we shall show that H n 124,\' is a translate of A 24 • More precisely, if

the points of H n 124 ,\ are arbitrarily labeled Po,PJ, ... , we shall show that
{Pi - Po : Pi E H n 124,d is a copy of A 24 •
Straightforward calculation shows that H n 124,\ contains all the points
mentioned in Fig. 26.1, consisting of 25 "fundamental vectors" f 0,"''/ 24
(in the notation of Chap. 24) forming a D24 diagram, plus two "glue
524 Chapter 26

go" (1,2,3"",23,241701 g,"-(1,2,3, ••• ,21,22,22,251 701

(0,1,-1,0 21 1 01 (0 '9 , I , - I , 0 3 101

-(4,5,7,9, '" ,45,47,51 11451

Figure 26.1. Hole diagram of type D24 for Leech lattice.

vectors" go and g\. Two points in the figure are joined by an edge (or a
directed edge) if they are distance ../6, all other pairs being distant .J4
apart.
Since Fig. 26.1 is a copy of the D24 hole diagram (see Chap. 24), it
follows that the set {Pi - po} contains A 24 . But A24 is unimodular, so this
lattice is the Leech lattice.
A similar argument establishes the next result.
Theorem 2. If u = (1,3,5, ... ,45,47,511145), an isotropic vector in 125 ,\.
then (u J. n 125 ) / u is a copy of the Leech lattice.
Proof Prefix all the points of Fig. 26.1 with an initial zero,
However still more elegant coordinates for A24 may be obtained as
follows, It is well known that the only solutions of
02 + 12 + 22 + ... + m 2 = n2

in positive integers are (m,n) = 0,1) and (24,70) ([Ljull, [Mor6,

p, 258], [Wat23]). In particular
w = (0,1,2,3, ... ,23,24170)

is an isotropic vector in 11 25 ,\. Let us define the Leech roots to be the

vectors r E 11 25 ,\ that satisfy
r'r=2, r·w=-l.

(The reason for this name will appear in the next chapter.) Our main
result is the following,
Theorem 3. (a) The Leech roots are in one-to-one correspondence with
the points of the Leech lattice. This correspondence is an isometry for the
metric defined by
d(r,s)2 = N(r -s) .

(b) Furthermore (wJ. n 11 25 ) /w is also a copy of the Leech lattice.

Lorentzian Forms for the Leech Lattice 525

Proof We know from the remarks in §l that (w.L n 11 25 ,1) /w is an

integral unimodular lattice. The hyperplane
H = {s E R25 ,1: s·w = -I} (3)

is a translate of w.L, and for any s E H, (s + AW)' (s + AW) = s·s - 2A.

For s E H n 11 25 ,1> s . s is even and A E Z can be chosen uniquely so that
s + AW is a Leech root. Thus the Leech roots are a set of coset
representatives for (w.L n 11 25 ,1) /w.

The set of Leech roots we have found can be embedded in a set of

Leech roots that contains the A 24 hole diagram (see Fig. 27.3), and so by
Chap. 24 (w.L n 11 25 ,1) /w is a copy of the Leech lattice.
Remarks. Although distances in H are correctly computed using
d (x ,y) = N(x - y) 1/2, the reader should note that norms and inner
products of individual vectors in H are not invariant under the addition of
w. If r ,s are distinct Leech roots then, since N(r -s) ~ 4, we have
r's ~ O.
We may also work with the isotropic representatives for the points of H.
Theorem 4. If u is any isotropic vector of H n 11 25 ,1. then
(u.L n 11 25 ,1) / u is a copy of the Leech lattice.
Proof We see that r = u -w is a Leech root, and that the reflection in r,
Rr : x --+ x - (x' r) r ,

is a symmetry of 11 25 ,1 that sends u to w, and (u.L n 11 25 ,1)/U to

(w.L n 11 25 ,1) /w.
Theorem 5. After identifying the Leech lattice with the set of Leech roots
in the hyperplane H (see (3)). let u denote the isotropic representative of
the center of a deep hole in H. Then (u.L n 11 25 ,1) /u is a copy of the
Niemeier lattice corresponding to that hole.
Proof We already know that (u.L n 11 25 ,1) /u is a 24-dimensional even
unimodular lattice, and so is either A24 or one of the Niemeier lattices.
We take a particular case, and suppose that U is the center of a deep hole
of type All D7E 6. The Leech lattice vectors closest to U are the
fundamental vectors
(f) f~,ft ,... ,ft2,fg,ff, ... ,ff ,f~,ff, ... ,ff ,

the superscripts picking out fundamental sets of roots for All, D 7 , E 6 in

mutually orthogonal spaces A, B, C (see Example 2 of Chap. 24). From
Eq. (1) of Chap. 23 we have
cfc
U = h1 ~
AfA
Cj j = h1 ~
B
Cj
B
fj = h1 ~ Cj j ,
I I I

where h is the Coxeter number of the hole and the Cj are defined in
Fig. 23.1. Since the covering radius of A24 is .J2,
526 Chapter 26

N(f: - U) = 2, N(f:) = 2,

so
f: . U = Y2 N(U) (5)

But from (4),

h N(U) = ~ctft· U = Y2 N(U) ~ct = Y2 h N(U).

Therefore N(U) = 0, so we can take u = U, and from (5) ft E u.L.

Thus all the fi are in u.L n 11 25 ,1> and the lattice is the Niemeier lattice
whose components agree with the components of the hole diagram.
Again, the sum of the vectors on the 25-gon visible in Fig. 27.3 is
( (I /2) 25 15/2), which is isotropic and proportional to the center of the
corresponding hole of type A 24 in the Leech lattice in H. Thus for
u = (1 25 15), we see that (u.L n 11 25 ,!) / u is a Niemeier lattice of type
A 24 •
In a similar way the vectors
U2 = (lI3, 3!211]),
U4= (l!5, 39 , 511]),

have (u/ n I1 25 ,!) lUi equal to the Niemeier lattices of types

A~,A f2,A 17E7,A !5D9 respectively, and other examples can be produced at
will.
The final lemma shows that in these examples it does not matter
whether we work in In,! or lin,!.
Lemma 6. If U = V2 (uo,u I> ... , u n-! Iun) E lin,! is isotropic and all Ui are
odd integers then
U.L n In,! = u.L n lIn,! . (6)

Xo + ... + Xn-! - Xn == 0 (mod 2) ,

and x E lin,!. Therefore, in (6), LHS ~ RHS. But det In,!

= det lin,! (=1), so LHS = RHS.
 27
The Automorphism Group of the
26-Dimensional Lorentzian Lattice
1. H. Conway

The automorphism group of the lattice 11 25 ,\ is shown to be a certain

infinitely generated Coxeter group extended by the negative of the identity
operation and the group of all automorphisms of the Leech lattice.

1. Introduction
The automorphism groups of the unimodular Lorentzian lattices I n,\ and
II n,\ (defined in the previous chapter) are of considerable interest. In the
present chapter we investigate 11 25 ,1; the odd lattices In,1 will be considered
in the succeeding chapter.
The study of these automorphism groups seems to have originated with
Coxeter and Whitrow [Cox30] (see also [Cox28]), who identified the group
of 13,\ as a hyperbolic reflection group. More recently, Vinberg [Vinl1-
[Vin4], [Vin7], Meyer [Mey21, and Vinberg and Kaplinskaja [VinI5] have
shown that the reflection subgroup of AutO n,\) has finite index only if
n ~ 19, and have given Coxeter diagrams for the reflection subgroups in
these cases. In [Vin7] Vinberg has also shown that the reflection
subgroups of AutOI9,\) and AutOI17) have finite index and are described
by the Coxeter diagrams shown in Figs. 27.1 and 27.2.
The vectors with positive time coordinate and negative or zero norm in
a Lorentzian space, taken modulo positive scalar factors, become the
ordinary or ideal points, respectively, of the associated hyperbolic space.
The symmetry group of the hyperbolic space can be identified with the
autochronous symmetries of the Lorentzian space (those symmetries not
interchanging the positive and negative time cones). The full group of
symmetries of the Lorentzian space is the direct product of the
autochronous subgroup with the group of order 2 generated by negation.
528 Chapter 27

If L is a Euclidean or Lorentzian lattice. the reflecting hyperplanes that

correspond to all the roots of L partition the corresponding Euclidean or
hyperbolic space into fundamental regions (§2 of Chap. 4). The roots
corresponding to the walls of anyone fundamental region are a set of
fundamental roots. for which the corresponding reflections generate the
smallest group that contains all reflections in L. The latter group we call
the reflection group of L. Since lIn.! is unimodular. the roots all have
norm 2.
We shall describe reflection groups by Coxeter diagrams using
Vinberg's conventions [Vin7. p. 325]. These diagrams contain a node for
each fundamental root R j • and the reflection group is defined by the
relations

R/ = 1.
(R j Rj )2 = 1. if nodes i and j are unjoined.
(R j Rj )3 - 1. if nodes i and j are joined.

There are no other defining relations. These relations correspond to

reflecting hyperplanes that intersect. Two nodes corresponding to non-
intersecting hyperplanes are joined by a heavy line if the reflecting
hyperplanes are parallel. or by a broken line if they are divergent.

2. The main theorem

Theorem 1. For n - 9.17.25 there are respectively 10.19. 00 fundamental
roots for lIn.!. which can be taken to be the vectors r E II n.1> with
r'r == 2 and r'Wn = -1 •

where the Weyl vectors (I) Wn are given by

W9 =- (0.1 •...• 8138). N(W9) == -1240 •
W17 - (0.1 •...• 16146). N(w17) = -620 •
w25 = (0.1 •...• 24170). N(W25) = 0 .

The group of all autochronous automorphisms of lIn.! is the split

extension of the Coxeter group whose generators are the corresponding
reflections. by
for n = 9. the trivial group.
for n - 17. a group of order 2. or
for n = 25. an infinite group abstractly isomorphic to the
group Co"" of all automorphisms of the Leech lattice
(including translations).
The corresponding Coxeter diagrams are displayed for n ... 9 in Fig. 27.1
and for n = 17 in Fig. 27.2. For n = 25. a portion of the diagram is

WIt would perhaps be more conventional to call -w the Weyl vector

(compare [BouID. but our choice is better in a hyperbolic space.
Automorphism Group of 26-Dimensional Lorentzian Lattice 529

shown in Fig. 27.3. The full diagram has one node for each Leech lattice
vector and (using Vinberg's conventions) two nodes r ,s are joined by

no line if N(r-s) = 4 ,
an ordinary line if N (r - s) = 6 ,
a heavy line if N (r - s) - 8 , or
a broken line if N(r-s) ~ 10

8 9 2 3 4 5 6 7
Figure 27.1. The Coxeter diagram for 119.1, The points are:
i : (oj, +1, -I, 07-; 10), for 0 ~ i ~ 7; 8: ((1,-2)91 Y2); 9: (-1 2,07 10).

17 18

16 0 2 3 4 5 6 7 8 9 10 II 12 13 14 15
Figure 27.2. The Coxeter diagram for 11 17 •1, The points are:
i : (oj, +1, -I, 01S-; 10), for 0 ~ i ~ 15; 16: (-liz, (Y2) 16 1 3/2);
17: (_1 2,0 15 I 0); 18 : (0 14, 13 I I).

Figure 27.3. A convenient set of 35 Leech roots, representing the Leech

lattice points closest to a deep hole of type A 24 . The coordinates of the
points are as follows: i: (OJ,+I,-1,023-i 10) for 0 ~ i ~ 23 ;
24: (-1/2,(1/2)23,3/215/2); 25: (-1 2,0 23 10); 26: (0 7,1 18 14) ;
27: «1/2)12,(3/2) 13 19/2); 28: «1/2)17,(3/2)819/2); 29: (0 22 ,1 31 J);
30: (0 5,1 14,26 16); .31: (0 10,1 14,214); 32: (04,1 11,2 10 17);
33: «(1/2)9, (3/2)11, (5/2)5115/2); 34: (0 14, 111 13).
530 Chapter 27

Proof It is easy to check that for n = 9 and 17 all r that satisfy the
conditions (1) are displayed in Figs. 27.1 and 27.2. Since the diagrams of
these figures agree with those given by Vinberg [Vin7, p. 347], the
assertion of the theorem holds for n = 9 and 17.
When n = 25, Theorem 3 of the previous chapter shows that there is a
one-to-one correspondence between the points of the Leech lattice and the
vectors r that satisfy (1), namely the Leech roots. This correspondence is
an isometry for the metric defined by d(r,s)2 = N(r-s). The assertion of
the theorem will therefore follow if we can show that the Leech roots are
precisely the fundamental roots for 11 25 ,[' since they have the joins and
symmetries indicated in the statement of the theorem.
We show this using the algorithm described by Vinberg for finding the
fundamental roots for any discrete hyperbolic reflection group ([Vin7], §4
of Chap. 28). We take Vinberg's vector Xo to be W25 (which satisfies his
conditions, except of course those entailing that the algorithm terminates
after finitely many vectors are produced), and define the height of a vector
r to be h (r) = - r . W25' The algorithm then proceeds as follows. The
vectors in 11 25 ,1 of norm 1 or 2 (for us, only norm 2) and positive height
are enumerated in increasing order of height· norm- l /2 (for us, in order of
height). Vectors of height zero have norm at least 4, by the result of the
previous chapter, and so need not be considered in our case. (In more
general cases such vectors require special treatment when starting the
algorithm.) A vector is rejected by the algorithm if it has strictly positive
inner product with any previously accepted vector; otherwise it is accepted
as a fundamental root. It is not necessary to test the condition between
vectors of the same height.
It is clear that in our case all vectors of height 1 and norm 2 are
accepted. We shall show that every other norm 2 candidate x is rejected.
Let h (x) = h, and define v = x /h . Then v lies in the affine
hyperplane of height 1 vectors and so by the previous chapter specifies a
point in the rational 24-dimensional space of the Leech lattice. By
Chap. 23 there is a Leech lattice vector r, which we can regard as a Leech
root, such that N(v-r) ~ 2. Thus
2 2 1
x'r>':
---x'r+2~2,
h2 h ~
~ h '

showing that x is rejected.

Remarks. (j) Abstractly, the reflection subgroup of Aut(II 25 I) is the

Coxeter group with a generator Rr for each Leech lattice vector rand
having the presentation

(Rr)2=1,
(RrRs)2 = 1 if N(r-s) = 4,
(RrRY = 1 if N(r-s) = 6
Automorphism Group of 26-Dimensional Lorentzian Lattice 531

(and no other relations). The group of all autochronous automorphisms of

is this Coxeter group extended by the group Co"" of its diagram
11 25 ,1
automorphisms.
(in Several alternative proofs are known, all of which, however, depend on
the main theorem of Chap. 23. It would be desirable to find a more direct
proof.
(iii) We can show that for n = 33,41,... there is no vector Wn having
constant inner product with the fundamental roots for IIn ,I'
(iv) The group Co"" of diagram automorphisms acts transitively on the
fundamental roots! Compare 119,\' where the diagram has no nontrivial
automorphism, and 11 17,\> where the automorphism group has order 2. It is
not known whether the group is transitive for any of 1133,\>.... (See also
lYin8 J-lYin 121.)
(v) Is there any connection with the Fischer-Griess Monster group?
(Compare Chap. 30.)
 28
Leech Roots and Vinberg Groups
J. H. Conway and N. J. A. Sloane

This chapter investigates the remarkable properties of the Leech roots,

which are the fundamental roots for the even unimodular lattice in
Lorentzian space R25 ,1, and correspond one for one with the points of the
Leech lattice. We give an extensive table of the Leech roots in both
Euclidean and hyperbolic coordinates. We also provide the first of what
promise to be many applications by showing that the Leech roots simplify
and explain the remarkable results of Vinberg, Kaplinskaja and Meyer on
the reflection groups of unimodular Lorentzian lattices in dimensions below
20. They also enable us to make some progress on the study of these
groups in the next few dimensions.

1. The Leech roots

The Leech roots are the vectors r in the even unimodular lattice 11 25 ,1 for
which
r . r - 2, r ' W = -1 ,

where w is the Weyl vector

w = (0,1,2, ... ,23,24170) .

Let T 1 be the affine hyperplane

TI = {x E R 25 ,1 : X·W = -I} .

--------------------------------------------I~
Table 28.1. Leech roots '1 in hyperbolic coordinates
,x241 X25). (The 27 columns of the table give the index i and
(Xo, Xl • .•.
then Xo, •.• ,X25, except that the roots with i < 0 are fractional and have
been multiplied by 2. For example r -1 - (_1/2,1/ 2, ... , 112,3/21512). The
table contains all roots with X25 < 25.)
 t"'
(1)
x. x. X, x. x, x. x, x, x, X. x •• Xu Xli X .. X .. Xu Xu X., X .. Xu X •• X" X" X,. X" Xl.
(1)
(')
0 -1 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ::r
1 1 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 U 0 0 0 0 0 ~
2 0 1 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0
3 0 0 1 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ....
4 0 0 0 1 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 '"
III
5 0 0 0 0 1 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 !:l
0-
6 0 0 0 0 0 1 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
7 0 0 0 0 0 0 1 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ~
!:l
8 0 0 0 0 0 0 0 1 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 cr
(1)
9 0 0 0 0 0 0 0 0 1 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ...,
I7Q
10 0 0 0 0 0 0 0 0 0 1 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
11 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
...,
0
12 0 0 0 0 0 0 0 0 0 0 0 1 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 c
'0
13 0 0 0 0 0 0 0 0 0 0 0 0 1 -1 0 0 0 0 0 0 0 0 0 0 0 0
14 0 0 0 0 0 0 0 0 0 0 0 0 0 1 -1 0 0 0 0 0 0 0 0 0 0 0
'"
15 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 -1 0 0 0 0 0 0 0 0 0 0
16 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 -1 0 0 0 0 0 0 0 0 0
17 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 -1 0 0 0 0 0 0 0 0
18 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 -1 0 0 0 0 0 0 0
19 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 -1 0 0 0 0 0 0
20 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 -1 0 0 0 0 0
21 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 -1 0 0 0 0
22 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 -1 0 0 0
23 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 -1 0 0
24 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0
25 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1
26 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 3
27 0 0 0 0 0 0 0 I 1 1 1 1 1 1 1 1 1 1 I 1 I 1 1 4
28 0 0 0 0 0 0 0 0 0 0 1 1 1 I 1 1 I 1 1 I 2 4
29 0 0 0 0 0 1 1 1 1 1 I 1 1 1 2 2 2 2 2 2 6
30 0 0 0 0 0 0 0 0 1 1 1 I 1 2 2 2 2 2 2 2 6
31 0 0 0 0 1 I I 1 1 1 2 2 2 2 2 2 2 2 2 2 7
32 0 0 0 0 0 0 1 1 1 I 2 2 2 2 2 2 2 2 3 7
33 0 0 0 0 0 1 1 2 2 2 2 2 2 2 2 2 2 3 3 8
34 0 0 0 0 1 2 2 2 2 2 2 2 2 2 2 3 3 3 3 9
Ul
35 0 0 0 1 1 1 2 2 2 2 2 2 2 3 3 3 3 3 9 w
w
 Table 28.1 {cont.} U1
W
.j>.
Xo Xl Xz X3 X .. X5 Xe X7 Xs x, X 10 Xu Xlt Xu Xu Xu XIS Xli X l8 Xu X 20 Xu X 22 Xn Xt.a x"
36 0 o o o o o 1 2 2 2 2 2 2 2 2 3 3 3 3 3 9
37 0 o o 1 I 1 2 2 2 2 2 2 2 2 3 3 3 3 3 3 10
38 0 o o o o 1 2 2 2 2 2 2 3 3 3 3 3 3 3 3 10
39 0 o o o 1 1 1 2 2 2 2 2 2 3 3 3 3 3 3 4 10
40 0 o o o 1 1 2 2 2 2 2 2 2 3 3 3 3 3 3 4 4 11
41 0 o o 1 1 1 2 2 2 2 2 3 3 3 3 3 3 3 4 4 11
42 0 o o 1 1 1 2 2 2 2 2 3 3 3 3 3 3 3 3 4 4 4 12
43 0 o o 1 1 2 2 2 2 2 2 2 2 3 3 3 3 3 4 4 4 4 12
44 0 o 1 1 1 2 2 2 2 2 3 3 3 3 3 3 4 4 4 4 12
45 0 o o o o 2 2 2 2 2 2 3 3 3 3 3 3 4 4 4 4 12
46 0 o o o 2 2 2 2 2 2 3 3 3 3 4 4 4 4 4 12
47 0 o 2 2 2 2 2 2 3 3 3 3 3 4 4 4 4 4 4 13

48 0 o o o 2 2 2 2 3 3 3 3 3 3 4 4 4 4 4 4 13
49 0 o o o 2 2 2 2 2 2 3 3 3 3 3 3 4 4 4 4 5 13
50 0 o o 1 1 2 2 2 2 2 2 3 3 3 3 4 4 4 4 4 5 13
51 0 o o 1 2 2 2 2 2 3 3 3 3 4 4 4 4 4 4 4 5 14
52 0 o o 1 2 2 2 2 3 3 3 3 3 3 3 4 4 4 4 5 5 14
53 0 o o 2 2 2 2 2 2 2 3 3 3 3 4 4 4 4 4 5 5 14
54 0 o 1 1 2 2 2 2 3 3 3 3 3 4 4 4 4 4 5 5 14
55 0 o o 2 2 2 2 2 2 3 3 3 3 3 4 4 4 4 4 5 5 5 15
56 0 o 1 1 2 2 2 3 3 3 3 3 3 4 4 4 4 4 5 5 5 15
57 0 o 1 1 2 2 2 2 2 2 3 3 3 4 4 4 4 4 5 5 5 15
58 0 o o o 1 2 2 2 2 3 3 3 3 4 4 4 4 4 4 5 5 5 15
59 0 o 2 2 2 2 2 3 3 3 3 3 3 4 4 4 5 5 5 5 15

60 0 o o o 1 2 2 2 2 2 3 3 3 3 4 4 4 4 5 5 5 5 15
61 0 o o 1 1 1 2 2 2 3 3 3 3 3 4 4 4 4 5 5 5 5 15
62 0 o 1 1 2 2 2 2 3 3 3 3 3 4 4 4 4 5 5 5 5 5 16
63 0 o o 1 2 2 2 3 3 3 3 4 4 4 4 4 5 5 5 5 5 16
64 0 o o 1 2 2 2 2 3 3 3 3 4 4 4 5 5 5 5 5 5 16
65 0 o o 1 2 2 2 3 3 3 4 4 4 4 4 4 5 5 5 6 16
66 0 o o 2 2 2 3 3 3 3 3 3 4 4 4 5 5 .5 5 6 16 (')
67 0 o 1 2 2 2 2 3 3 3 3 4 4 4 4 5 5 5 5 6 16 ::r
68 0 o o 2 2 2 3 3 3 4 4 4 4 4 5 5 5 5 5 6 17 .§
69 0 o 1 1 2 2 3 3 3 3 4 4 4 .5 .5 .5 5 5 6 17 ..,en
70 0 o o 2 2 3 3 3 3 3 4 4 4 4 4 5
" 5 5 6 6 17 N
1 2 2 2 3 3 3 4 4 4 4 4 4 5 5 5 6 6 17 00
71 0 o
 Table 2801 (conto) t"'
(1)
(1)
x. Xl X, X, X, x, X, x, x, X, Xl. Xu Xl3 Xu Xl5 X16 X 17 XIS Xu X 21 X" X" X" (')
X" x,. X"
::r'
72 0 0 1 1 2 2 2 3 3 3 3 3 3 4 4 4 5 5 5 5 6 6 17 :;0
73 0 0 1 1 2 2 2 2 2 2 3 3 3 4 4 4 4 5 5 5 5 6 6 17 0
74 0 0 0 0 1 2 2 2 2 3 3 3 3 4 4 4 4 5 5 5 5 6 6 17 ~
75 0 0 1 1 2 2 2 2 3 3 3 3 4 4 4 5 5 5 5 5 5 6 6 18 '"
III
76 0 0 1 2 2 2 2 3 3 3 4 4 4 4 4 4 5 5 5 6 6 6 18 ::l
0-
77 0 0 1 1 2 2 2 3 3 3 3 3 3 4 4 4 5 5 5 5 6 6 6 18
78 0 0 1 2 2 2 2 2 2 3 3 3 4 4 4 4 5 5 5 5 6 6 6 18 <
50
79 0 1 1 1 1 2 2 2 3 3 3 3 4 4 4 4 5 5 5 5 6 6 6 18 cr'
(1)
80 0 0 0 0 2 2 2 2 3 3 3 3 4 4 4 4 5 5 5 5 6 6 6 18 ..,
(fC<
81 0 0 0 2 2 2 2 3 3 4 4 4 4 4 5 5 5 5 6 6 6 18
82 0 0 0 2 2 2 3 3 3 3 3 4 4 5 5 5 5 5 6 6 6 18 (l
..,
83 0 0 0 2 2 2 3 3 3 3 4 4 4 4 4 5 5 6 6 6 6 18 0
C
"0
84 0 0 0 1 2 2 2 2 2 3 3 3 3 4 4 4 5 5 5 6 6 6 6 18
85 0 0 1 2 2 2 2 3 3 3 3 4 4 4 4 5 5 5 5 6 6 6 6 19
'"
86 0 0 0 1 2 2 2 2 3 3 3 4 4 4 5 5 5 5 5 6 6 6 6 19
87 0 0 0 1 2 2 2 3 3 3 3 3 4 4 4 5 5 5 6 6 6 6 6 19
88 0 0 1 1 1 2 2 2 3 3 3 4 4 4 4 5 5 5 6 6 6 6 6 19
89 0 0 0 1 2 2 2 3 3 3 3 4 4 4 4 4 5 5 5 6 6 6 7 19
90 0 0 0 2 2 2 2 2 3 3 3 3 4 4 4 5 5 5 5 6 6 6 7 19
91 0 0 1 1 1 2 2 3 3 3 3 3 4 4 4 5 5 5 5 6 6 6 7 19
92 0 0 1 1 2 2 2 2 2 3 3 4 4 4 4 5 5 5 5 6 6 6 7 19
93 0 0 1 1 2 2 2 2 3 3 3 3 4 4 4 4 5 5 6 6 6 6 7 19
94 0 0 0 2 2 2 2 3 3 3 4 4 4 4 5 5 5 5 6 6 6 6 7 20
95 0 0 2 2 2 3 3 3 3 4 4 5 5 5 5 5 6 6 6 6 7 20

96 0 0 1 2 2 2 3 3 3 4 4 4 4 4 5 5 6 6 6 6 6 7 20
97 0 0 2 2 2 2 2 3 3 3 4 4 4 5 5 5 6 6 6 6 6 7 20
98 0 0 2 2 2 3 3 3 4 4 4 4 5 5 5 5 5 6 6 7 7 20
99 0 0 2 2 3 3 3 3 3 4 4 4 4 5 5 5 6 6 6 7 7 20
100 0 0 2 2 2 2 2 3 3 4 4 4 4 4 5 5 5 6 6 6 7 7 20
101 0 0 2 2 2 2 3 3 3 3 3 4 4 5 5 5 5 6 6 6 7 7 20
102 0 1 1 1 2 2 2 2 3 3 3 4 4 4 5 5 5 5 6 6 6 7 7 20
103 0 0 0 1 1 2 2 3 3 3 3 4 4 4 5 5 5 5 6 6 6 7 7 20
104 0 0 0 1 2 2 2 2 3 3 3 4 4 4 4 5 5 6 6 6 6 7 7 20
105 0 0 1 2 2 2 2 3 3 4 4 4 4 5 5 5 5 6 6 6 6 7 7 21
U1
106 0 0 1 2 2 2 3 3 3 3 4 4 4 4 5 5 6 6 6 6 6 7 7 21 W
107 0 0 0 2 2 2 3 3 3 4 4 4 5 5 5 6 6 6 6 6 7 7 21 U1
 (Jl
Table 28.1 (cont.) w
x, x. x, x, x. x, x, x,. Xu x .. Xli Xu x,. X18 Xu x., Xu x .. x .. x .. 0-
Xo x. x. x" x" x"
108 0 0 1 1 2 2 2 3 3 3 3 4 4 4 5 5 5 5 5 6 6 7 7 21
109 0 0 1 2 2 2 2 2 3 3 3 4 4 4 4 5 5 5 6 6 6 7 7 7 21
110 0 1 1 1 1 2 2 3 3 3 3 4 4 4 4 5 5 5 6 6 6 7 7 7 21
111 0 0 0 1 2 2 2 3 3 4 4 4 4, 4 5 5 5 6 6 6 7 7 7 21
112 0 1 1 2 2 2 2 3 3 3 3 4 4 5 5 5 5 6 6 6 7 7 7 21
113 0 0 0 1 2 2 3 3 3 3 3 4 4 5 5 5 5 6 6 6 7 7 7 21
114 0 0 0 2 2 2 2 2 3 3 4 4 4 5 5 5 5 6 6 6 7 7 7 21
115 0 0 1 1 2 2 3 3 3 4 4 4 5 5 5 5 6 6 6 7 7 7 21
116 0 0 0 2 2 2 3 3 3 3 4 4 4 5 5 6 6 6 6 7 7 7 21
117 0 0 1 1 2 2 2 2 3 3 4 4 4 4 5 5 6 6 6 6 7 7 21
118 0 0 1 2 2 2 3 3 3 3 4 4 4, 5 5 5 6 6 7 7 21
119 0 2 2 2 3 3 3 4 4 4 4 5 5 5 6 6 6 7 7 7 22
120 0 0 0 2 2 2 3 3 3 3 4 4 5 5 5 5 6 6 6 7 7 22
121 0 0 2 2 2 3 3 4 4 4 5 5 5 5 6 6 6 7 7 7 7 22
122 0 0 1 2 2 3 3 3 3 4 4 4, 5 5 6 6 6 6 7 7 7 7 22
123 0 0 1 2 2 2 2 3 3 3 4 4 4 5 5 5 6 6 7 7 7 7 7 22
124 0 0 0 2 2 2 3 3 3 4 4 4 4 5 5 5 6 6 6 6 7 7 8 22
125 0 0 1 2 2 3 3 3 3 4 4, .5 5 5 5 6 6 6 6 7 7 8 22
126 0 0 2 2 2 2 3 3 3 4 4 4 5 5 6 6 6 6 6 7 7 8 22
127 0 0 1 2 2 3 3 3 4 4 4 4 5 5 5 5 6 6 7 7 7 8 22
128 0 0 2 2 2 2 3 3 3 4 4 5 5 5 5 5 6 6 7 7 7 8 22
129 0 0 2 2 2 2 3 3 4 4 4 4 4 5 5 6 6 6 7 7 7 8 22
130 0 0 2 2 2 3 3 3 3 3 4 4, 5 5 5 6 6 6 7 7 8 22
131 0 2 2 2 3 3 3 4 4 4 5 5 5 6 6 6 7 7 8 22
132 0 0 2 2 2 3 3 3 4 4 4 5 5 6 6 6 6 6 7 7 8 23
133 0 0 2 2 2 3 3 3 4 4 5 5 5 5 5 6 6 7 7 7 8 23
134 0 0 1 2 2 2 3 3 4 4 4 4 4 5 5 6 6 6 7 7 7 7 8 23
135 0 0 1 2 2 3 3 3 3 3 4 4 5 5 5 6 6 6 7 7 7 7 8 23
136 0 0 2 2 2 2 2 3 3 4 4 4 5 5 5 6 6 6 7 7 7 7 8 23
137 0 1 1 1 2 2 3 3 3 4 4 4 5 5 5 6 6 6 7 7 7 7 8 23
138 0 0 1 2 2 3 3 4 4 4 4 5 5 5 5 6 6 6 7 7 8 8 23 (")
139 0 0 1 1 2 2 3 3 3 3 4 4 4 4 5 5 6 6 6 6 7 7 8 8 23 ::r-
III
140 0 0 1 2 2 2 2 3 3 3 3 4 4 5 5 5 6 6 6 6 7 7 8 8 23 "0
141 0 1 1 1 2 2 2 2 3 3 4 4 4 5 5 5 6 6 6 6 7 7 8 8 23 t1>
.....,
142 0 0 0 1 1 2 2 3 3 3 4 4 4 5 5 5 6 6 6 6 7 7 8 8 23 I>.)
143 0 0 1 2 2 2 2 3 3 3 4 4 4 4 5 5 5 6 6 7 7 7 8 8 23 00
 Table 28.1 (cont.) t""'
<1>
<1>
x, Xl X, X, x, X, X, x, x, X, XIO Xu Xu X13 XI< X" Xu X l7 X18 Xu X" X" X" X" X" x" ::r
"
144 0 I I 2 2 3 3 3 3 4 4 5 5 5 5 6 6 7 8 8 23 :;>j
145 0 0 0 I 2 2 2 2 3 3 4 4 4 5 5 5 5 6 6 7 7 8 8 23 0
• 0
,..
146 0 0 0 1 2 2 3 3 3 3 4 4 4 5 5 6 6 6 7 7 8 8 23
147 0 0 1 I 2 2 2 3 3 4 4 4 4 5 5 6 6 6 7 7 8 8 23 '"
24 ~
148 0 0 I 2 2 2 3 3 4 4 4 5 5 5 6 6 6 6 7 8 8 '0-"
149 0 0 I 2 3 3 3 3 4 4 5 .5 5 5 6 6 7 7 8 8 24
150 0 1 2 2 2 3 3 4 4 4 5 5 5 .5 6 6 7 7 7 8 8 24
<:
151 0 1 I 2 2 3 3 3 3 4 4 4 5 5 6 6 6 7 7 7 8 8 24
:r
r:r
152 0 0 0 2 2 2 3 3 4 4 4 4 5 5 6 6 6 7 7 7 7 8 8 24 ..,<1>
aq
153 0 0 1 2 2 3 3 3 4 4 .') 5 5 6 6 6 7 7 7 7 8 8 24
154 0 0 2 2 2 3 3 3 4 4 4 4 5 5 .') 6 6 6 7 7 8 8 8 24 ..,0
155 0 2 2 3 3 3 3 4 4 5 5 5 5 6 6 6 7 7 8 8 8 24 0
C
"1:l
156 0 0 0 1 2 2 2 3 3 4 4 4 5 5 5 5 6 6 6 7 7 8 8 8 24 '"
157 0 I I 2 2 2 2 3 3 3 4 4 4 5 5 6 6 6 6 7 7 8 8 8 24
158 0 0 0 2 2 3 3 3 3 4 4 4 5 5 6 6 6 6 7 8 8 8 24
159 0 0 I 2 2 3 3 4 4 4 4 5 5 6 6 6 6 7 7 8 8 8 24
160 0 0 I 2 2 2 2 3 3 4 4 5 5 5 6 6 6 6 7 7 8 8 8 24
161 0 0 0 2 2 2 3 3 3 4 4 4 5 5 5 6 6 7 7 7 8 8 8 24
162 0 0 1 2 3 3 3 3 4 4 4 5 5 5 6 6 7 7 8 8 8 24
163 0 0 2 2 2 2 3 4 4 4 4 5 5 5 6 6 7 7 8 8 8 24
164 0 0 2 2 2 3 3 3 3 4 5 5 5 5 6 6 7 7 7 8 8 8 24
165 0 0 2 2 2 3 3 3 4 4 4 4 5 6 6 6 7 7 7 8 8 8 24
166 0 0 2 2 3 3 3 4 4 4 5 5 5 6 6 6 7 8 8 8 8 24
167 0 0 2 2 2 3 3 3 4 4 4 5 5 5 6 6 7 7 7 8 9 24

-1 -1 I 1 1 1 1 1 1 1 1 1 1 1 1 3 5
-2 1 1 I 1 1 1 1 1 3 3 3 3 3 3 3 3 9
-3 1 1 1 3 3 3 3 3 3 3 3 3 3 3 3 3 11
-4 1 3 3 3 3 3 3 3 3 3 3 5 5 5 5 5 15
-5 -I 1 1 3 3 3 3 3 3 3 3 5 5 5 5 5 5 15
-6 -1 1 3 3 3 3 3 3 3 5 5 5 5 5 5 5 5 5 17
-7 I I 3 3 3 3 3 3 3 5 5 5 5 5 5 5 5 7 7 19
-8 3 3 3 3 3 3 3 3 5 5 5 5 5 5 5 7 7 7 7 21
-9 -I I 1 3 3 3 3 3 5 5 5 5 5 5 5 5 7 7 7 7 21
-10 -1 1 3 3 3 3 3 3 3 3 5 5 5 5 5 7 7 7 7 7 21
<.n
-11 -1 3 3 3 3 3 3 3 5 5 5 5 5 7 7 7 7 7 7 7 23 w
-12 I 3 3 3 3 3 3 5 5 5 5 5 5 7 7 7 7 7 9 23
"
 (It
Table 28.1 (cont.) w
00
Xo Xl Xz X3 Xc X& Xe X7 Xe Xg XIO Xu Xu Xl3 Xu Xu Xu x 17 Xu Xu ,f20 Xu XII X Z3 Xu x,.
-13 I I 3 3 3 3 3 5 5 5 5 5 5 5 7 7 7 799 25
-14 -I I 1 3 3 3 3 3 5 :; 5 I) 5 7 7 7 7 799 25
-15 1 3 3 3 3 3 5 5 5 5 !i 5 7 7 7 7 7 9 9 9 27
-16 -1 I 3 3 3 3 3 5 5 i3 !i 7 7 7 7 7 7 9 9 9 27
-17 -I I 3 3 3 3 .:> 5 5 :; 5 I) 5 7 7 7 9 9 9 9 27
-18 -I 3 3 3 3 3 3 3 .5 5 .') 7 7 7 79999 27
-19 1 1 1 3 3 3 3 5 I) 5
" 5 5 7 7 7 7 7 999 9 27
-20 -1 3 3 3 3 3 I) 5 5 7 7 7 7 9 999 9 29
-21 1 I I 3 3 3 3 5 I)
" 5 I) fi 7 7 7 9 99999 29
-22 1 1 3 3 3 3 3 3 5 5 5 5 7 7 7 9 9 9 9 II 29
-23 1 1 3 3 3 3 5 5 5 5 5 7 9 9 9 9 9 9 11 31
-24 3 3 3 3 3 3 .5 5 5 ."i 7 7 7 9 9 9 9 11 11 31
-25 -1 1 1 3 3 3 Ii 5 5 .'i 7 7 9 9 9 9 11 11 31
-26 -1 I 3 3 3 3 3 Ii i)
" 5 .i :; 9 9 9 9 9 11 11 31
-27 I 3 3 3 3 .1 5 ;) 5 5 9 9 9 9 II II 11 33
-28 -1 3 3 3 3 5 !i 5 7 7 9 9 9 9 11 11 11 33
-29 -I I 3 3 3 5 !) :, 5 ;') j') 9 9 9 9 9 11 11 11 33
-30 -1 3 3 3 3 3 !i ;; I) 7 7 9 9 9 9 9 II 11 11 33
-31 1 I 1 3 3 3 :,3 Ii 5 r, 7 7 7 9 9 9 9 9 11 11 11 33
-32 -1 3 3 3 3 3 5 !i r, 5 :; 7 9 9 q II II II II 33
-33 1 1 3 3 3 3 3 ,; ~ [) 7 7 9 9 \) II II II 11 33
-34 -1 3 3 3 3 ;) 0) .5 .5 7 7 7 7 9 9 9 11 II II 11 11 3:,
-35 3 3 3 5 ;) .') r, 5 7 7 9 9 9 9 II II 11 11 11 35
-36 3 3 3 ;; Ii ii 5 9 9 9 9 11 II II 13 35
-37 I 3 3 3 3 3 5 !) [) 5 9 999 9 II 11 11 13 35
-38 -I I 3 3 3 3 5 r; :; :; 999 11 11 11 11 13 35
-39 I 3 3 3 3 5 ii ;') 7 9 9 9 II II II II 11 13 37
-40 -1 I 3 3 3 5 5 5 !i 7 7 9 9 9 9 II II 11 II 11 13 37
-41 1 3 3 3 .1 Ii 5 5 7 7 7 7 9 999 11 11 11 13 13 37
-42 - I I 3 3 3 .1 5 !i 7
" 7 7 7 7 9 999 II 11 II 13 13 37
-43 -1 3 3 3 3 3 5 I) 5 7 7 9 9 999 II 11 II 13 13 37 ('l
-44 -1 3 3 3 3 5 5 5 5 ."i 7 9 9 9 11 11 11 II 13 13 37 ::r'
III
-4,') I I I 3 3 3 3 5 I) 5 7 7 9 9 9 11 11 II II 13 13 37 "S.
<1>
-46 -1 I 3 3 3 3 .5 ;; I) 7 9 9 9 9 11 11 11 11 11 13 13 39 ..,
-47 1 3 3 3 3 5 5 .'i 7 7 7 9 9 9 9 11 11 11 13 13 13 39 tv
-48 -1 3 3 3 ij I) 7 7 9 9 9 9 11 11 11 13 13 13 39 00
 Table 28.1 (cont.) t"
(l)
(l)
x. Xl x, x, x, x, x. x, Xs x, Xl. X ll Xu Xl3 Xu x" x .. X 17 x .. Xu x,. X 21 x" X 23 Xu x" (')
::>"
-49 - I I 3 3 3 5 5 5 5 579 9 9 9 9 II II II 13 13 13 39 :;0
-50 I I I 3 3 3 5 5 5 7 9 9 9 9 9 II II II 13 13 13 39 o
-51 - I I 3 3 3 .> 5 5 5 7 7 7 9 9 II II II II 13 13 13 39 S.
V>
-52 - I 3 3 3 3 3 5 5 5 5 7 7 9 9 9 II II II II 13 13 13 39 OJ
-53 I I 3 3 5 5 5 5 5 9 9 9 II II II II 13 13 13 39 ::l
0-
-54 I 3 3 3 3 3 5 5 7 7 7 9 9 9 II II II II 13 13 13 39
-55 I 3 3 3 3 5 5 5 5 7 7 9 9 9 9 II II 13 13 13 13 39 <
-56 - I .'i 7 9 II II II II 13 13 13 13 41
S·
3 3 3 3 5 5 5 9 9 0-
(l)
-57 I 3 3 3 3 5 5 7 9 9 9 tI II II II 13 13 13 13 41 ...,
-58 I 3 5 5 5 .) 9 9 9 9 II II 13 13 13 13 13 41 ~
3 3
-59 3 3 3 5 5 5 ;) 9 9 9 II II II II II 13 13 15 41 o
-60 3 3 3 5 5 5 7 9 9 9 II II II 13 13 13 15 41 dc
"0
-61 I 3 3 3 3 3 5 5 5 7 9 9 9 9 II II II 13 13 13 15 41 V>

-62 - I I I 3 3 5 5 5 5 7 7 9 9 9 9 11 11 II 13 13 13 15 41
-63 - I I 3 3 3 3 5 .'i 5 7 7 9 9 11 II II II 13 13 13 15 41
-64 I 3 3 5 5 5 5 7 9 9 9 9 II II II 13 13 13 13 15 43
-65 I 3 3 3 3 5 5 7 9 9 II 11 II II 13 13 13 13 15 43
-66 - I I 3 3 3 5 5 5 5 7 9 9 9 II II II II 13 13 13 13 15 43
-67 -I I 3 3 3 5 5 5 7 7 7 9 9 9 II II 13 13 13 13 13 15 43
-68 I 3 3 3 3 5 5 5 7 9 9 9 9 II II II 11 13 13 15 15 43
-69 -I 3 3 3 5 5 5 7 7 7 7 9 9 II II II II II 13 13 15 15 43
-70 - I 3 3 3 5 5 5 7 7 9 9 9 9 9 II II 13 13 13 15 15 43
-71 - I I 3 3 5 5 5 5 5 9 9 9 II II II 13 13 13 15 15 43
-72 -I 3 3 3 3 3 5 5 7 9 9 9 II II II 13 13 13 1.5 15 43

-73 I 3 3 5 5 5 7 7 9 9 9 II II II 13 13 13 15 15 43
-74 I 3 3 3 3 5 5 .5 7 9 9 9 9 II II II 13 13 13 15 15 43
-75 3 3 3 5 5 5 7 '7 7 9 9 9 II II II 13 13 13 13 15 15 45
-76 -I I 3 3 5 5 5 5 7 7 9 9 9 9 11 II II 13 13 13 13 15 15 45
-77 -I 3 3 3 3 5 5 5 7 7 7 9 9 II II II 11 13 13 13 13 15 15 45
-78 I I 3 3 3 5 5 5 7 7 7 9 9 9 II II 13 13 13 13 13 1.5 15 45
-79 -I I 3 3 5 5 5 7 7 7 7 9 9 9 II II II II 13 13 15 15 15 45
-80 -I 3 3 3 3 .5 5 5 7 7 9 9 9 9 II II II II 13 13 15 15 15 45
-81 I I 3 3 3 5 5 5 7 7 9 9 II II II II II 13 13 15 I.> 1.5 45
-82 -I 3 3 3 3 5 5 7 7 7 9 9 9 9 II II 13 13 13 15 1.5 15 45
Ul
-83 I I 3 3 3 5 5 5 9 9 9 9 9 II II 13 13 13 15 1.5 15 45 W
-84 -I 3 3 3 5 5 5 5 7 9 9 II II II 13 13 13 15 15 15 45 'D
 Table 28.1. (cont.) VI
.j>
o
Xo Xl XI X3 X" Xs Xs X7 XM Xg XlO Xn Xu Xu Xu Xl5 x IG X 17 X 18 Xl. X zo Xn X zz X I3 x z" x"
-85 1 3 3 3 5 5 7 7 9 9 11 11 11 13 13 13 15 15 I.'i 45
-86 1 3 3 5 5 ., 5 5 9 9 9 11 11 11 13 13 13 15 15 15 4.5
-87 3 3 3 3 3 5 5 7 9 9 9 11 11 11 13 13 13 15 15 15 45
-88 -1 1 1 3 3 5 5 5 7 7 7 9 9 9 11 11 11 13 13 13 15 15 15 45
-89 -1 3 3 3 5 5 5 7 7 7 9 9 9 11 11 II 11 13 13 15 15 15 15 47
-90 1 1 3 3 5 5 5 7 7 9 9 9 9 11 11 13 13 13 15 15 15 15 47
-91 1 3 3 3 3 5 7 7 7 7 9 9 11 11 II 13 13 13 15 15 15 15 47
-92 -1 1 3 3 5 5 7 7 9 9 9 11 11 11 13 13 13 15 15 15 15 47
-93 1 1 3 3 .,3 ;"j 5 7 7 7 9 9 9 11 11 11 II 13 13 13 15 15 17 47
-94 1 3 3 3 5 7 7 7 9 9 9 9 11 II 13 13 13 13 15 15 17 47
-95 1 3 3 3 .3, 5 5 5 7 7 7 9 9 11 11 II 13 13 13 13 15 15 17 47
-96 -1 3 3 3 5 5 7 7 7 9 9 11 11 II 13 13 13 13 15 15 17 47
-97 1 3 3 3 5 .'i 5 5 9 9 9 9 11 II 11 13 13 15 15 15 17 47
-98 -1 1 3 3 3 5 5 7 9 9 9 9 II 11 II 13 13 15 15 15 17 47
-99 -1 1 3 3 5 5 5 5 7 7 9 9 11 11 11 II 13 13 15 15 15 17 47
-100 -1 3 3 3 3 5 5 ., 7 7 9 9 9 11 II 13 13 13 15 15 15 17 47
-101 1 3 3 3 5 5 5 7 7 9 9 9 9 11 11 13 13 13 13 15 I., 15 17 49
-102 -1 1 3 3 5 5 5 7 7 9 9 11 11 11 13 13 13 13 15 15 15 17 49
-103 1 3 3 3 Ii 5 7 7 7 9 9 9 11 II 11 13 13 15 15 15 15 17 49
-104 -1 1 3 3 5 5 5 7 7 9 9 9 9 11 II 11 13 13 15 15 15 15 17 49
-105 -1 3 3 3 3 .'i 5 9 9 11 11 11 II 13 13 15 15 15 15 17 49
-106 -1 3 3 3 5 5 ;, 9 9 9 II 11 13 13 13 15 15 15 15 17 49
-107 1 3 3 3 5 5 7 9 9 9 11 II 13 13 13 15 15 15 15 17 49
-108 3 3 5 5 5 5 7 9 9 9 11 11 11 13 13 13 15 15 17 17 49
-109 -1 1 3 3 5 ., 7 7 9 9 9 II 11 II 13 13 13 15 15 17 17 49
-110 -1 3 3 3 3 ., 5 7 9 9 9 9 II 11 11 13 13 13 15 15 17 17 49
-111 -1 3 3 3 5 ,"j 5 5 7 7 9 9 11 11 II 11 13 13 13 15 15 17 17 49
-112 1 1 3 3 3 5 5 7 7 9 9 11 11 11 11 13 13 13 15 15 17 17 49
-113 -1 3 3 3 ;"j 5 5 7 7 9 9 11 II 13 13 13 13 15 15 17 17 49
-114 1 1 3 3 5 5 5 5 7 9 9 9 II 11 13 13 13 13 15 15 17 17 49
(')
-115 -1 3 3 3 5 ., 5 7 9 9 9 9 11 II 13 13 15 15 15 17 17 49 ::T
-116 1 3 3 5 ;) 5 7 7 9 9 11 11 11 13 13 15 15 15 17 17 49 III
'0
-117 3 3 3 3 .'i 5 5 9 9 9 11 11 II 13 13 15 15 15 17 17 49
...,Cb
'"00
Leech Roots and Vinberg Groups 541

We saw in Chap. 26 that, modulo w, TI may be regarded as a 24-

dimensional Euclidean space, in which the Leech roots form a (nonlinear)
set isometric to the Leech lattice. The main result of Chap. 27 is that the
reflection subgroup of Aut(H25,1) is generated by the reflections in the
Leech roots.
In the present chapter we enumerate some of the Leech roots: see
Tables 28.1 and 28.2. We show in §§3-5 how the fundamental roots for
In,1 (n ~ 19) may be easily derived from the Leech roots (see especially
Theorems 1 and 2 in §6). The description in terms of Leech roots provides
a natural explanation for the striking symmetries observed in the
fundamental regions of AutOI4,1), ... ,AutOI9,1) by Vinberg [Vinl1-[Vin4],
[Vin7]. Meyer [Mey2] and Vinberg and Kaplinskaja [VinI51. We also
give a partial description of the fundamental roots for 120,10 ... , 124 ,1 and their
relationship to the Leech roots (see Table 28.3). Figures 28.1a-h show
(part of) the Coxeter-Vinberg diagrams for AutOn,I), 13 ~ n ~ 20.

2. Enumeration of the Leech roots

We discovered the relationship between the groups AutOn,l) and the Leech
lattice after examining an extensive list of Leech roots. Since these roots
promise to have many other applications (see for example the final
chapter), we have thought it desirable to provide a substantial tabulation.
In Table 28.1 the roots are labeled (in the left-hand column) by
nonnegative or negative integers according to whether their coordinates are
integral or fractional. The ordering is otherwise lexicographic, first by the
value of the timelike coordinate t = X25, then by 1X241, 1X231, ....
It is in fact true that every Leech root has t ~ 0, and since this is not
entirely obvious we provide a proof. Let z denote the vector
z = _1 (02511)
70 '

where the notation indicates that the first 25 coordinates are zero. Since
Z'w - -1 (i.e. z E T I), z can be regarded as a point in the real 24-space
that contains the Leech lattice. If x "" (XO .... ,X24 1t) is another point of
T I of norm 2 we have
t 1
N(x-z) - 2 + 35 - 4900 '

showing that t increases linearly with the squared distance from x to z. If

x is one of the particular Leech roots TO, ... ,T24 (see Table 28.1), we have
t = 0, and so
1
N (x - z) - 2 - 4900 .

The coordinates of TO, .... T24 show that they do not lie in a 23-space, and so
they form a simplex whose circumcenter is z. In the language of Chap. 23
these points form an ordinary Dynkin diagram d 25, and they are all the
vertices of a shallow hole, of radius (2-1/4900)112, centered at z. Other
542 Chapter 28

Leech roots r must therefore give strictly larger values of N(r-z), and so
have strictly positive values of t.

The Leech roots (r) in Table 28.1 include all those with t ~ 25, and
were found by a backtracking program. We shall refer to the entries in
this table as the hyperbolic coordinates for the roots. In Table 28.2 we
give Euclidean coordinates for these roots. (The blocks of four numbers
correspond to columns in the MOG arrangement of coordinates - see
Chap. 11.)

Table 28.2 Euclidean coordinates (Yo, ... ,Y23) for the same Leech roots
rj as in Table 28.1. (The columns give i, Yo, ... ,Y23' The roots r_j have
not been multiplied by 2. Here + = +1, - = -1, b = -2, d = -4.)

o 4 0 0 0 o0 0 0 o0 0 0 000 0 o 000 o 000

1 000 0 d 0 0 0 o0 0 0 o 000 o0 0 0 o 0 () 0
2 ++++ ++++ ++++ ++++
3 o0 0 0 o 000 4 0 0 0 o0 0 () o0 0 0 o0 0 0
4 +--- -+++ -+++ -+++ -+++ -+++
5 000 0 o0 0 0 o0 0 0 o0 0 0 4 000 o0 0 0
6 +--+ --++ +-+- ++++ -+-+ ++++
7 o0 0 0 o4 00 o 0 0 0 o0 0 0 o 0 0 () o 0 0 0
8 ++-- --++ ++-- --++ ++++ ++++
9 o 000 000 0 o0 0 0 o4 0 0 o0 0 () 000 0
10 +--- -+++ +-++ +-++ +-++ +-++
11 () 0 0 0 o 000 o 000 o 000 000 0 o4 0 0
12 +--+ -+-+ ++-- ++++ ++++ --++
1., 000 0 o 040 o 000 o 000 o 000 o 0 0 0
14 +-+- -+-+ +-+- -+-+ ++++ ++++
1[, o 000 000 0 000 0 o0 4 0 o 0 0 0 o () 0 ()
it) +--- -+++ ++-+ ++-+ ++-+ ++-+
17 o 000 o0 0 0 000 0 o 0 0 0 000 0 004 0
18 ++-- -++- +-+- ++++ ++++ -+-+
19 o0 0 0 o0 0 4 o0 0 0 o 0 0 0 o0 () 0 o 0 0 ()
20 +--+ -++- +--+ -++- ++++ ++++
21 o0 0 0 000 0 o0 0 0 o0 0 4 000 0 () 0 0 ()
22 +--- -+++ +++- +++- +++- +++-
23 o 000 o 000 o0 0 0 () 0 0 0 o0 () 4 o () 0 0
24 +--- -+++ +--- -+++ +--- -+++
25 o0 0 0 o0 0 0 000 0 o0 0 0 o0 0 0 000 4
26 o0 d 0 o 000 o0 0 0 o0 0 0 o 000 o0 0 0
27 +-+- --++ +--+ ++++ ++++ -++-
28 o0 0 0 o0 0 0 o0 0 0 () 0 0 0 o4 () 0 o 000
29 +-+- -++- ++-- ++++ --++ ++++
30 od 0 0 o0 0 0 o0 0 0 () 000 000 () o 0 0 0
31 ++-+ -+++ +--- ++-+ +-++ +++-
32 o0 0 0 o0 0 0 o0 0 0 000 0 004 0 o0 0 0
33 ++-- -+-+ +--+ ++++ -++- ++++
34 +++- -+++ +--- +++- ++-+ +-++
35 ++-- -+-+ -++- ++++ +--+ ++++
3tl o 0 0 d o0 0 0 o0 0 0 o0 0 0 o0 0 0 o0 0 0
37 +-++ -+++ -+-- -+++ ++-+ +++-
38 +--- -+++ +--- -+++ -+++ +---
39 o0 0 0 o0 0 0 o 0 0 0 o0 0 0 () 0 0 0 4 000
40 +-++ -+++ +--- +-++ +++- ++-+
41 ++-+ -+++ --+- -+++ +++- +-++
42 ++-+ -+++ ---+ +-++ ++-+ -+++
4:1 +-++ -+++ --+- +++- +-++ -+++
44 ++-+ -+++ +++- -+-- ++-+ -+++
4.5 +--- -+++ +--- +--- -+++ -+++
Leech Roots and Vinberg Groups 543

Table 28.2 (cont.)

46 o0 0 0 000 0 o0 0 d o0 0 0 o0 0 0 o0 0 0
47 +--+ -++- ++++ ++++ +--+ -++-
48 +--- -+-- +--- +-++ ++-+ +++-
49 o0 0 0 o0 0 0 o0 0 0 4 0 0 0 o0 0 0 o0 0 0
50 +--+ -++- -++- +--+ ++++ ++++
51 +--+ -+-+ --++ ++++ ++++ ++--
52 +-+- -++- --++ ++++ ++-- ++++
53 +--+ -:...-++ -+-+ ++++ +-+- ++++
54 +--+ -+-+ ++++ --++ ++-- ++++
55 +++- -+++ ---+ -+++ +-++ ++-+
56 +-++ -+++ +-++ -+++ ++-+ ---+
57 ++-+ ---+ +-++ -+++ +-++ -+++
58 o0 0 0 000 0 od 0 0 o0 0 0 o0 0 0 o 000
59 +-+- --++ ++++ -++- +--+ ++++
60 +--- ---+ +--- +++- +-++ ++-+
61 +--- -+++ --+- --+- ++-+ ++-+
62 +++- -+++ +++- -+++ +-++ --+-
63 +--- -+++ --+- ++-+ ++-+ --+-
64 +--- --+- --+- -+++ +-++ +++-
65 ++-- --++ --++ ++-- ++++ ++++
66 +-+- --++ -++- ++++ ++++ +--+
67 o 000 o0 0 0 004 0 o0 0 0 o0 0 0 o 000
68 ++-- -++- -+-+ ++++ ++++ +-+-
69 +--+ --++ ++++ -+-+ ++++ +-+-
70 +++- -+++ -+-- ++-+ +++- -+++
71 ++-- --++ ++++ ++++ ++-- --++
72 +-++ --+- +++- -+++ +++- -+++
73 ++-- -++- ++++ -+-+ +-+- ++++
74 +--- --+- +--- ++-+ +++- +-++
75 ++-+ -+++ ++-+ -+++ +++- -+--
76 +++- -+-- ++-+ -+++ ++-+ -+++
77 +-++ -+++ ++-+ ---+ +-++ -+++
78 ++-+ -+++ ++-+ -+++ ---+ +-++
79 ---+ +-++ ++-+ -+++ ++-+ -+++
80 o0 0 0 o0 0 0 o0 d 0 000 0 o0 0 0 o0 0 0
81 +--- ---+ ---+ -+++ ++-+ +-++
82 +--+ --++ -+-+ ++++ -+-+
83 +--- --+- -+-- +++- ++-+ -+++
84 +--- -+++ -+-- -+-- +-++ +-++
85 2 000 b 202 2 000 o0 2 0 o0 0 2 o2 0 0
86 ++-- -+-+ -++- ++++ -++-
87 +--- -+++ -+-- +-++ +-++ -+--
88 +--- ---+ +++- -+++ ++-+ -+--
89 +-+- -+-+ -+-+ +-+- ++++ ++++
90 ++-+ -+++ -+-- +++- -+++ ++-+
91 +--+ --++ ++++ +-+- ++++ -+-+
92 ++-- -+-+ ++++ -++- ++++ +--+
93 o0 0 0 o0 0 0 o4 0 0 o0 0 0 o0 0 0 o0 0 0
94 +++- -+++ --+- +-++ -+++ +++-
95 ++-- -+-+ ++++ +--+ ++++ -++-
96 +-+- -++- ++++ --++ ++++ ++--
97 ++-+ -+-- +++- -+++ -+++ +++-
98 +++- -+++ +-++ --+- +++- -+++
99 +-+- -+-+ ++++ ++++ +-+- -+-+
100 +++- -+++ +++- -+++ -+-- ++-+
101 ++-+ -+++ +++- +-++ --+- -+++
102 -+-- ++-+ +++- -+++ +++- -+++
103 +--- ---+ --+- +-++ +++- -+++
104 +--- -+-- -+-- -+++ +++- ++-+
105 2 000 b 220 2 000 o0 0 2 o2 0 0 o0 2 0
106 +-++ -+++ +-++ -+++ --+- +++-
544 Chapter 28

Table 28.2 (cont.)

107 +--- -+++ ---+ ---+ +++- +++-
108 +++- -+++ +-++ ++-+ ---+ -+++
109 ++++ -++- +-+- --++ +--+ ++++
110 --+- +++- +-++ -+++ +-++ -+++
111 +-+- -++- --++ ++++ --++
112 o0 b 0 o2 0 0 2 000 o0 0 2 o () 0 2 022 2
113 +--- -+-- ---+ ++-+ +-++ -+++
114 ++-- -++- -+-+ -+-+ ++++
115 +--- --+- +-++ ---+ ++-+ -+++
116 +--+ -+-+ --++ --++ ++++
117 o0 0 0 o0 0 0 o0 0 0 d 0 0 0 o () () (I o0 0 0
118 +--- -+-- +++- --+- +-++ -+++
119 o bOO 022 2 2 000 o2 0 0 o0 \J 2 o0 2 0
120 +--- -+++ ---+ +++- ---+ f++-
121 +--- -+++ ++-+ --+- ++-+ --+-
122 +--- -+++ +-++ -+-- +-++ -+--
123 + -+++ +++- ---+ ---+ +++-
124 +-++ -+++ ---+ ++-+ -+++ +-++
125 o0 0 0 o0 0 0 o 0 0 4 o0 0 0 o0 0 0 o0 0 0
126 ++-+ -+++ +-++ ---+ -+++ ++-+
127 +-+- -++- ++++ ++-- ++++ --++
128 ++-- -++- ++++ +-+- -+-+ ++++
129 +-++ -+++ +++- --+- -+++ +-++
130 +--+ -+-+ ++++ ++-- --++ ++++
131 ++++ ++++ ++++ ++++
132 ++-+ -+++ +-++ +++- -+++ --+-
133 +-+- -+-+ ++++ ++++ -+-+ +-+-
134 +-++ -+++ +++- ++-+ -+++ -+--
135 ++-- --++ ++++ ++++ --++ ++--
136 ++-+ -+++ +++- +-++ ++-+ +---
137 ++++ ++++ ++++ ++++
138 +-++ -+++ ++-+ +++- -+-- -+++
139 200 0 b 022 2 0 0 0 o2 0 0 o0 2 0 000 2
140 ++-+ -+++ +-++ +++- +--- ++-+
141 o0 b 0 o2 2 2 2 000 002 0 o2 0 0 o0 0 2
142 +--- -+++ ---+ +++- +++- ---+
143 +-++ -+++ +++- ++-+ +--- +-++
144 ++++ ++++ ++++ ++++
145 +--- -+++ -+-- +-++ -+-- +-++
146 +--- -+++ --+- ++-+ --+- ++-+
147 +--- -+++ +++- ---+ +++- ---+
148 ++++ -+-+ +-+- ++++ ++-- -++-
149 2 0 0 0 bOO 0 200 0 o2 2 2 2 0 0 0 2 000
150 o bOO o2 0 0 2 000 022 2 o2 0 0 o2 0 0
151 o0 b 0 o0 2 0 2 000 o2 2 2 o0 2 0 o0 2 0
152 +--+ -++- ++++ -++- -++-
153 +--- - + --- +-++ -+++ +++- --+-
154 ++++ --++ ++-- ++++ +--+ -+-+
155 o 0 0 b 000 2 2 000 022 2 o0 0 2 o0 0 2
156 +-+- -+-+ ++++ -+-+ -+-+
157 -+-+ ++++ +-+- ++++ +--+ --++
158 ++-- --++ ++++ --++ --++
159 o 000 o0 0 0 o0 0 0 o0 0 0 o0 0 0 d 0 0 0
160 ++-- -+-+ +-+- --++ -+-+ --++
161 +--+ -++- ++++ +--+ +--+
162 +--- --+- ++-+ -+++ +-++ ---+
163 +--+ -++- ++-- --++ -+-+ -+-+
164 +--- -+-- +-++ -+++ ---+ ++-+
165 +--+ -++- +-+- -+-+ --++ --++
166 +--- -+++ +++- +++- ---+ ---+
167 +--+ -++- ++++ ++++ -++- +--+
Leech Roots and Vinberg Groups 545

Table 28.2 (cont.)

-1 ++++ ++++ ++++ ++++ ++++ ++++
-2 ---+ -+++ +++- -+++ +-++ ++-+
-3 -+-- -+++ +-++ -+++ ++-+ +++-
-4 --+- -+++ +++- +-++ ++-+ -+++
-5 +--- +++- +++- -+++ ++-+ +-++
-6 +--+ ++++ +-+- --++ ++++ -++-
-7 ---+ -+++ +-++ ++-+ +++- -+++
-8 -+-- -+++ ++-+ +++- +-++ -+++
-9 +--- ++-+ +-++ +++- ++-+ -+++
-10 +--- +-++ +++- ++-+ +-++ -+++
-11 +--- ++-+ ++-+ -+++ +-++ +++-
-12 ++++ ++++ ++++ ++++
-13 --+- -+++ ++-+ -+++ +++- +-++
-14 +--- +-++ +-++ -+++ +++- ++-+
-15 -+++ -+++ +--- -+++ -+++ -+++
-16 ++-- ++++ +--+ -+-+ ++++ --++
-17 +-+- ++++ ++-- -++- ++++ -+-+
-18 ++-- ++++ +-+- -++- --++ ++++
-19 --++ +-+- -++- ++++ --++
-20 ++-- ++++ ++-- ++++ -+-+ -++-
- 21 -++- ++-- -+-+ ++++ -++-
-22 -+-- -+++ +++- ++-+ -+++ +-++
-23 ---+ -+++ ++-+ +-++ -+++ +++-
-24 -+-+ -++- ++-- ++++ ++-- ++++
-25 +--- +++- ++-+ +-++ +++- -+++
-26 +--+ ++++ ++-- -+-+ -++- ++++
-27 --++ -+-+ +--+ ++++ +--+ ++++
-28 +-+- ++++ +--+ --++ -+-+ ++++
-29 +--+ ++++ +--+ ++++ --++ -+-+
-30 ++-+ ++-+ +--- -+++ ++-+ ++-+
-31 -+-+ +--+ --++ ++++ -+-+
-32 +--+ ++++ ++-- +-+- +--+ ++++
-33 -+-+ ++-- -++- -+-+ ++++
-34 +-+- ++++ +-+- ++++ +--+ ++--
-35 -+-+ +-+- ++++ --++ -++-
-36 --+- -+++ +-++ +++- -+++ ++-+
-37 -+-+ -+-+ +-+- +-+- ++++ ++++
-38 +--- ++-+ +++- +-++ -+++ ++-+
-39 --++ -++- +-+- ++++ ++++ +-+-
-40 +--- +++- +-++ ++-+ -+++ +++-
-41 -++- --++ +-+- ++++ +-+- ++++
-42 +-+- ++++ +-+- ++++ -++- --++
-43 ++-- ++++ +-+- +--+ ++-- ++++
-44 +--+ ++-- +-+- ++++ +-+- ++++
-45 -++- +-+- --++ -++- ++++
-46 +--+ ++++ +--+ ++++ ++-- +-+-
-47 o0 0 0 b 0 2 0 o0 0 0 o2 0 2 200 2 002 2
-48 +-++ +++- +--- ++-+ ++-+ -+++
-49 ++-- +-+- +--+ ++++ + - -- + ++++
-50 -++- +--+ ++++ -+-+ --++
-51 +-++ +-++ +--- -+++ +-++ +-++
-52 +--+ ++++ --++ -+-+ +--+ ++++
-53 --++ +--+ -+-+ --++ ++++
-54 ---+ -+++ +--- ---+ ++-+ +-++
-55 -++- +-+- ++-- +--+ ++++
-56 2 0 b 0 o0 2 2 o0 0 0 o2 2 0 2 002 o0 0 0
-57 ---+ -+++ +--- +++- ++-+ -+--
-58 --++ ++-- ++++ +--+ +-+-
-59 -+-+ --++ +--+ ++++ ++++ +--+
-60 --++ --++ ++-- ++-- ++++ ++++
546 Chapter 28

Table 28.2 (cont.)

- 61 ---+ -+++ -+++ +++- ++-+ +-++
-62 +--- +-++ ++-+ +++- -+++ +-++
-63 +--+ ++++ +-+- ++-- ++++ +--+
-64 -++- -+-+ ++-- ++++ ++++ ++--
-65 -+-+ --++ -++- ++++ ++++ -++-
-66 ++-- +--+ +-+- ++++ ++++ +-+-
-67 +--+ +-+- ++-- ++++ ++++ ++--
-68 o0 0 0 bOO 2 o0 0 0 o 2 2 0 2 2 0 0 o2 0 2
-69 ++-+ +-++ +--- +++- +++- -+++
-70 +-+- +--+ ++-- ++++ ++-- ++++
-71 ++-- ++++ ++-- ++++ +-+- +--+
-72 +--+ ++++ -++- ++++ ++-- -+-+
-73 --++ ++-- ++++ -++- -+-+
-74 -+-+ +-+- ++++ ++-- +--+
-75 o0 0 0 b 200 o0 0 0 002 2 202 0 o2 2 0
-76 +++- +++- +--- -+++ +++- +++-
-77 2 0 b 0 o0 0 0 o0 0 0 o 2 0 2 2 2 0 0 o 220
-78 ---+ ---+ +--- -+++ +++- +++-
-79 +++- ++-+ +--- +-++ +-++ -+++
-80 +++- +-++ --+- -+++ ++-+ -+++
-81 -+-- ---+ +--- ++-+ ++-+ -+++
-82 2 bOO o0 0 0 o0 0 0 002 2 2 002 o 2 0 2
-83 --+- --+- +--- -+++ ++-+ ++-+
-84 ++-+ +++- -+-- -+++ +-++ -+++
-85 ---+ --+- +--- +-++ +-++ -+++
-86 -+-- -+-- +--- -+++ +-++ +-++
-87 ---+ -+-- --+- -+++ ++-+ -+++
-88 +--- +--- +--- -+++ -+++ -+++
-89 2 0 0 b o2 2 0 o0 0 0 o0 0 0 2 002 o2 2 0
-90 --+- -+++ +--- --+- +-++ +++-
-91 ob b 0 bOO 2 o0 0 0 o0 0 0 2 0 0 2 o 2 2 0
-92 +--- +-++ +--- -+-- ++-+ +++-
-93 -++- -++- +--+ +--+ ++++ ++++
-94 --++ --++ --++ --++ ++++ ++++
-95 -+-+ -+-+ -+-+ -+-+ ++++ ++++
-96 +--+ +--+ +--+ +--+ ++++ ++++
-97 -++- -++- -++- -++- ++++ ++++
-98 +-+- +-+- +-+- +-+- ++++ ++++
-99 ++-- ++-- ++-- ++-- ++++ ++++
-100 +--+ +--+ -++- -++- ++++ ++++
-101 o0 0 0 b 222 o0 0 0 o0 0 0 2 2 2 2 o0 0 0
-102 ++-- ++++ +--+ +-+- ++++ ++--
-103 --+- -+++ -+++ ++-+ +-++ +++-
-104 +-+- ++++ ++-- +--+ ++++ +-+-
-105 +--- +-++ -+++ +-++ ++-+ +++-
-106 ++-- ++++ -++- -+-+ ++++ ++--
-107 --++ +-+- +--+ ++++ ++--
-108 o0 0 0 b 2 2 2 o0 0 0 o0 0 0 000 0 2 2 2 2
-109 +-+- ++++ +--+ ++-- +-+- ++++
-110 2 bOO 002 2 o 000 o0 0 0 2 2 0 0 002 2
-111 ++-- ++++ -+-+ -++- ++-- ++++
-112 --++ +--+ +-+- ++-- ++++
-113 2 0 b 0 o2 0 2 o0 0 0 o0 0 0 2 0 2 0 o2 0 2
-114 -+-- -+++ +--- -+-- +++- ++-+
-115 +-+- ++++ -++- --++ +-+- ++++
-116 -+-+ ++-- +--+ +-+- ++++
-117 --++ -++- -+-+ ++-- ++++
Leech Roots and Vinberg Groups 547

3. The lattices 1n,1 for n ~ 19

Examination of the coordinates given by Vinberg [Vin4] reveals that the
fundamental roots v = (vo, ... , vII-II VII) for 111 ,10 n ~ 17, can be arranged in
the following two "batches": (a) the vectors v (of norm 2) for which
v+ = (0, ... ,0,0, Vo, ... , vII-II VII)

is a Leech root, and (b) the vectors v (of norm 1) for which
v+ = (0, ... ,0,1, Vo,···, vII-II VII)

is a Leech root. Similarly the partial lists of coordinates given by Vinberg

and Kaplinskaja [VinlS] suggest that for 118,1 the fundamental roots are
(a) and (b), and (c) the vectors v (of norm 1) for which
v++ ... (± 1, 1, 1, 1, 1, 1, 1, vo, ... , vl71 V18)

is a vector obtained by doubling a Leech root with fractional coordinates;

and also that for 1191 the fundamental roots are (a) and (b), and (c') the
vectors v (of norm 2) for which
v++ ... (±I,I,I,I,I,I,vo, ... ,vI8IvI9)

is obtained by doubling a Leech root with fractional coordinates.

These observations, proved below, provide an immediate way to decide
if a specified vector v is a fundamental root for 111 ,1 (n ~ 19): does the
supplemented vector r = v + or 'h v ++ satisfy r . r - 2 and r . w - -I?

4. Vinberg's algorithm and the initial batches of fundamental

roots
Vinberg [Vin7] describes an algorithm for finding a set of fundamental
roots for the reflection subgroup of any discrete hyperbolic group. For
AutOIl,I) the algorithm proceeds as follows. For a suitable vector Xo
(which may be called the controlling vector), we first determine the
subgroup H generated by the reflections that fix Xo. We then choose a set
of fundamental roots VIo ... ,VII for H. Further vectors VII +1,VII +2, ••• ,Vt.... are
determined inductively by the following conditions:
(j) Vk E 111 ,10 N(Vk) = 1 or 2,
(ii) the Vk are enumerated in increasing order Of-Vk·XO/N(Vk)1I2,
(iii) Vk·Vj ~ 0 for alII ~ j < k.
We shall regard 111 ,1 (1 ~ n ~ 25) as embedded in 125 ,1 by preceding its
vectors by an initial string of 2S-n zeros. Our observations stem from the
fact that w is then a suitable choice for Xo. This is equivalent to using the
projection
wll ... (2S-n, ... ,23,24170)
548 Chapter 28

of w into In,b and this does not affect inner products with elements of In,I'
It is easy to check that Wn does satisfy the conditions on Xo given in
Vinberg [Vin7].
It is convenient to group the candidate vectors v E In,1 that are to be
considered by the algorithm into batches according to their norm and
height h (v) = -v' w. We shall use nL,h to denote the set of all v E In,1
that have height h and norm 2 (i.e. are long roots), and nS,h to denote the
height h roots of norm 1 (short roots).
A candidate v is accepted by the algorithm (with w as the controlling
vector) if and only if v has non-positive inner products with the vectors of
all strictly earlier batches taken in the order nL,]' ns,], nL,2, nS,2, nL,3, nL,4,
ns, 3,,,, specified by condition (ij). Theorems 1 and 2 determine the first
batches of long and short roots, respectively, that are non-empty.
Theorem 1. For n ~ 2, the first batch of long roots is nL,I, and consists
of the vectors v = (vo, ... , vn-II vn ) for which v+ = (0, ... ,O,vo, ... , Vn-I 1 vn ) is
a Leech root. All candidates from this batch are accepted.

Proof The smallest possible value for h is 1, and the vectors of nL, I are
easily checked to be Leech roots (when extended by O's). Since nL, I is the
earliest possible batch, no candidate is rejected. The Leech root r24 is in
In, I for n ~ 2 and shows that the batch exists (i.e. is non empty). This
completes the proof.
Vinberg takes for his controlling vector Xo the vector z = (on 1 1). It is
easy to see that the algorithm with his controlling vector and initial vectors
accepts sufficiently many more vectors from our first batch to span the
space Rn,l. It follows that the two forms of the algorithm determine the
same fundamental region.
Theorem 2. For 1 ~ n ~ 24 the first non-empty batch of short roots is
nS,25-n, and consists of the vectors v = (vo, ... , vn-II vn ) for which
v+ = (0, ... ,0,1, Vo, ... , vn-II vn ) is a Leech root. Again all candidates from
this batch are accepted.
Proof If v is in ns, h where 1 ~ h ~ 25 - n then the vector
v + = (Oh-I , 1, 025-n-h ,Vo, ... ,Vn-l 1 Vn )

has norm 2 and height 1 and so is a Leech root. This implies that
h = 25 - n, since otherwise the inner product of v + and the particular
Leech root (Oh-I,I,-1,024-h 10) would be positive. Thus v has the
required form.
The batch nS,25-n exists since it contains the vector (-1,0, ... ,010) of
In,I' We have not been able to find a uniform proof that all candidates
from this batch are accepted. For n ~ 19 this can be read from the
results of Vinberg and Kaplinskaja. For 20 ~ n ~ 24 we observe that the
corresponding set of vectors v+ forms a single orbit under the stabilizer of
Itl in the automorphism group of the Leech lattice. So if one member of
this batch is accepted, all must be. But (-1,0, ... , 010) is accepted by
Leech Roots and Vinberg Groups 549

Table 28.3. The first two batches of roots

Dimension Known values of h for which

n InL,11 InS,25-n I nL,h ~0 nS,h ~ 0

1 0 1 none 24 only
2 1 2 1 only 23 only
3 3 1 I only 22 only
4 4 I 1 only 21 only
5 5 1 1 only 20 only
6 6 1 1 only 19 only
7 7 1 1 only 18 only
8 8 1 1 only 17 only
9 9 1 1 only 16 only
10 10 2 1 only 15 only
11 12 1 1 only 14 only
12 13 1 1 only 13 only
13 14 1 1 only 12 only
14 15 2 1 only 11 only
15 17 1 1 only 10 only
16 18 2 1 only 9 only
17 20 2 1 only 8 only
18 22 3 1 only 7, 23 only
19 25 5 1, 17 only 6 only
20 30 12 1,17, ... 5,13, ...
21 42 56 1,9,17, ... 4,8, ...
22 100 1100 1,5,9, ... 3,5,7, ...
23 4600 953856 1,3,5, ... 2,3,4, ...
24 00 00 1,4,5,6, ... 1,2,3, ...

Vinberg's form of the algorithm, and therefore will be accepted by ours.

This completes the proof.
The sizes InL,11 and InS,25-h I of the first non empty batches of long
and short roots can be computed from knowledge of the stabilizer of Itl in
the automorphism group of the Leech lattice, and are shown in Table 28.3.
The first batches of long roots can be described as follows. For n = 18
and 19, nL,1 is the subset of roots called ~~ by Vinberg and Kaplinskaja.
In particular, the Coxeter diagram for the roots in 19 L ,1 is the incidence
graph of the Petersen graph. The subset 20 L ,1 contains 30 roots forming a
copy of Tutte's 8-cage graph (see [Har2, p. 174] and Figs. 28.1g, 28.1h
below). The 42 roots in 21 L,I correspond to the points and lines of a
projective plane of order 4, joined by incidence, while the 100 points of
22 L ,1 are a copy of the Higman-Sims graph (see §3.5 of Chap. 10). The
last finite diagram in this sequence is 23 L ,J, whose 4600 points may be
regarded as the points of the Leech lattice at the minimal distance from
two points which are themselves at the minimal distance. They are
550 Chapter 28

therefore invariant under the group CO 2 (see Chap. 10). The first batches
of short roots could be described similarly.

5. The later batches of fundamental roots

We have much less understanding of the later batches of fundamental
roots. A computer program was prepared to run Vinberg's algorithm (with
controlling vector Xo = z rather than w) on the Cray-l computer at Bell
Laboratories. For n = IS and 19 the algorithm ran to completion and
confirmed the results (obtained partly by calculation and partly by
analysis) of Vinberg and Kaplinskaja [Vinl5]. For n = IS there are 12
more fundamental roots, belonging to the batch IS s .23 , as described in §4.
For n = 19 there are 20 more, belonging to 19 L.17' Since the algorithm is
very time-consuming, and since Vinberg and Kaplinskaja do not list all the
roots for n = IS and 19, we have made Tables 2S.1 and 2S.2 extensive
enough to include the corresponding Leech roots v++.
For n ~ 20, Vinberg [Vin7] has shown that the reflection subgroup of
AutO n •l) has infinite index(I) and so there are infinitely many fundamental
roots. Our extensive but still incomplete computations show that some
vectors are accepted from at least the batches shown in Table 2S.3. In
20 L •17 and 20s .13 , for example, the algorithm appears to accept the vectors
v = (vo, ... , vl91 V20) for which
v +++ = (± 1, 1, 1, 1,2, vo, ... , v 191 v20)

can be expressed as 2r + s, where rand s are Leech roots.

We have also shown that there are symmetries of the fundamental
region for 11 20 •1 that do not preserve the vector w. One such symmetry
takes W20 to
W'20= (13,14,15, ... ,30,31,4SIIlO)

and transforms the initial batches 20 L. I and 20S.5 to image batches 20'£.1
and 20'S.5' The two Socage graphs 20 L,1 and 20'L,1 intersect in 24 roots,
and it seems likely that the full set of long fundamental roots is a union of
S-cages intersecting in this way, while the set of short roots is the union of
infinitely many disjoint images of 20S,5' It would be nice to know mord 2)
The 100 fundamental roots in 22 L ,1 are obtained from the Leech roots
that begin 000.... The latter consist of
42 beginning 0000 ... (producing 21 £. I) ,
56 beginning 0001... (producing 21s,4), and
2 beginning 0002 ...

(J)More extensive information about unimodular lattices in the Euclidean

space R19 upon which this theorem is based can be found in Chap. 16.
(2)R. E. Borcherds (personal communication) has found more, and in
particular has confirmed this guess about the case n = 20.
 t"'
ro
ro
n
:;J"
:;.J
0
::?
Table 28.4. Further Leech roots in hyperbolic coordinates. (This
'"til
::l
together with Table 28.1 includes the 100 Leech roots beginning 000... , 0-

and therefore includes the first two batches of fundamental roots for 121 ,1 <
S·
0-
and the first batch of fundamental roots for 122•1,) .,ro
[fQ

Xo x, x, X3 x, x, x, x, x, x, X ,O Xll X 12 X13 X 14 Xl5 X16 X 17 X18 X19 X'O X 2l x" x" x" x,. .,00
0 0 0 2 3 3 3 4 4 4 5 ,1 5 6 6 6 7 8 8 9 25 C
'J .) 8 8 8 9 25 '0
0 0 0 - 2 3 3 3 4 4 5 5 5 6 6 6 7
0 0 0 2
-3 3 3 4 4 4 ,1 5 6 6 6 8 8 8 8 9 26 '"
0 0 0 3 3 3 4. 4 5 5 5 5 6 6 8 8 9 9 26
0 0 0 2 3 3 4 4 4 ,'i 5 6 6 6 8 8 9 9 26
0 0 0 3 3 3 4 4 5 5 6 6 6 8 8 8 9 9 27
0 0 0 2 3 3 4 4 4 5 5 5 6 6 6 7 7 8 8 9 9 9 27
0 0 0 2 3 3 4. 4 4 5 5 6 6 6 7 8 8 8 9 9 10 28
0 0 0 2 3 3 4 4 ,1 5 5 6 6 7 8 8 8 9 9 9 10 29
0 0 0 2 2 3 3 3 4 4 4 5 5 6 6 7 7 8 8 9 9 10 10 29
0 0 0 2 3 3 4 4 4 5 5 6 6 6 7 8 8 9 9 9 10 10 30
0 0 0 2 2 3 3 3 4 4 5 5 6 6 8 8 8 9 9 10 10 11 31
0 0 0 2 3 3 4 4 5 ,5 6 6 6 8 8 9 9 10 10 10 11 32
0 0 0 2 3 3 4 4 5 5 6 6 7 8 8 9 9 10 10 11 II 12 34
0 0 0 2 3 4 4 5 5 6 6 8 8 9 9 10 10 11 11 12 12 36

til
til
 552 Chapter 28

lat n 'l,3 Ibl n -I.

20 20

19

I.
13

" 23

16 24 I. 2.
let ,. -15 (dl n -I.
20 20

19

22

23

16 2. 16 2.

Figure 28.1. Coxeter-Vinberg diagrams for AutO l3 ,I) - Aut0 2o,I)' The
numbers attached to the nodes indicate the corresponding Leech roots.

All except fifteen of those roots can be found in Table 28.1. The last
fifteen occur much further along in our ordering of the Leech roots, and
are given separately in Table 28.4.
Figures 28.la-h show the Coxeter-Vinberg diagrams for the
fundamental roots of In , [, 13 ~ n ~ 20. The fundamental roots in nL, I
are represented by single circles and those in nS ,2S-n by double circles.
The remaining roots (for n = 18 - 20) are not shown. The numbers
attached to the roots in nL I and ns 2S-n give the subscripts i of the
corresponding Leech roots v+' = ri, whi~h must then be projected onto Rn,1
as described in §5 , For example, node 9 in the diagram for 116, 1
(Fig, 28,ld) refers to the Leech root
r9 = (08,+1,-1,0 15 10) .

Since this node has a double circle it is a short root, so r9 must be

projected onto (-1,0 15 10), which is the fundamental root for 116 ,1
corresponding to node 9. In the next figure node 9 carries a single circle,
and we project r9 onto (+1,-1,0 15 10), a fundamental root for 117 ,1 '
Leech Roots and Vinberg Groups 553

If) n " 8
20
20

2•

•8

23
01

'6 2'
PLUS 12 MORE
POI TS
Is) n "9
20
20

22

23

PLUS 20 MORE
POI TS
PLUS IX) MORE
POINTS

Figure 28 .1 (Cont,)

Every pair of roots with double circles should be joined by a heavy line,
although for simplicity these lines have been omitted from the figure. The
figures have been drawn as subgraphs of the 8-cage, which (except for ten
missing edges) can be seen in full in Fig. 28.1h. The missing edges join
the diametrically opposite pairs of nodes 36 and 6, ... ,29 and 28, and only
their ends have been shown. The reader should note that for each
dimension greater than 13 the diagram has a nontrivial symmetry.
In the center of Fig. 28.1 h there is a circle of ten short roots enclosing
two more. The ten nodes around this circle should be labeled 34, 39, 40,
46, 48, 49, 58, 60, 74 and 80, and two in the center should be labeled 5
and 31. The double lines from node 40 to 9, 17, 23, 26 and 36, and from 5
to 6, 29, 33, 38 and 45 have been indicated by long double arrows. Node
31 should be joined to 15, 18, 25, 28 and 36 by double lines, while the
remaining joins from the nodes 34, 39, ... can be found by rotating the
figure.
 29
The Monster Group
and its 196884-Dimensional Space
J. H. Conway

The Monster simple group was first constructed by R. L. Griess in 1981 as

the group of automorphisms of a certain algebra in Euclidean space of
dimension 196884. This chapter describes the simplified construction given
in [ConI5]. The construction is based on the binary Golay code, the Leech
lattice, and a remarkable nonassociative loop (closely related to the Golay
code) discovered by R. A. Parker.

1. Introduction
The Monster (or Friendly Giant or Fischer-Griess) simple group of order
808017424794512875886459904961710757005754368000000000
= 246320597611213317. 19.23.29. 31 .41 ·47· 59·71 (I)

was first constructed by R. L. Griess [Gri4], [Gri5]. This group M is also

studied in [ConI6]-[ConI8], [Con24a], [Frel]-[Fre5], [Gri2], [Gri6]-
[Gri8], [GrilO], [Gril0aJ, [Nor5], [Nor6], [Smil3], [Soi2], [Th04]-[Th06],
[Tit8], [Tit9]. The main goal of this chapter is to describe more briefly the
simplified construction given in [Con15]. The verification that the
construction works reduces to elementary but tedious calculations. The
often repeated phrase "explicit calculation" is an implicit reference to the
Appendices of [ConI5], where these calculations are performed in detail.
Figure 29.1 shows various subgroups of the Monster that arise in ot!!:
construction, with some information about their structures. The group N
is the normalizer of a certain fourgroup in the Monster, whose non-trivial
elements are X_I> .v_I> Z-l. ~ncefl £.ermutes these, it has an obvious
homomorphism onto S3, and N x , Ny, N z are the preimages of the thr~
subgroups S2 of this S3 - in other words they consist of the elements of N
The Monster Group and its 196884-Dimensional Space 555

STRUCTURES:
+---+ THE MONSTER

+---+ 21+24

Figure 29.1. Some subgroups of the Monster.

that commute with X-I> Y_I> Z-l respectively. The intersection of these
three groups, which is also the intersection of any two of them, is the group
g.yz..!hat centralizes the fourgroup X-I> Y_I> Z-l. The groups Q, Qx, ~,
Q., Q..yz are normal 2-subgroups of the above groups, while the groups Gx ,
~Y' G..: ar: the full centralizers in the Monster of the respective involutions
X-I> Y-I> Z-l·

The time has come to explain all those bars, and quickly describe our
construction. The main idea is that we give an explicit construction for a
certain group N of order
201889334420601569280 = 248 . 34 . 5 . 7 . 11 . 23,

as a group of permutations of triples of elements in a non-associative loop

f#> discovered by R. A. Par~r [Par2]. This group N turns out to be a
fourfold cover of the group N described above. The definition of N makes
it very easy to define certa.i!t reQI"ese.!!tations of its subgroups N x , Ny, N z
(the preimages in N of N x , Ny, N z ) on some quite large-dimensional
Euclidean spaces, a!!.d we can put these representations together to build a
representation for N on a space of dimen~n 196884, on which we also
define an algebra structure invariant under N.
In fact we give three different definitions of this algebra, which make it
~ear ~ha~ it is actually invariant und~ each of three rather large groups
Gx , Gy , Gz th~t are not subgroups of N, and we can define the Monster to
~ the group G that these generate. We give a surprisingly easy proof that
G is finite, but the computation of its order, and its identification with the
abstract group already known as the Monster, depend on a theorem of
Stephen D. Smith.
556 Chapter 29

2. The Golay code ~ and the Parker loop ~

As usual, we use r.t1 for the binary Golay code, consisting of 212 binary
codewords of length 24, with weight distribution
0 1 8759 12 2576 16 759 241

(see (2.8.2) of Chap. 3). In this chapter we shall use multiplicative rather
than additive notation for the Golay code r.t1, but will continue as in
previous chapters to identify elements of r.t1 with the corresponding sets
(r.t1 -sets). The Parker loop ~ is a (non-associative) system with a center
{l, -I} of order 2, modulo which it becomes a copy of r.t1. We shall write
a, b, c, d, e, f ,... for typical elements of ~, and use d -+ d for the
homomorphism onto r.t1. Thus d and -d are the two preimages of the
r.t1 -set d. In particular, nand - n are the two preimages of the universal
n.
r.t1-set
So you already know how to multiply in ~ up to sign. The exact rules
for the signs are rather subtle (Appendix 1), but are not needed for an
overview of the construction.

3. The Mathieu Group M 24; the Standard Automorphisms of ~

The automorphisms of the Golay code r.t1 form the Mathieu group M 24 of
permutations of n. (Regarded as an abstract group, of course, r.t1 has
other automorphisms.) The group M 24 is 5-transitive on n,
of order
24·23·22·21·20·16·3 = 244823040.
The standard automorphisms of ~ are those that become elements of
M 24 when we ignore the signs of Parker loop elements. They form a group
we call SF. (Regarded merely as an abstract group, ~ has some other
automorphisms.) A standard automorphism is called even if it fixes both
nand - n, and odd if it interchanges them.
We shall use 7r for the typical standard automorphism of ~, and 1r for
the corresponding element of M 24. The homomorphism 7r -+ 1r is 212_to-l
(Appendix 1), and 9'has structure 2 12 . M 24 •

4. The Golay Cocode rr5* and the Diagonal Automorphisms

The 7r for which 1r is trivial are the diagonal automorphisms - they
merely change the signs of certain elements of P. The diagonal
automorpqisms correspond one-for-one with the elements of the Golay
cocode r.t1 of weight distribution
0 1 124 2276 32024 41771,

and we shall give them the same names 0, E, L.... As usual we take the
elements of r.t1* to be subsets of n
modulo addition of r.t1-sets. The rule is
that 0 changes the sign of the preimages of just those r.t1-sets it intersects
oddly - in other words
 The Monster Group and its 196884-Dimensional Space 557

As a standard automorphism 0 is even or odd according as 101 is even or

odd.
The diagonal automorphisms form an elementary abelian normal
subgroup of order 212 in the group!/' of all standard automorphisms of fffJ.
In fact!/' has no subgroup isomorphic to M 24 - its structure is a non-split
extension 212 . M 24 of the diagonal automorphisms by M 24.

5. The Group N of Triple Maps

We define N to be the group of permutations of triples of elements of ~
generated by the following maps (which we specify by giving the image of
a typical triple (a, b, c»:
Xd: (dad, db ,cd) x .. : (a",b",c") or «a"")-I, (C,,)-I, (b .. )-I)

Yd: (ad, dbd ,de) Y.. : (a",b",c") or «C .. )-I, (b"")-I, (a,,)-I)

Zd : (da , bd,dcd) z .. : (a", b", c,..) or «b .. )-I, (a,,)-I, (c .. )-I)

(according as 7r is even or odd),

where d ranges over the Parker loop ~, and 7r over its standard
automorphism group S.
There is an obvious "triality" automorphism that cyclically permutes x,
Y, Z in these and later notations. In future we shall often make only one
of three statements or definitions related by triality, and write "(&c)" to
indicate that the reader should derive the others for himself.

6. The Kernel K and the Homomorphism g - g

The success of our construction depends vitally on the fact that when taken
modulo a certain kernel K of order 4, N becomes the normalizer of a
certain fourgroup in the Monster. The kernel K consists of the identity
and the three elements

We shall write g - g for the canonical homomorphism N - N /K, and

r = r/(K n r) for the image of a subgroup r of N under this
homomorphism, and in particular, if for N /K.
On a first reading, it is wise to ignore elements of the kernel, and so
identify every element g with the corresponding g, even though this is not
consistently possible.

7. The Structures of Various Subgroups of N

The structures we give here can be verified by explicit calculations, but
most of them become obvious in terms of the representations defined in
later sections.
558 Chapter 29

The group N itself has structure

(2 2 x 22) 211 222 (S3 x M 24 )

kx X_I X6 Xd X6 X".
ky Y-I Y6 Yd Yo Y".
kz Z_I Zo Zd Zo Z".

(0 even) (0 odd)

(Below the various composition factors we have given generators modulo

earlier factors. We note that X6 = Y6 = Zo (0 even), and Xd Yd Zd = U Of
course the structure of if is obtained by omitting the first factor 22.

The homomorphism onto S3 is obvious - any of our triple maps takes a

typical triple (a, b, c) to another triple (A, B ,C) in which each of A, B, C
depends on just one of a, b, c. The elements for which A depends only on
a form a subgroup N x (&c), and the intersection of N x , Ny, N z (which is
also the intersection of any two of them) is called N xyz ' The group N x has
a normal subgroup Qx == <Xd,Xo> (&c), and the intersection of all (or
any two) of Qx, Qy, Qz is called Qxyz'

8. The Leech Lattice A24 and the Group Qx

The reader should already be familiar with the Leech lattice A24 . For our
purposes, it is convenient to note that it is generated by the vectors (where
the usual factor of 8- 1/ 2 is to be understood)
"11. d = (2on Ii> Oelsewhere) ,

"11.; = (3on;, -lelsewherJ ,

as d ranges through ~ and i through fl (of course, "1I.d depends only on

d). If i, j, k ,... are elements of fl, we write
"1I.;jk ... = "11.; +"1I. j +"1I. k + ....
It can be checked that if {i ,j, k ,... } is a ~-set, then "1I.;jk ... E 2A 24 . So
when we work modulo 2A 24, we can regard "1I.;jk ... as determined even when
o - {i ,j, k ,... } is specified only modulo addition of ~ -sets, and so we call
it "11.6 (OE ~*).
There is a wonderful homomorphism, with kernel {XJ,X-I}, from
Qx = <Xd,Xo> onto A 24/2A 24 . It is defined by

Xd -+ Xd , Xo -+ Xo,
where Xr denotes the image of">-r in A 2J2A 24 • We note that every element
of A24/2A24 can be expressed in the form
(d E ~,OE ~*),

the expression being unique apart from the identity Xd = X-d'

The Monster Group and its 196884-Dimensional Space 559

9. Short Elements
We shall write Xdob = Xd + X b , Xdob = XdXb (&c), and will use r for the
generic subscript d . () that occurs here. A vector A or Ar that is a minimal
vector in the Leech lattice will be called short, as will the corresponding
objects X, X" x" X" Xr (&c).
Here are all the short vectors up to sign, with the names of the
corresponding elements of A 24 /2A 24 :

Xij 400 ;, -400j , OelsewherJ,

A nojj 400 j, 400j , Oelsewhere) ,

Xdoj 300 j, 100 d, -1 elsewhere) , (i ¢ d),

X(n)H (-200 b, 200 rest of d ' Oelsewhere) ,

where in the last line d is an octad, () is an even subset of d, and the factor
n is present just when V21{)1 is odd. (The actual vectors of A24 displayed
above are Aj - Aj for X jj , Ad - Anojj for Xndojj, and Ad - Anoij - Anokl
for Xdojjkl.) So a subscript r indicates shortness just if it appears in the
above list.

10. The Basic Representations of N x

The group N x has the following representations (for proof see
Appendix 4).
A representation on a 24-space 24x spanned by vectors ix (i En). In
this, Xd and Xb act trivially, Yd and Zd both act by changing the sign of
those ix for which i E d, and x .. acts as the permutation if.
A representation on a 4096-space 4096 x spanned by vectors (d); and
(d); that satisfy the relations, in which u denotes a sign - or +:
(-d)~ = -(d)~, (und)~ - (d)~.

The actions of the generators on 4096x are given in Table 29.1.

A representation on a 98280-space 98280 x spanned by vectors Xr that
are permuted exactly like the short elements xr
of Qx under conjugation,
and satisfy X-r = -Xr .
The entries in T.able 29.1 are easily checked from the exact definitions
in Appendix 4, and show in particular that the elements k x , k y , k" X-I all
act as plus or minus the identity in these representations, according to the
display:

+ +
+
+ + + +
560 Chapter 29

We can now build the desired representation 196884x of N x as

300x + 98304x + 98280x ,
where 300x is the symmetric tensor square of 24x , and 98304x the tensor
product 4096 x ® 24x . The above display shows that k x , k y , kz all act
trivially in 196884x~hich we can therefore regard as a representation of
the quotient group N x = Nx/K.

Table 29.1. Action of the generators of N on basis vectors.

Notes: D - d-l, 0' - J n i, u - If;} nil, m = 10 n e'l. n - lI2lo'l + m.
g Action of g on the vectors
X d .•
(0 even) (0 odd)
Xe X(-U"d .• X(-l)"'ede .• ix (ed); (de);-
Ye X(-M"d ...' X(-O)"'ed .• (-J)uix (ede); (ed);-
Ze XO"d ...' XOMde .• (-J)uix (de); (ede );-

X E =YE:=: ZE (. even) X d ,.• ix (d'); (d');-

x, X D ,.• ix (D');- (D');

y, (. odd) ZD' .• iz (D'): (D');
Z, Y D ,.• iy (D'); (D');

X']I' = Y .. = Z1t ('Ir even) X d • .••

(i r ) x (d"); (d");-

11. The Dictionary

Here is a dictionary that identifies the three spaces 196884x , 196884y ,
196884z :
(u)x (;Oy (;j)z = (;0, say,
(ij)x Yij + Y o.jj Zjj - Z o.jj (&c),

Xd·j d+ ®y j d- ®z j (&c),

1 1
X(n)d'6 8" ~ - 6,. Y(O)d-E 8" ~ - .,6 Z (O)d" (&c).
• •
Here in line I, (jj)x means (Ox ® (;)x,
in line 2, (ij)x means (Ox ® (j)x + (j)x ® (;)x,
in line 3, d" ®x j means (d); ® (Ox; and,
in line 4, IJI = 8, () and E range over the 64 C€* -sets that are
representable by even subsets of d, and -6,. = (_I)16.1/2+1<I!2
There are fairly obvious inner products on these spaces, under which
basis vectors are orthogonal except when equal or opposite, and the basis
vectors have norm 1 except that (ij)x - (Ox ® (j)x + (j)x ® (;)x has
norm 2 (&c).
The identifications given by the dictionary equate these inner products.
One can also check by explicit calculation that the actions of elements of
N (as given in Table 29.I), are preserved by the dictionary.
The Monster Group and its 196884-Dimensional Space 561

12. The Algebra

From now on, we can regard ourselves as having defined a single space of
dimension 196884, a single inner product on that space, and a well-defined
action of N upon it. In the same spirit, we next define an algebra
multiplication (.) on the space, by making three definitions for • on
196884 x , 196884y , 196884 z , that are checked to be equivalent by explicit
calculations. Here is the definition for 196884x :

Mx • N x = 2(MN+NM)x,

(X r · s =X±I),
(x r .s short),
(otherwise) ,

d U ®x A • e T ®xJl = ~ [(A,Jl) - 2(A,A r )(Jl,A r ) ]Xr

($)

the last term being present just if (d) ~ = ± (e)~.

Here A and Jl denote vectors of 24x , and if M is a 24 x 24 symmetric

matrix, then Mx is the corresponding element of 300x ' The vectors (d)~
and (e) ~ are typical basis elements of 4096 x , and ($) is the condition that
Xr is a short element taking (d)~ to (e)~. We only enter one of two equal
products u • v and v • u.

13. The Definition of the Monster Group G, and its Finiteness

The group Nx has structure 2 1 + 24 . 2 11 M 24 , in which the quotient group
211M24 can be regarded as the group of monomial autom2!phisms of A24
taken modulo 2A 24 . It can be enlarged to a group Gx of structure
21 +24 . COl by replacing the group of monomial automorphisms of A24 by
the group COl of all automorphisms of A24 taken modulo 2A 24 .
When this is done (see Appendix 5), it is obvious that the group Gx
still has a representation on the space 196884x , and that 2ur definition of
the algebra on 1.2.6884x is actually invariant under Gx . (The main
difference is that N x permutes the vectors Xr in three orbits according to
the shapes (± 42022), (± 2 80 16), (+ 3 ± 123 ) of the corresponding Leech
lattice vectors, whereas Gx fuses these three types.)
562 Chapter 29

Of course the algebra is equaily invariant under the analogous groups

ib" (£. y:!e define the Monster group M to be the group G generated by
Gx , Gy , Gz (in fact it is generated by any two of them). This group G
plainly preserves the algebra multiplication and the inner product defined
on our 196884-space.
The fact that this group is finite is almost a triviality. Explicit
calculations show that if t is the vector

then the map v - v * t is diagonalizable with a single eigenvalue of I (at

1), and that all the other eigenvalues are at most 1/4 and correspond to
eigenvectors orthogonal to t .
It follows that t can have only finitely many images under G. To see
this, observe that if t had infinitely many images, then one of them, I' say,
would be very close to I and have the same norm. So we could write
tI = t + Ow + 0 (0 2 ),

where w is a nonzero vector orthogonal to I and 0 is a small real number.

But then
II * II = t *t + 20w * I + 0 (0 2 ) ,
and since Ilw*11I ~ 1/4 1Iwll, (2) and (3) prohibit t!!.e necessary equation
t * t = t Since the images of v span the space, G is isomorphic to the
I I I •

group of permutations it induces upon them, and so is a finite group.

14. Identifying the Monster

In order to identify G_with the Monster .&roup, we first note that the
centralizer of X-I in G plainly contains Gx , for which the irreducible
constituents of 196884x have dimensions I, 299, 98280, and 983Q1. In fact
these spaces are invariant under the full centralizer of X-I in G. This is
because the 98304-space is just the space on which X-I has eigenvalue -I,
and explicit calculations show that on this space the mean squared
eigenvalue of the map v - v * u is respectively
2 I 3
3"(u,u), 2'4(u,u), 64"(u,u) ,

if u is a vector in one of the constituents I, 299, 98280. The 2 x 98280

vectors X, are now determined as the unit vectors that are eigenvectors for
all the maps v - v * M, M in 300x . The only non-trivial element fixing
all the X, is X-I itself. It follows that the centralizer of X-I in G is exactly
the group 2 1 +24 . COl we can see. Stephen D. Smith [SmiI2] has shown
that a finite group that has an involution with this centralizer has order
given by (I), and has another involution ~ose centralizer is 2.F2' where
F 2 is the Baby Monster - in other words G is the group usually known as
the Fischer-Griess Monster.
The Monster Group and its 196884-Dimensional Space 563

Appendix 1. Computing in r!J>.

In r!J> we have the relations (the sign rules):

a2 = (_0101/4,

ba = (-0 10 n b1/2. ab,

a(bc) = (-0 10 n b n EI . (ab)c,

and it is easy to see that in fact these provide an abstract definition for r!J>
(if it exists). For let us suppose that eJ, e2, ... , el2 are preimages of the
12 elements of some basis of C{j, so that any element of r!J> can be written
as ± F (e I, . . . , e 12)' where F is some bracketed product of some sequence
of its argument~, possibly with repetitions or omissions.
Then an obvious condition that an equation between two such products
should hold in r!J> is that it hold in C{j. On the other hand, C{j is an
elementary abelian 2-group, and the category of such groups is defined by
the laws a 2 = 1, ba =ab, a(bc) = (ab)c. This means that if the two
products are equal in C{j we can verify this fact by repeated applications of
these laws. Using the sign rules above, we can therefore check whether the
two products have the same or opposite sign in r!J>.
The argument also shows that any element ii- of M 24 lifts to 212 distinct
standard automorphisms. For if ii- takes el, ... ,e12 to iI, ... ,i12 we
can pick (in any of 212 ways), preimages I I, ... ,f 12 for iI, ... ,i12. No
matter how this is done, II, ... ,/12 will satisfy the same relations as
e I, . . . , e 12, so there is an automorphism 7r: ei - Ii.
Appendix 2. A Construction for r!J>.
Here is the shortest explicit construction we know for r!J>. The elements of
r!J> will be just the elements of C{j, equipped with formal signs. The loop
multiplication A . B is defined in terms of the addition A + B in C{j by
the formula
A· B = (_oIO(A)nBI (A+B)

where 0 is a certain function from C{j to C{j*. The sign rules for ;JjJ then
reduce to the following requirements on 0 (the congruences being
modulo 2):-
p(A): 10(A) nAI == 1/4IAI,
q(A,B): 10(A) nBI + 10(B) nAI == v21A nBI.

r(A,B,C):
10(A) n BI + 10(A +B) n cl + 1==IAnBncl.
+ 10(B) n cl + 10(A) n (B+c)1
564 Chapter 29

We satisfy these as follows. We first require that 0 have the property

oU +B) - O(A) + O(B) + (A n B)
(the addition here being in C€*), which says that it is a quadratic function
from C€ to C€* whose associated symmetric bilinear function is A n B.
In characteristic 2 such a requirement still leaves us free to specify 0 at a
basis EJ, E 2 • . . • • E\2 for C€, so we can demand further that for
n = 1,2 •.... 12 in turn we have:

I == j
lh IE; n En I - 10(E;) n En I (j < n),
10(En) n E; IJ4IEn I (j =n),
0, say, (j > n).
(Recall that a C€* -set is exactly specified by giving the parities of its
intersections with the E;, since C€ and C€* are dual groups.)
These equations ensure thatp(A) and q(A,B) hold when A and Bare
among the E;. But then, using the quadratic property and the well known
formulae
Ix + yl ... Ixl + Iyl - 21x n yl , (X + y) n z = x nz +y nz ,
we find that rU,B,C) holds universally, that r(A,B,C) enables us to
deduce qU,B+C) from qU,B) and q(A,C), and finally that qU,B)
enables us to deduce p(A+B) from pU) and p(B). Inductively, this
shows that all of p(A), q(A,B), r(A,B,C) hold universally.
Appendix 3. Some Relations in Qx.
By direct computation with triples, one can verify that the elements Xd and
Xa satisfy the relations:
[xa,x,] = 1,

[x d, x]
e
Ii nil+lil·liI/4 ,
- x -1

e -
[X d, X] x -I
Ii n e-1/2 ,

X ...
.Y -
, - x a" x d x e -- zli
-I
nil/2 . x de . x-d n e'

where in the last relation J n e is to be regarded as a C€* -set. We shall

take the last (and hardest) of these as an example. We have
XdXe: (a,b,c) - (***,e.db,cd.e) ,

Xde : (a,b,c) - (***,de.b,c.de) , (5)

where the symbols *** indicate some more complicated expressions whose
exact values won't be needed. Now since Z_I : (a,b,c) - (-a,-b,c) and
[d,e] = (_I)IJ ne I/2 we have
The Monster Group and its 196884-Dimensional Space 565

line-1/2 .Xde'.(a, b , c )-(*** , edb

Z-I . , c. d)
e ,

which needs only some reassociation to become (4). But the diagonal
automorphism J n i actually reassociates any three-term product two of
whose terms are d and e. For example, since IJ n il is even, we have
(c.de)Jne = cJni.de = (_l)lcnJnel.c.de = cd.e

from the definitions in Appendix 1. Accordingly

Ii n e-I/2 .Xde,xd-
Z_I '
ni' (a, b ,c ) - (*** ,e.,c
db d .e )

agreeing with (4). (We need not check the first coordinate, since it IS
easily checked that triples with abc = 1 are taken to other such triples.)
This relation shows that L I and kx = x nLI are in Qx, and from the
previous relations we see that modulo Z_I and kx the group Qx becomes
abelian, after which it is easy to check that its order is at most 226 , and
that of Qx at most 225.
But the relations show much more; that indeed there is a 2:1
homomorphism from Qx onto A 24/2A 24, taking xr
to Ar . To see this, one
only has to check the images of the above relations. Once again we
consider only the last one. We have:

and so

Now J n i is an even subset of ii, and so can be expressed as the union

of m 2-element sets ij, where m = IJ n i1/2. From the definitions,

which imply the desired relation

Ad + Ae = Ade + mAn + AJne'

(We remark that Z_I == Xn modulo K, since z-Ixn = k x , so that the

last of the relations with which we introduced this Appendix reads
- - - -
Xd' xe = Xnm'de' Xdne

in Qx.)
566 Chapter 29

The relations we have given therefore show that the group Qx is at

most a double cover of A 2J2A 24 , and so has structure (at most) 21+24.
But it is easy to see that no non-trivial element of this double cover fixes
every triple (a, b, C), so this is the exact structure. We shall see in
Appendix 4 that Qx is the kernel of the representation 24x of N x , the
image being a group of structure 212 :M24 . This gives the structure of N x ,
and the other structures in §7 are easily deduced.
Appendix 4. Constructing Representations for N x'
The representation 24x . We extend f!jJ by adjoining a formal zero
satisfying
dO = Od = 00 = 0 (d E f!/J) .

Obviously we can regard N as acting on triples of the larger system

f!jJo = f!jJ U {oL We now define formal objects ix, -ix
ix = {(d,O,O):i ~ ti)' -ix = {{d,O,O):i Ed} (&c).

It is easy to see that N x acts on these as follows

Yd, Zd: ix - {_l)IJnilix, x.".: ix - (r'') x ,

so demonstrating that a quotient group of structure 212:M24 acts faithfully,

and that the kernel is exactly Qx.
The formal objects ± ix are permuted exactly like the coordinate
vectors
(± Ion i, Oelsewhere)

for the space of the Leech lattice, and so we can abstractly identify them
with these vectors, so defining a 24-dimensional representation 24x of N x .
The representation 4096 x ' We define
d; = ({O,d,O), (O,- Od,O)}, di = {(O,O,d), (O,O, Od)} (&c),

for each d E f!JJ. Then there are 2·4096 distinct objects d;, which are
permuted by N x , and there is an invariant pairing on these, given by
changing the sign of d. We can therefore identify these objects with
2·4096 vectors d; spanning a 4096-dimensional space 4096x by adding the
relations (-d); = -(d);. The definitions have been arranged so that kx
(which takes (a,b,c) to (a,-Ob,Oe)) acts trivially, and in fact the
kernel of this representation is the fourgroup {l,y n,z-n,kxL (The
elements ky and k z both act as -1.)
The representation 98280 x ' The group Qx has 2·98280 short elements xr
corresponding~o the 98280 short elements Ar of A24/2A24· Under
conjugation, N x permutes these in such a way that the pairing x" X-r is
preserved.
We can think of this as an action of N x on a 98280-dimensional space
98280 x spanned by vectors Xr that are permuted just like the x" and
The Monster Group and its 196884-Dimensional Space 567

satisfy X-r = -Xr . We usually prefer to regard it as a representation of

N x in which K acts trivially. One can check that the full kernel of this
representation is the group of order 8 generated by K and X-I'
Appendix 5. Building the Group Gx .
Now in all our constructions we enlarge the group 212:M24 of
monomial automorphisms of A24....!0 the group 2 . COl of~ll automorphl,sms
of A24 , and so obtain a group Gx strictly larger than N x . In fact, Gx is
essentially defined just by making sure that this process works for 24x and
4096 x - we shall see that 98280 x takes care of itself.
It is well-known that an extraspecial 2-group 21+ 2n has a u!!;ique 2n_
dimensional representation up to equivalence. In particular, Qx has a
unique 4096-dimensional representation.
It now follows that if {3 is an automorphism of Qx, then there is a
matrix M (unique up to scalar factors, by Schur's lemma), such that
m4096(g) = M- I . m4096(g) . M

for all g in Qx, where m4096(g) is the matrix representing g in 4096 x ' We
can make M unique up to sign by demanding that it have real entries and
determinant 1. The two such matrices will be called the representors of (3.
Every automorphism of this group 21+ 24 yields an automorphism of the
quotient group 224 C=A 2i2A 24 ) when we factor by X-I' The
automorphisms of 224 that happen in this way are precisely those that fix
the quadratic form (modulo 2) defined on A 24/2A 24 . In particular, there is
such an automorphism a*, say, for each automorphism a of the real Leech
lattice A24 .
We define Gx to be the group of ordered pairs of matrices of the form
(24 x 24 matrix of a, 4096 x 4096 matrix of a representor of a*).

If we restrict the elements a to be the monomial automorphisms, we get a

subgroup that is almost the same as N x . More precisely, it is isomorphic
to Nx/<k x >, since kx is represented trivially in 24x and 4096 x '
The group Gx obviously has 24 and 4096-dimensional representations
extending those of Nx/<k x >. However, it also has a 98280 dimensional
x
representation, since it still permutes the short r • (To see this, observe
that Gx = Gx/<ky> normalizes Qx, and therefore permutes by
x
conjugation the 2· 98280 short elements r • In the usual way, we regard
this as an action of Gx in which ky acts trivially. So Gx has a
representation of degree 98280 extending that of N x .)
We note that Gx has elements that intermix the Xr corresponding to
Leech lattice vectors Ar of the three shapes (± 42,0 22 ), (± 28,0 16),
(:n, ± 123 ).
 30
A Monster Lie Algebra?
R. E. Borcherds, J. H. Conway,
L. Queen and N. J. A. Sloane

We define a remarkable Lie algebra of infinite dimension, and conjecture

that it may be related to the Fischer-Griess Monster group.
The idea was mooted in Monstrous Moonshine [Con17] that there might
be an infinite-dimensional Lie algebra (or superalgebra) L that in some
sense "explains" the Fischer-Griess "Monster" group M. In this chapter
we produce some candidates for L based on properties of the Leech lattice
described in the preceding chapters. These candidates are described in
terms of a particular Lie algebra L= of infinite rank.
We first review some of our present knowledge about these matters. It
was proved by character calculations in [Con 17, p. 317] that the
centralizer C of an involution of class 2A in the Monster group has a
natural sequence of modules affording the head characters (restricted to
C). In [Kac41, V. Kac has explicitly constructed these as C-modules.
Now that Atkin, Fong and Smith ([AtkI1, [FonI1, [Smi13D have verified
the relevant numerical conjectures of [Con17] for M, we know that these
modules can be given the structure of M -modules. More recently, Frenkel,
Lepowsky and Meurman [Frel1-[FreS] have given a simple construction
for the Monster along these lines, but this sheds little light on the
conjectures.
Some of the conjectures of [Conl7] have analogs in which M is
replaced by a compact simple Lie group, and in particular by the Lie
group E 8. Most of the resulting statements have now been established by
Kac and others. However, it seems that this analogy with Lie groups may
not be as close as one would wish, since two of the four conjugacy classes
of elements of order 3 in E8 were shown in [Que7]-[Que9] to yield
examples of modular functions neither of which are the Hauptmodul for
any modular group. This disproves the conjecture made on p. 267 of
[Kac3], and is particularly distressing since it was the Hauptmodul
A Monster Lie Algebra? 569

property that prompted the discovery of the conjectures in [ConI7]. and it

is this property that gives those conjectures almost all their predictive
power. (Other references to recent work on the Monstrous Moonshine
conjectures will be found in §1.4 of Chap. 1.)
The properties of the Leech lattice that we shall use stem mostly from
the facts about "deep holes" in that lattice reported in Chap. 23. Let w ...
(0. 1.2.3 •...• 24 I 70). The main result of Chap. 26 is that the subset of
vectors r in 1125 ,1 for which r' r - 2. r' W ... -1 ("Leech roots") is
isometric to the Leech lattice. under the metric defined by
d(r,s)2 = norm(r-s). The main result of Chap. 27 is that AutOh5,1) is
obtained by extending the Coxeter subgroup generated by the reflections in
these Leech roots by its group of graph automorphisms together with the
central inversion -1. It is remarkable that the walls of the fundamental
region for this Coxeter group (which correspond one-for-one with the
Leech roots) are transitively permuted by the graph automorphisms. which
form an infinite group abstractly isomorphic to the group COo. of all
automorphisms of the Leech lattice. including translations.
Vinberg [Vin7] shows that for the earlier analogs 119,1 and 11 11,1 of
1125 ,1 the fundamental regions for the reflection subgroups have respectively
10 and 19 walls. and the graph automorphism groups have orders 1 and 2.
For the later analogs 11 33 ,1> •••• there is no "Weyl vector" like w. so it
appears that 1125 ,1 is very much a unique object.
We can use the vector w to define a root system in 1125 ,1' If v E 11 25 ,1
then we define the height of v by -v·w. and we say that v is positive or
negative according as its height is positive or negative. We now define a
Kac-Moody Lie algebra Loo. of infinite dimension and rank. as follows: Loo
has three genera tors e (r). f (r) • h (r) for each Leech root r. and is
presented by the following relations:
[e(r).h(s)] - r's e(r) •
[f(r).h (s)] - -r's f(r).
[e (r),f(r)] - h (r).
[e (r),f(s)] = O.
[h (r),h (r)] - 0 ... [h(r).h(s)),
e (r) {ad e (s)P-r-· - 0 - f(r) {ad f(S)}I-r-.,

where rand s are distinct Leech roots. (We have quoted these relations
from Moody's excellent survey article [Mo021. Moody supposes that the
number of fundamental roots is finite. but since no argument ever refers to
infinitely many fundamental roots at once. this clearly does not matter.
See also [Kac5J.)
Then we conjecture that Loo provides a natural setting for the Monster,
and more specifically that the Monster can be regarded as a subquotient of
the automorphism group of some naturally determined subquotient algebra
of Loo.
The main problem is to "cut Loo down to size". Here are some
suggestions. A rather trivial remark is that we can replace the Cartan
570 Chapter 30

subalgebra H of Loa by the homomorphic image obtained by adding the

relations

for Leech roots rhr2, ... , whenever ChC2, •.. are integers for which
Cl rl + C2r2 + ... = 0.

A more significant idea is to replace Loa by some kind of completion

allowing us to form infinite linear combinations of the generators, and then
restrict to the subalgebra fixed by all the graph automorphisms. The
resulting algebra, supposing it can be defined, would almost certainly not
have any notion of root system.
Other subalgebras of Loa are associated with the holes in the Leech
lattice, which are either "deep" holes or "shallow" holes (see Chap. 23).
(j) By Chap. 23, any deep hole corresponds to a Niemeier lattice N, which
has a Witt part which is a direct sum of root lattices chosen from the list
An (n = 1,2, .. .), Dn (n = 4,5, ... ), E 6, E7 and E 8 • Only 23 particular
combinations arise, and we shall take AllD7E6 as our standard example.
The graph of Leech roots contains a finite subgraph which is the disjoint
union of extended Dynkin diagrams corresponding to these Witt
components W of N, and so our algebra Loa has a subalgebra L[N) which
is a direct sum of the Euclidean Lie algebras E (W) corresponding to those
components (see [Kacl1, [Kac5), [Mool)). For example, Loa has a
subalgebra

Each such subalgebra of Loa can be extended to a larger subalgebra L *[N)

having one more fundamental root, corresponding to a "glue vector" of the
appropriate hole (see Chap. 25). In the corresponding graph, the new node
is joined to a single special node (§2 of Chap. 4) in each component. The
graph for L *[A ll D 7E 6 ) is shown in Fig. 30.1. These hyperbolic algebras
L *[N), having finite rank, are certainly more manageable than Loa itself.

Figure 30.1 The fundamental root diagram for L *[AIID7E6J.

A Monster Lie Algebra? 571

Since the 23 Niemeier lattices yield 23 constructions for the Leech lattice
(Chap. 25), it is natural to ask if we can obtain 23 different constructions
for the Monster using the Lie algebras L[N1 or L *[N1.
GO Each shallow hole in the Leech lattice (see Chap. 24) corresponds to a
maximal subalgebra of LOG of finite rank.
We have made various calculations concerning LOG (finding the
multiplicities of certain roots via the Weyl-MacDonald-Kac formula, etc.).
It is worth noting that these calculations are facilitated by the remarkable
recent discovery that the Mathieu group M12 is generated by the two
permutations
t -12tl , t - 11-t (mod 23) ,

of the set {a, 1,2,3,4,5,6,7,8,9,10, 11}, where Ix I denotes the unique y in

this set for which y == ±x (mod 23) (see Chap. 11 for this result and its
history). The simplest transformation (see Chap. 28) between the usual
Euclidean coordinates for the Leech lattice and its Lorentzian coordinates
uses this description of M 12.
 Bibliography

We have tried to give an adequate bibliography for the major topics discussed in
this book. The numbers in square brackets at the end of an entry give the chapters
where it is cited, and thus also serve as an author index. [B) indicates that the
reference is cited elsewhere in this bibliography. Certain items, not otherwise
mentioned in the book, have been included for completeness. There are about
1550 references. Further references on number theory and many related subjects
will be found in [LeV2) and [Guyl J, and [Mac6) contains an extensive
bibliography on coding theory up to 1977. The books [Grula), [Lekll and the
survey articles in [Toll) and [Gru2) also contain relevant bibliographies.

Abbreviations

AJM - American Journal of Mathematics.

AMAH - Acta Mathematica Academiae Scientiarum Hungaricae.
AMM - American Mathematical Monthly.
ASUH - Abhandlungen Mathematischen Seminar der Universitiit Hamburg.
BAMS - Bulletin of the American Mathematical Society.
BLMS - Bulletin of the London Mathematical Society.
BSTJ = AT &T Technical Journal (formerly Bell System Technical Journal).
CJM - Canadian Journal of Mathematics.
CMB = Canadian Mathematical Bulletin.
COMM = Institute of Electrical and Electronics Engineers, Transactions on
Communications (formerly Transactions on Communication Technology).
CR = Comptes Rendus Hebdomadaires des Seances de I' Academie des
Sciences, Paris, Serie A.
DAN = Doklady Akademii Nauk SSR.
Discr. Math. = Discrete Mathematics.
DMJ - Duke Mathematical Journal.
IANS - Izvestiya Akademii Nauk SSSR, Seriya Matematicheskaya.
IC = Information and Control.
11M = Illinois Journal of Mathematics.
Izv. = Mathematics of the USSR - Izvestiya.
J. Alg. - Journal of Algebra.
JAMS - Journal of the Australian Mathematical Society.
JCT - Journal of Combinatorial Theory.
JLMS - Journal of the London Mathematical Society.
Bibliography 573

JNT - Journal of Number Theory.

JRAM - Journal fiir die Reine und Angewandte Mathematik.
JSM - Journal of Soviet Mathematics.
KNAW - Proceedings of the Koninklijke Nederlandse Akademie van
Wetenschappen.
LNM - Lecture Notes in Mathematics (Springer-Verlag).
LOMI - Zapiski Nauchnykh Seminarov Leningradskogo Otdeleniya
Matematicheskogo Instituta imeni V. A. Steklova Akademii
Nauk SSSR.
Mat. Sb. - Matematicheskii Sbornik.
Math. - Mathematika.
MTAC - Mathematics of Computation (formerly Mathematical Tables and
Aids to Computation).
PCPS - Mathematical Proceedings of the Cambridge Philosophical Society
(formerly Proceedings of the Cambridge Philosophical Society).
PGIT - Institute of Electrical and Electronics Engineers, Transactions on
Information Theory.
PIEEE - Proceedings of the Institute of Electrical and Electronics Engineers.
PIT - Problems of Information Transmission.
PJM - Pacific Journal of Mathematics.
PLMS - Proceedings of the London Mathematical Society.
PNAS - Proceedings of the National Academy of Sciences of the U.S.A.
PPI - Problemy Peredachi Informatsii.
PRS - Proceedings of the Royal Society, London.
PSIM - Proceedings of the Steklov Institute of Mathematics.
PSPM - Proceedings of Symposia in Pure Mathematics, American
Mathematical Society, Providence, RI.
Sb. = Mathematics of the USSR - Sbornik.
SIAD - SIAM Journal on Algebraic and Discrete Methods.
SIAJ - SIAM Journal on Applied Mathematics.
SMD = Soviet Mathematics, Doklady.
Springer-Verlag- Springer-Verlag: New York, Berlin, Heidelberg, London, Paris
and Tokyo.
TAMS = Transactions of the American Mathematical Society.
TMIS - Trudy Matematicheskogo Instituta imeni V. A. Steklova.
A

[Aall) M. J. Aaltonen, Linear programming bounds for tree codes. PGIT 25

(1979),85-90 [91.
[Abrl) M. Abramowitz and I. A. Stegun, Handbook of Mathematical
Functions. National Bureau of Standards Appl. Math. Series 55, U. S.
Dept. Commerce, Washington DC, 1972 [3, 9, 13, 14, 16).
[Adol) J.-P. Adoul, La quantification vectorielle des signaux: approche
alge'brique. Ann. Ttl!6commun. 41 (1986), 158-177 [2,201.
[Affl) L. Affierbach, Minkowskische Reduktionsbedingungen fur positiv
definite quadratische Formen in 5 Variablen. Monatsh. Math. 94
(1982), 1-8 [2, 151.
[Aff2) - and H. Grothe, Calculation of Minkowski-reduced lattice bases,
Computing 35 (1985), 269-276 [2,15).
[Ahml) F. R. Ahmed, K. Huml and B. Sedlacek, editors, Crystallographic
Computing Techniques, Munksgaard, Copenhagen, 1976 [21.
[Airl) T. J. Aird and J. R. Rice, Systematic search in high dimensional sets.
SIAM J. Num. Anal. 14 (1977), 296-312 [11.
574 Bibliography

[Akil] I. Y. Akimova, The problem of optimal arrangement and generalization

of a theorem of Fejes Tdth, Tekh. Kibern. 20 (No.2, 1982) - Eng.
Cybernetics 22 (No.2, 1982), 149 [21.
[Aki2] Application of Voronoi diagrams in combinatorial problems (a
survey), Tekh. Kibern. 22 (No.2, 1984), 102-109 = Eng. Cybernetics
22 (No.4, 1984), 6-12 [21.
J. M. Alexander: see T. L. Saaty.
[Aiel] A. D. Alexandrov, Convex Polyhedra, Moscow, 1950; German
translation, Akademie-Verlag, Berlin, 1958 [211.
[AllI] E. L. Allgower and P. H. Schmidt, Computing volumes of polyhedra,
MTAC 46 (1986), 171-174 [211.
[AndIl J. B. Anderson and R. de Buda, Better phase-modulation performance
using trellis phase codes, Elect. Lett. 12 (1976),587-588 [31.
[And2] S. Andrilli, Existence and uniqueness of O'Nan's simple group, Ph.D.
Dissertation, Rutgers University, 1979 [101.
[Apol) T. M. Apostol, Introduction to Analytic Number Theory, Springer-
Verlag, 1976 [151.
[Arnl) V. I. Arnold and A. L. Krylov, Uniform distribution of points on a
sphere and some ergodic properties of solutions of linear ordinary
differential equations in a complex region, DAN 148 (1963), 9-12 -
SMD 4 (1963), 1-5 [11.
[Art1) E. Artin, The orders of the classical simple groups, Comm. Pure Appl.
Math. 8 (1955),455-472 = Coil. Papers pp. 398-415 [101.
M. Aschbacher: see R. L. Griess Jr., G. Mason.
[Ashl) A. Ash, Eutacticforms, CJM 29 (1977), 1040-1054.
[Ash2] - On the existence of eutactic forms, BLMS 12 (1980),192-196.
[Ash3) P. F. Ash and E. D. Bolker, Recognizing Dirichlet tessellations, Geom.
Dedic. 19 (1985), 175-206 [21.
T. Ashida: see S. R. Hall.
[AsiIl D. Asimov, The grand tour-a tool for viewing multidimensional data,
SIAM J. Sci. Stat. Comput. 6 (1985), 128-143.
[AskIl R. Askey, Orthogonal polynomials and positivity, Studies in Applied
Math. 6, Special Functions and Wave Propagation, Soc. Indust. Appl.
Math., Phil. PA, 1970, pp. 64-85 [91.
[Ask2) Orthogonal polynomials and special functions, Soc. Indust. Appl.
Math., Phil. PA, 1975 [91.
[Ask2a) - editor, Theory and Application of Special Functions, Ac. Press, NY,
1975 [al.
[Ask3) Orthogonal polynomials old and new, and some combinatorial
connections, in DacIl, 1984, pp. 67-84.
[Ask4) - and J. Wilson, A set of orthogonal polynomials that generalize the
Racah coefficients or 6-j symbols, SIAM J. Math. Anal. 10 (1979),
1008-1016 [91.
[AssIl S. F. Assmann and N. J. A. Sloane, in preparation [91.
[Ass2) E. F. Assmus, Jr., and H. F. Mattson, Jr., Perfect codes and the
Mathieu groups, Archiv. Math. 17 (1966),121-135 [3, Ill.
[Ass3) - - New 5-designs, JCT 6 (1969), 122-151 [3,71.
[Ass4) - - Coding and combinatorics, SIAM Rev. 16 (1974), 349-388 [31.
[Astl) J. T. Astola, The Tietiiviiinen bound for spherical codes, Discr. App1.
Math. 7 (1984),17-21 [1,91.
B. S. Atal: see also J. L. Flanagan, M. R. Schroeder.
[Ata1) - Predictive coding of speech at low bit rates, COMM 30 (1982),
600-614 [21.
Bibliography 575

[Ata2] - and M. R. Schroeder, Predictive coding of speech signals and

subjective error criteria, IEEE Trans. Acoust. Speech, Signal Proc. 27
(1979),247-254 [21.
[Ata3] - - Stochastic coding of speech signals at very low bit rates, in
Links for the Future, ed. P. Dewilde and C. A. May, IEEE Press, NY,
1984, pp. 1610-1613 [2].
A. O. L. Atkin: see also J. Larmouth.
[Atkl] - P. Fong and S. D. Smith, personal communication [301.
[Atk2] M. D. Atkinson, editor, Computational Group Theory, Ac. Press, NY,
1984 [B].
[AusIl L. Auslander, An account of the theory of crystallographic groups,
PAMS 16 (1965), 1230-1236 [41.
[Avil] D. Avis and B. K. Bhattacharya, Algorithms for computing d-
dimensional Voronoi diagrams and their duals, in [Prela], 1983,
pp. 159-180.
B
[BabIl L. Babai, On Lovasz' lattice reduction and the nearest lattice point
problem, Combinatorica 6 (1986), 1-13 [21.
[Bab2] V. F. Babenko, On the optimal error bound for cubature formulas on
certain classes of continuous functions, Analysis Math. 3 (1977), 3-9
[I].
[BacIl G. Bachman, Introduction to p-Adic Numbers and Valuation Theory,
Ac. Press, NY, 1964 [I5].
[Bac2] P. Bachmann, Zahlentheorie IV, Die Arithmetik der Quadratische
Formen 2, Teubner, Leipzig, 1923 [2].
[Bac3] R. Backhouse, Writing a number as a sum of two squares: a new
solution, Info. Proc. Lett. 14 (1982), 15-17 [15].
R. Bajcsy: see R. Mohr.
[Ball] W. W. R. Ball and H. S. M. Coxeter, Mathematical Recreations and
Essays, Univ. Toronto Press, Toronto, 12th ed., 1974 [41.
[Baml] R. P. Bambah, On lattice coverings by spheres, Proc. Nat. Inst. Sci.
India 20 (1954), 25-52 [2, 4].
[Bam2] - Lottice coverings with four-dimensional spheres, PCPS 50 (1954),
203-208 [2, 4].
[Bam3] - and N. J. A. Sloane, On a problem of Ryskov concerning lattice
coverings, Acta Arith. 42 (1982), 107-109 [2,41.
[Ban 11 E. Bannai, On some spherical t-designs, JCT A26 (1979), 157-161 [3].
[Ban2] - Orthogonal polynomials, algebraic combinatorics and spherical t-
designs, PSPM 37 (1980), 465-468 [31.
[Ban3] - On the weight distribution of spherical t-designs, Europ. J. Comb. 1
(1980), 19-26 [3].
[Ban4] - Spherical t-designs which are orbits of finite groups, J. Math. Soc.
Japan 36 (1984), 341-354 [31.
[Ban5] - Spherical designs and group representations, Contemp. Math. 34
(1984),95-107 [3].
[Ban6] - A. Blokhuis, J. J. Seidel and P. Delsarte, An addition formula for
hyperbolic space, JCT A36 (1984), 332-341 [3].
[Ban7] - and R. M. Damerell, Tight spherical designs I, J. Math. Soc. Japan
31 (1979), 199-207 [3, 141.
[Ban8] - - Tight spherical designs II, JLMS 21 (1980), 13-30 [3, 14].
[Ban9] - and S. G. Hoggar, On tight t-designs in compact symmetric spaces of
rank one, Proc. Jap. Acad. A61 (1985),78-82 [31.
576 Bibliography

[BanIO) Tight t-designs and squareJree integers, Europ. J. Combin. to

appear 131.
[Banll) - and T. Ito, Algebraic Combinatorics I: Association Schemes,
Benjamin, Menlo Park CA, 1984 [3, 9, 141.
[BanI2) - - Current research on algebraic combinatorics, Graphs Combin., 2
(1984),287-308 [3,9).
[Ban13) - and N. J. A. Sloane, Uniqueness oj certain spherical codes, CJM 33
(1981),437-449;;;; Chap. 14 of this book [1, 3, 141.
E. P. Baranovskii: see also S. S. Ryskov.
[BarI) - Pac kings, coverings, partitionings and certain other distributions in
spaces oj constant curvature, Itogi Nauki-Ser. Mat. (Algebra,
Topologiya, Geometriya) 1969, 185-225 = Progress in Math. 9 (1971),
209-253 [11.
[Bar2) - and S. S. Ryskov, Primitive five-dimensional parallelohedra, DAN
212 (1973), 532-535 - SMD 14 (1973) 1391-1395.
[Bar3) - - and S. S. Shushbaev, A geometric estimate oj the number oj
representations oj a real number by a positive quadratic Jorm and an
estimate oj the remainder oj the multidimensional zeta-Junction, TMIS
158 (1981) 3-8 - PSIM (1983, No.4) 1-7.
[Bar4) E. S. Barnes, Note on extreme Jorms, CJM 7 (1955), 150-154 [21.
[BarS) - The covering oj space by spheres, CJM 8 (1956),293-304 (2).
[Bar6) - The perJect and extreme senary Jorms, CJM 9 (1957), 235-242 11,
2,8).
[Bar7) The complete enumeration oj extreme senary Jorms, Phil. Trans.
Royal Soc. London A249 (1957), 461-506 [1, 2, 81.
[Bar8) - On a theorem oj Voronoi, PCPS 53 (1957), 537-539 [21.
[Bar9) The construction oj perJect and extreme Jorms, Acta Arith. 5
(1959) 57-79 and 205-222 [2, 61.
[BarIO) - CriteriaJor extreme Jorms, JAMS 1 (1959), 17-20 [21.
[BarIl) - Minkowski's Jundamental inequality Jor reduced positive quadratic
Jorms, JAMS A26 (1978),46-52 [21.
[BarI2) - and M. J. Cohn, On the reduction oj positive quaternary quadratic
Jorms, JAMS A22 (1976), 54-64 (2).
[Bar13) - - On Minkowski reduction oj positive quaternary quadratic Jorms,
Math. 23 (1976), 156-158 [21.
[Barl4) - and T. J. Dickson, Extreme coverings oj n-space by spheres, JAMS 7
(1967), 115-127; 8 (1968),638-640 [21.
[BarIS) - and N. J. A. Sloane, New lattice packings oj spheres, CJM 35
(1983), 117-130 [1, 8).
[BarI6) The optimal lattice quantizer in three dimensions, SIAD 4
(1983), 30-41 [2,211.
[BarI7) - and D. W. Trenerry, A class oj extreme lattice-coverings oj n-space
by spheres, JAMS 14 (1972),247-256 [21.
[Barl8) - and G. E. Wall, Some extreme Jorms defined in terms oj Abelian
groups, JAMS 1 (1959),47-63 [1,4,5,6,7,81.
[BasIl L. A. Bassa1ygo, New upper bounds Jor error-correcting codes, PPI 1
(No.4, 1965),41-44 = PIT 1 (No.4, 1965),32-35 [91.
[Bas2) - and V. A. Zinoviev, Some simple consequences oj coding theory Jor
combinatorial problems oj packings and coverings, PPI 34 (No.2,
1983),291-295 = PIT 34 (1983), 629-631 [91.
[BatI ) P. T. Bateman, On the representations oj a number as the sum oj three
squares, TAMS 71 (1951),70-101 [41.
[BauI) J. R. Baumgardner and P. O. Frederickson, Icosahedral discretization oj
Bibliography 577

the two-sphere, SIAM J. Num. Anal. 22 (1985), 1107-1115 [ll.

[BayIl E. Bayer-Fluckiger, Definite unimodular lattices having an
automorphism of given characteristic polynomial, Comment. Math.
Helvet. S9 (1984), 509-538 [2,4,7,8).
[Bec1) J. Beck, Sums of distances between points on a sphere, Math. 31
(1984),33-41 [1).
[BeeIl G. F. M. Beenker, PGIT 30 (1984),403-405 [71.
[Bell) Bell Telephone Laboratories, Transmission Systems for
Communications, Bell Telephone Labs, Murray Hill NJ, 1964 [31.
L. K. Bell: see C. E. Briant.
[Bel2) R. Bellman, A Brief Introduction to Theta-Functions, Holt, Rinehart
and Winston, NY, 1961 (4).
[BenO) M. Benard, Characters and Schur indices of the unitary reflection group
[321P, PJM S8 (1975),309-321 [41.
[Benl) c. Bender, Bestimmung der grossten Anzahl gleich Kugeln, welche sich
auf eine Kugel von demselben Radius, wie die iibrigen, aUfiegen lassen,
Archiv Math. Physik (Grunert) S6 (1874), 302-306 [ll.
[Ben2) J. W. Benham and J. S. Hsia, Spinor equivalence of quadratic forms,
JNT 17 (1983), 337-342 [151.
[Ben3) C. H. Bennett, Serially depOSited amorphous aggregates of hard
spheres, J. Appl. Phys. 43 (1972),2727-2734 [ll.
[Ben4) G. H. Bennett, Pulse Code Modulation and Digital Transmission,
Marconi Instruments, St. Albans, England, 1978 [31.
C. T. Benson: see L. C. Grove.
[BerIl T. Berger, Rate Distortion Theory, Prentice-Hall, Englewood Cliffs NJ,
1971 [21.
[Ber2) - Optimum quantizers and permutation codes, PGIT 18 (1972), 759-
765 [21.
[Ber3) - F. Jelinek and J. K. Wolf, Permutation codes for sources, PGIT 18
(1972),160-169 (2).
E. R. Berlekamp: see also C. E. Shannon.
[Ber4) Algebraic Coding Theory, McGraw-Hill, NY, 1968 [3, 5).
[Ber5) - Coding theory and the Mathieu groups, IC 18 (1971),40-64 [51.
[Ber6) - editor, Key Papers in the Development of Coding Theory, IEEE
Press, NY, 1974 [31.
[Ber7) - F. J. MacWilliams and N. J. A. Sloane, Gleason's theorem on self-
dual codes, PGIT 18 (1972),409-414 (7).
[Ber8) - R. J. McEliece and H. C. A. van Tilborg, On the inherent
intractability of certain coding problems, PGIT 24 (1978),384-386 [21.
[Ber9) J. D. Berman and K. Hanes, Volumes of polyhedra inscribed in the unit
sphere in E 3, Math. Ann. 188 (1970), 78-84 [ll.
[BerlO) - - Optimizing the arrangement of points on the unit sphere, MTAC
31 (1977), 1006-1008 [ll.
[Berll) J. D. Bernal, The structure of liquids, PRS A280 (1964), 299-322 [ll.
[Besl) M. R. Best, Binary codes with a minimum distance of four, PGIT 26
(1980),738-742 [3, 5, 91.
[Bes2) - and A. E. Brouwer. The triply shortened binary Hamming code is
optimal, Discr. Math. 17 (1977), 235-245 (9).
[Bes3) - - F. J. MacWilliams, A. M. Odlyzko and N. J. A. Sloane, Bounds
for binary codes of length less than 25, PGIT 24 (1978), 81-93 [3,5,91.
[Betl) T. Beth, W. Fumy and H. P. Reiss, Der Wunderschone Oktaden-
Generator, Mitt. Math. Sem. Giessen 163 (1984), 169-179 [111.
[BezIl A. Bezdek, Solid packing of circles in the hyperbolic plane, Studia Sci.
578 Bibliography

Math. Hungar. 14 (1979), 203-207 [11

[Bez2] - Uber /onenpackungen, Studia Sci. Math. Hungar. 18 (1983), 277-
285.
[Bez3] - Circle packings into convex domains of the Euclidean and hyperbolic
plane and sphere. Geom. Dedic. 21 (1986),249-255 [11
B. K. Bhattacharya: see D. Avis.
[BieI] L. Bieberbach, Uber die Bewegungsgruppen der Euklidischen Riiume.
Math. Ann. 70 (1911), 297-336; 72 (1912),400-412 [4].
[Big I] N. L. Biggs, Perfect codes in graphs. JCT BIS (1973), 289-296 [9].
[Big2] - Algebraic Graph Theory. Camb. Univ. Press, 1974 [9].
[Big3] - Designs. factors and codes in graphs. Quart. J. Math. Oxford 26
(1975),113-119 [9].
[Big3a] - A. G. Boshier and J. Shawe-Taylor, Cubic distance-regular graphs.
JLMS 33 (1986), 385-394 [9].
[Big4] - and D. H. Smith, On trivalent graphs. BLMS 3 (1971), 155-158 [9].
[Big5] and A. T. White, Permutation Groups and Combinatorial
Structures. Camb. Univ. Press, 1979 [9].
[Bill ] E. Biglieri and M. Elia, On the existence of group codes for the
Gaussian channel. PGIT 18 (1972), 399-402 [3].
[Bi12] - - Optimum permutation modulation codes and their asymptotic
performance. PGIT 22 (1976),751-753 [3].
[BiB] - - Cyclic-group codes for the Gaussian channel. PGIT 22 (1976),
624-629 [3].
[BinI] N. H. Bingham, Positive definite functions on spheres. PCPS 73 (1973),
145-156 [9].
B. J. Birch: see J. Larmouth.
[Birl] G. Birkhoff and S. MacLane, A Survey of Modern Algebra. Macmillan,
NY, 4th ed., 1977 [2].
[Bla1] N. M. Blachman, The closest packing of equal spheres in a larger
sphere. AMM 70 (1963), 526-529 [I].
[Bla2] R. E. Blahut, Theory and Practice of Error Control Codes. Addison-
Wesley, Reading MA, 1983 [3].
I. F. Blake: see also I. B. Gibson.
[Bla3] - The Leech lattice as a code for the Gaussian channel. IC 19 (1971),
66-74 [3].
[Bla4] - Distance properties of group codes for the Gaussian channel. SIAJ
23 (1972), 312-324 [3].
[Bla5] - editor, Algebraic Coding Theory History and Development. Dowden,
Stroudsburg PA, 1973 [3].
[Bla6] - Configuration matrices of group codes. PGIT 20 (1974),95-100 [3].
[Bla7] - Properties of generalized Pless codes. in Proc. 12th Allerton Con!
Circ. Syst. Theory. Univ. Ill., Urbana, 1974, pp. 787-789 [3].
[Bla8] - On a generalization of the Pless symmetry codes. IC 27 (1975),
369-373 [3].
[Bla9] - and R. C. Mullin, The Mathematical Theory of Coding. Ac. Press,
NY, 1975 [3].
I. Blech: see D. Shechtman.
[BleI] M. N. Bleicher, Lattice coverings of n-space by spheres. CJM 14
(1962), 632-650 [21.
H. F. Blichfeldt: see also G. A. Miller.
[BliI] - Finite Collineation Groups. Univ. Chicago Press, 1917 [7].
[Bli2] On the minimum value of positive real quadratic forms in 6
variables. BAMS 31 (1925), 386 [11
Bibliography 579

[Bli3) - The minimum value of quadratic forms, and the closest packing of
spheres, Math. Ann. 101 (1929),605-608 (1).
[Bli4) - The minimum values of positive quadratic forms in six, seven and
eight variables, Math. Zeit. 39 (1935), 1-15 [1,4,6).
[Bli5) F. van der Blij, History of the octaves, Simon Stevin 34 (1961), 106-125
[4).
[Bli6) - and T. A. Springer, The arithmetics of octaves and of the group G2,
KNAW A62 (1959),406-418 [4).
A. Blokhuis: see E. Bannai.
[Bocl) S. Bochner, Hilbert distances and positive definite functions, Ann.
Math. 42 (1941),647-656 [9).
[Boel) A. H. Boerdijk, Some remarks concerning close-packing of equal
spheres, Philips Res. Rep. 7 (1952), 303-313 (1).
E. D. Bolker: see P. F. Ash.
B. BolloMs: see J. J. Seidel.
[Booll W. M. Boothby and G. L. Weiss, editors, Symmetric Spaces, Dekker,
NY, 1972 [9, B).
[Borl) R. E. Borcherds, The Leech lattice and other lattices, Ph.D.
Dissertation, Univ. of Cambridge, 1984 [2, 171.
[Bor2) The Leech lattice, PRS A398 (1985),365-376 [24).
[Bor3) - The 24-dimensional odd unimodular lattices, Chap. 17 of this book
[2, 17).
[Bor4) - J. H. Conway, L. Queen and N. J. A. Sloane, A Monster Lie
algebra? Adv. in Math. S3 (1984),75-79;;; Chap. 30 of this book [30).
[Bor5) Z. I. Borevich and I. R. Shafarevich, Number theory, Nauka, Moscow,
1964; English translation, Ac. Press, NY, 1966 (15).
[Bor6) D. Borwein, J. M. Borwein and K. F. Taylor, Convergence of lattice
sums and Madelung's constant, J. Math. Phys. 26 (1985), 2999-3009
[2).
J. M. Borwein: see D. Borwein.
[BarIl K. Boriiczky, Packing of spheres in spaces of constant curvature (in
Hungarian), Mat. Lapok 2S (1974),265-306; 26 (1975),67-90 [I).
[Bor2) Packing of spheres in spaces of constant curvature, AMAH 32
(1978),243-261 [I, 13).
[Bor3) - The problem of Tammes for n = 11, Studia Sci. Math. Hung. 18
(1983), 165-171 [11.
[Bor4) Closest packing and loosest covering, in Diskrete Geometrie, 3rd
Kol/oq., Inst. f. Math., Univ. Salzburg, 1985, pp. 329-334.
[Bosi) A. Bos, Sphere-packings in Euclidean space, in [Sch8), 1979, pp. 161-
177.
[Bos2) - Upper bounds for sphere packings in Euclidean space, preprint [I).
[Bos3) - J. H. Conway and N. J. A. Sloane, Further lattice packings in high
dimensions, Math. 29 (1982), 171-180 [I, 8).
A. G. Boshier: see N. L. Biggs.
[Botl) F. V. Botya, Bravais types of five-dimensional lattices, TMIS 163
(1984) = PSIM 163 (No.4, 1985),23-27 [1,3).
[BouO) P. Bougerol, Un Mini-Cours sur les Couples de Guelfand, Pub. du
Laboratoire de Statistique et Probabilites, Univ. Paul Sabatier, Toulouse,
1983 [9).
[Boul) N. Bourbaki, Groupes et Algeores de Lie, Chapitres 4, 5 et 6, Hermann,
Paris, 1968 [4, 18,21,23,24,25,27).
[Bou2) M. Bourdeau and A. Pitre, Tables of good lattices in four and five
dimensions, Num. Math. 47 (1985), 39-43 [11.
580 Bibliography

[Bowl) V. J. Bowman, Permutation polyhedra, SIAJ 22 (1972), 580-589 [211.

[Bow2) A. Bowyer, Computing Dirichlet tessellations, Compo J. 24 (1981),
162-166 [2).
H. Brandstriim: see L. H. Zetterberg.
[Brall H. Brandt and O. Intrau, Tabellen reduzierter positiver terniirer
quadratischen Formen, Abh. Sachs. Akad. Wiss. Math.-Nat. Kl. 4S
(No.4, 1958),261 pp. [15).
[Bra2) R. Brauer and C.-H. Sah, editors, Theory of Finite Groups, Benjamin,
NY, 1969 [B).
[Bra3) H. Braun, Zur Theorie der hermitischen Formen, ASUH 14 (1941),
61-150 [16).
[Bril) C. E. Briant, B. R. C. Theobald, J. W. White, L. K. Bell,
D. M. P. Mingos and A. J. Welch, Synthesis and X-ray structural
characterization of the centered icosahedral gold cluster compound
[AuI3 (PMe2Ph) IO Ch)(PF6) 3; the realization of a theoretical prediction,
J. Chern. Soc., Chern. Comm. (1981), 201-202 [11.
[Bri2) E. Brieskorn, Milnor lattices and Dynkin diagrams, PSPM 40 (1983),
153-165 [2).
J. Brillhart: see M. A. Morrison.
[BroO) P. L. H. Brooke, On matrix representations and codes associated with
the simple group of order 25920, J. Aig. 91 (1984), 536-566 (3).
[BroIl Tables of Codes Associated with Certain Finite Simple Groups,
Ph.D. Dissertation, Univ. of Cambridge, June 1984 (3).
[Br02) On the Steiner system S(2,4,28) and codes associated with the
simple group of order 6048, J. Aig. 97 (1985), 376-406 (3).
[Br03) W. Brostow, J.-P. Dussault and B. L. Fox, Construction of Voronoi
polyhedra, J. Compo Phys. 29 (1978), 81-92 [2).
[Br04) M. Broue, 1£ reseau de Leech et Ie groupe de Conway, These de 3e
cycle, Paris, 1970 [4).
[Br05) - Codes correcteurs d'erreurs auto-orthogonaux sur Ie corps d deux
e1ements et formes quadratique entieres deJinies positives d discriminant
+1, Discr. Math. 17 (1977), 247-269 [7, 16).
[Br06) - and M. Enguehard, Polynomes des poids de certains codes et
fonctions theta de certains reseaux, Ann. Sci. Ecole Norm. Sup. S
(1972), 157-181 [7, 16).
[Br07) Une famille infinie de formes quadratiques entieres; leurs
groupes d'automorphismes, Ann. Sci. Ecole Norm. Sup. 6 (1973), 17-
52 [7).
A. E. Brouwer: see also M. R. Best.
[Br08) - A few new constant weight codes, PG IT 26 (1980), 366 [9).
[Br09) - P. Delsarte and P. Piret, On the (23,14,5) Wagner code, PGIT 26
(1980), 742-743 [9).
[BrolO) H. Brown, R. Biilow, J. Neubiiser, H. Wondratschek and H. Zassenhaus,
Crystallographic Groups of Four-Dimensional Space, Wiley, NY, 1978
(3).
[BroIl) J. Neubiiser and H. J. Zassenhaus, On integral groups I: the
reducible case, Num. Math. 19 (1972), 386-399 (3).
[Brol2) - - - On integral groups II: the irreducible case, Num. Math. 20
(1972),22-31 (3).
[Brol3) - - - On integral groups III: normalizers, MTAC 27 (1973) 167-
182 (3).
[8ro14) W. S. Brown, ALTRAN User's Manual, Bell Laboratories, Murray Hill,
4th ed., 1977 [16).
Bibliography 581

[Brul) V. Brun, On regular packing of equal circles touching each other on the
surface of a sphere, Comm. Pure Appl. Math. 29 (I976), 583·590 [I).
J. A. Bucklew: see also N. C. Gallagher, Jr.
[Buell Compounding and random quantization in several dimensions,
PGIT 27 (I981), 207·211 (2).
[Bue2) - Upper bounds to the asymptotic performance of block quantizers,
PGIT 27 (I981), 577·581 (2).
[Bue3) - and G. L. Wise, Multidimensional asymptotic quantization theory
with rth power distortion measures, PGIT 28 (I982), 239·247 (2).
[Budll P. de Buda, Encoding and decoding algorithms for an optimal lattice·
based code, in Conference Record·lnternat. Con! Commun. ICC '81,
IEEE Press, NY, Vol. 3, 1981, pp. 65.3.1·65.3.5 [2, 20).
R. de Buda: see also J. B. Anderson.
[Bud2) The upper error bound of a new near·optimal code, PGIT 21
(I975), 441·445 (3).
[Bud3) - and W. Kassem, About lattices and the random coding theorem, in
Abstracts of Papers, IEEE Inter. Symp. Info. Theory 1981, IEEE Press,
NY 1981, p. 145 (3).
R. BUlow: see also H. Brown.
[Biill) - J. Neubiiser and H. Wondratschek, On crystallography in higher
dimensions, Acta Cryst. A27 (I971), 517·535 (3).
[Buml) C. Bumiller, On rank three graphs with a large eigenvalue, Diser. Math.
23 (I978), 183·187 (9).
H. Burzlaff: see also H. Zimmerman.
[Burl) - and H. Zimmerman, On the metrical properties of lattices, Z. Krist.
170 (I985), 247·262 (2).
[Bur2) - - and P. M. de Wolff, Crystal lattices, in [Hahl), 1983, 734·744
(2).
[Busll P. Buser, A geometric proof of Bieberbach's theorems on
crystallographic groups, L'Enseignement Math. 31 (1985), 137·145 (4).
[ButI) G. Butler, The maximal subgroups of the sporadic simple group of
Held, J. Alg. 69 (1981), 67·81 (10).
A. Buzo: see Y. Linde.
c
J. W. Cahn: see D. Shechtman.
[Call) A. R. Calderbank and W. M. Kantor, The geometry of two·weight
codes, BLMS 18 (1986),97·122 (9).
[Ca12) - T.·A. Lee and J. E. Mazo, Baseband trellis codes with a spectral
null at zero, PGIT to appear (3).
[Ca13) - and J. E. Mazo, A new description of trellis codes, PGIT 30 (1984),
784· 792 (3).
[Ca14) - - and H. M. Shapiro, Upper bounds on the minimum distance of
trellis codes, BSTJ 62 (I983), 2617·2646 (3).
[Ca15) - - and V. K. Wei, Asymptotic upper bounds on the minimum
distance of trellis codes, COMM 33 (I985), 305·309 (3).
[Ca16) - and N. J. A. Sloane, Four·dimensional modulation with an eight·
state trellis code, BSTJ 64 (1985), 1005·1018 (3).
[Ca17) - - An eight·dimensional trellis code, PIEEE 74 (1986), 757·759 (3).
[Ca18) - - New trellis codes based on lattices and cosets, PGIT 33 (1987),
177·195 (3).
[Ca19) - and D. B. Wales, A global code invariant under the Higman·Sims
group, J. Alg. 75 (1982), 233·260 (3).
582 Bibliography

P. J. Cameron: see also N. J. A. Sloane.

[Camll - Another characterization of the small Janko group, J. Math. Soc.
Japan 25 (1973), 591-595 [101.
[Cam2] - and J. H. van Lint, Graphs, Codes and Designs, Camb. Univ. Press,
1980 [3, 9].
[Cam3] - J. A. Thas and S. E. Payne, Polarities of generalized hexagons and
perfect codes, Geom. Dedic. 5 (1976), 525-528 [9].
C. M. Campbell: see R. A. Wilson.
[Cam4] C. N. Campopiano and B. G. Glazer, A coherent digital amplitude and
phase modulation scheme, COMM 10 (1962),90-95 [31.
[Carl] R. D. Carmichael, Introduction to the Theory of Groups of Finite
Order, Ginn, Boston, 1937; Dover, NY, 1956 [ill.
[Car2] R. W. Carter, Simple Groups of Lie Type, Wiley, NY, 1972 [101.
[Car3] - Finite Groups of Lie Type, Wiley, NY, 1985 [101.
[Casl] J. W. S. Cassels, Uber die Aquivalenz 2-adischer quadratischer Formen,
Comment. Math. Helvet. 37 (1962),61-64 [151.
[Cas2] - An Introduction to the Geometry of Numbers, Springer-Verlag, 1971
[1,2,6].
[Cas3] - Rational Quadratic Forms, Ac. Press, NY, 1978 [2, 3, 4, 7, 15, 16].
[Cas4] - Rational quadratic forms, in Proc. Intern. Math. Con!, ed.
L. H. Y. Chen et aI., North-Holland, Amsterdam, 1982, pp. 9-26 [l51.
[Cas5] - and A. Frohlich, editors, Algebraic Number Theory, Ac. Press, NY,
1967 [8, B].
[Chal] J. H. H. Chalk, Algebraic lattices, in [Gru2], 1983, pp. 97-110 [21.
[Cha2] R. C. Chang and R. C. T. Lee, On the average length of Delaunay
triangulations, BIT 24 (1984),269-273 [21.
[Cha3] G. Chapline, Unification of gravity and elementary particle interactions
in 26 dimensions, Phys. Lett. BI58 (1985),393-396 [11.
[Cha4] T. W. Chaundy, The arithmetic minima of positive quadratic forms,
Quart. J. Math. 17 (1946), 166-192, but see also the review:
H. S. M. Coxeter, Math. Rev. 8 (1947), 137-138 [61.
[Chell C. L. Chen, Computer results on the minimum distance of some binary
cyclic codes, PGIT 16 (1970),359-360 [3, 51.
L. H. Y. Chen: see J. W. S. Cassels.
[Che2] D.-Y. Cheng, A. Gersho, B. Ramamurthi and Y. Shoham, Fast search
algorithms for vector quantization and pattern matching, in Proc. Inter.
Conf. Acoust. Speech Sig. Proc. ICASSP '84, IEEE Press, NY, 1984,
Vol. 1, pp. 9.11.1-9.11.4 [201.
[Che3] Y. Cheng and N. J. A. Sloane, Codes from symmetry groups, in
preparation [81.
[Che4] A. G. Chernyakov, An example of a 32-dimensional even unimodular
lattice, LOMI86 (1979), 170-179 - JSM 17 (1979), 2068-2075 [71.
[Choll C. Choi, On subgroups ofM 24, TAMS 167 (1972), 1-27 and 29-47 [10].
[Clall B. W. Clare and D. L. Kepert, The closest packing of equal circles on a
sphere, PRS A405 (1986), 329-344 [11.
[Cohl] A. M. Cohen, Finite complex reflection groups, Ann. Sci. Ecole Norm.
Sup. 9 (1976), 379-436 [3,41.
[Coh2] - Finite quaternionic reflection groups, J. Alg. 64 (1980),293-324
[3, 81.
[Coh3] G. D. Cohen, M. R. Karpovsky, H. F. Mattson, Jr., and J. R. Schatz,
Covering radius-survey and recent results, PGIT 31 (1985), 328-
343 £31.
[Coh4] - A. C. Lobstein and N. J. A. Sloane, Further results on the covering
Bibliography 583

radius of codes, PGIT 32 (1986) 680-694 [31.

[Coh4a] J. Cohen and T. Hickey, Two algorithms for determining volumes of
convex polyhedra, J. Assoc. Comput. Mach. 26 (1979),401-414 [21.
[Coh5] H. Cohn, A Second Course in Number Theory, Wiley, NY, 1962 [151.
[Coh6] - A Classical Invitation to Algebraic Numbers and Fields, Springer-
Verlag, 1978 [8, BI.
M. J. Cohn: see E. S. Barnes.
S. Cohn-Vossen: see D. Hilbert.
[Coil] R. R. Coifman and G. Weiss, Representations of compact groups and
spherical harmonics, L'Enseignement Math. 14 (1968), 121-173 [91.
[Coi2] - - Analyse harmonique non-commutative sur certains espaces
homogenes, LNM 242, 1971 [9].
[Coil] M. J. Colbourn, Some new upper bounds for constant weight codes,
PGIT 26 (1980),478 [9].
[CoI2] M. J. Collins, editor, Finite Simple Groups II, Ac. Press, NY, 1980 [B].
[Conll R. Connelly, The rigidity of circle and sphere packings, J. Structural
Topology 14 (May 1987) and 16 (May, 1988), to appear.
J. H. Conway: see also R. E. Borcherds, A. Bos, R. A. Parker.
[Con2] - A perfect group of order 8,315,553,613,086,720,000 and the
sporadic simple groups, PNAS 61 (1968), 398-400 131.
[Con3] - A group of order 8,315,553,613,086,720,000, BLMS 1 (1969), 79-
88 [3, 101.
[Con4] - A characterisation of Leech's lattice, Invent. Math. 7 (1969), 137-
142 == Chap. 12 of this book [7, 12].
[Con5] - Three lectures on exceptional groups, in [Powl], 1971, pp. 215-247
.. Chap. 10 of this book [101.
[Con6] - Groups, lattices, and quadratic forms, in Computers in Algebra and
Number Theory, SIAM-AMS Proc. IV, Amer. Math. Soc., Providence
RI, 1971, pp. 135-139.
[Con7] - A construction for the smallest Fischer group, in [Gag2], 1973,
pp. 27-35 [101.
[Con8] Invariants for quadratic forms, JNT 5 (1973), 390-404 [15].
[Con9] - The miracle octad generator, in [CurO], 1977, pp. 62-68.
[Con 10] - A quaternionic construction for the Rudvalis group, in [CurO], 1977,
pp. 69-81 [101.
[Conlll - Monsters and moonshine, Math. Intelligencer 2 (No.4, 1980), 165-
172 Ill.
[ConI2] - The hunting of J 4, Eureka 41 (1981),46-54 [10, Ill.
[Con13] - The automorphism group of the 26-dimensional even unimodular
Lorentzian lattice, J. Alg. 80 (1983), 159-163 .. Chap. 27 of this book
[15,271.
[ConI4] - Hexacode and tetracode - MOG and MINIMOG, in [Atk2], 1984,
pp. 359-365.
[ConI5] - A simple construction for the Fischer-Griess Monster group, Invent.
Math. 79 (1985),513-540 [291.
[ConI6] - R. T. Curtis, S. P. Norton, R. A. Parker and R. A. Wilson, ATLAS
of Finite Groups, Oxford Univ. Press, 1985 [3,4,8, 10,291.
[ConI6a] - and K. M. Knowles, Quasiperiodic tiling in two and three
dimensions, J. Phys. A 19 (1986),3645-3653 [Il.
[ConI7] - and S. P. Norton, Monstrous moonshine, BLMS 11 (1979), 308-339
[1, 10, 25, 29, 30].
[ConI8] - - and L. H. Soicher, The Bimonster, the group Y m , and the
projective plane of order 3, Proceedings of "Computers in Algebra"
584 Bibliography

Conference, Chicago, 1985, to appear [291.

[Conl9] - A. M. Odlyzko and N. J. A. Sloane, Extremal self-dual lattices
exist only in dimensions 1 to 8, 12, 14, 15,23 and 24, Math. 25 (1978),
36-43 ;;;; Chap. 19 of this book [7, 19].
[Con20] - R. A. Parker and N. J. A. Sloane, The covering radius of the Leech
lattice, PRS A380 (1982),261-290;;;; Chap. 23 of this book [2, 6, 231.
[Con211 - and V. Pless, On the enumeration of self-dual codes, JCT A28
(1980),26-53 [4,7].
[Con22] - - On primes dividing the group order of a doubly-even (72, 36, 16)
code. and the group order of a quaternary (24, 12, 10) code, Discr.
Math. 38 (1982), 143-156 [71.
[Con23] - - Monomials of order 7 and 11 cannot be in the group of a
(24, 12 10) self-dual quaternary code. PGIT 29 (1983), 137-140 [71.
[Con24] - - and N. J. A. Sloane, Self-dual codes over GFO) and GF(4) of
length not exceeding 16, PGIT 25 (1979),312-322 [4, 5, 71.
[Con24a] - and A. D. Pritchard, Hyperbolic reflections for the Bimonster and
3Fi 24 , J. Alg. to appear [29].
[Con25] - and L. Queen, Computing the character table of a Lie group, in
[McK3], 1982, pp. 51-87 [211.
[Con26] - and N. J. A. Sloane, Fast 4- and 8-dimensional quantizers and
decoders. in Nat. Telecomm. Con! Record-1981, IEEE Press, NY,
1981, Vol. 3, pp. F4.2.1 to F4.2.4 [201.
[Con27] - - On the enumeration of lattices of determinant one, JNT IS
(1982),83-94 [2, 4, 7, 16].
[Con28] - - Voronoi regions of lattices. second moments of polytopes. and
quantization, PGIT 28 (1982), 211-226 == Chap. 21 of this book [2, 4, 6,
211.
[Con29] - - Fast quantizing and decoding algorithms for lattice quantizers
and codes, PGIT 28 (1982), 227-232 [2,201.
[Con30] - - Twenty-three constructions for the Leech lattice, PRS A381
(1982),275-283 ;;;; Chap. 24 of this book [241.
[Con311 - - Lorentzianformsfor the Leech lattice, BAMS 6 (1982), 215-217
== Chap. 26 of this book [26].
[Con32] - - Laminated lattices, Ann. Math. 116 (1982), 593-620 ;;;; Chap. 6
of this book [6].
[Con33] - - Leech roots and Vinberg groups, PRS A384 (1982), 233-258 ;;;;
Chap. 28 of this book [15, 281.
[Con34] - - The unimodular lattices of dimension up to 23 and the
Minkowski-Siegel mass constants, Europ. J. Combinatorics 3 (1982),
219-231 ;;;; Chap. 16 of this book [2, 7, 15, 161.
[Con35] - - A fast encoding method for lattice codes and quantizers, PGIT 29
(1983),820-824 [2, 201.
[Con36] - - Complex and integral laminated lattices, TAMS 280 (1983),
463-490 [2, 6, 221.
[Con37] - - The Coxeter-Todd lattice. the Mitchell group. and related sphere
packings, PCPS 93 (1983), 421-440 [2,4,7,8,221.
[Con38] - - On the Voronoi regions of certain lattices, SIAD 5 (I984), 294-
305 [2, 4, 20, 211.
[Con39] - - A lower bound on the average error of vector quantizers, PGIT
31 (1985), 106-109 [2].
[Con40] - - Soft decoding techniques for codes and lattices. including the
Golay code and the Leech lattice, PGIT 32 (1986), 41-50 [2, 4, 20).
[Con411 - - Lexicographic codes: error-correcting codes from game theory,
Bibliography 585

PGIT 32 (1986), 337-348 [6,9, Ill.

[Con42J - Low-dimensional lattices I, II, III, PRS to appear [I, 2, 31.
[Con43J - - Covering space with spheres, in preparation [2, 22J.
[Con44J - and D. B. Wales, Construction of the Rudvalis group of order
145,926,144,000, J. Alg. 27 (1973), 538-548 [101.
[Con45J - - Matrix generators for h J. Alg. 29 (1974),474-476 [101.
[Coo 11 G. R. Cooper and R. W. Nettleton, A spread-spectrum technique for
high-capacity mobile communications, IEEE Trans. Veh. Tech. 27
(1978),264-275 [ll.
D. J. Costello, Jr.: see Shu Lin.
[Cos1l P. J. Costello & J. S. Hsia, Even unimodular 12-dimensional quadratic
forms over Q(.J5) , Adv. in Math. 64 (1987), 241-278 [2,8,161.
H. S. M. Coxeter: see also W. W. R. Ball, T. W. Chaundy. A partial
list of H. S. M. Coxeter's publications is given on pp. 5-13 of [Dav3].
[CoxlJ The pure Archimedean polytopes in six and seven dimensions,
PCPS 24 (1928), 1-9 [4, 211.
[Cox2J - Groups whose fundamental regions are simplexes, JLMS 6 (1931),
132-136 [41.
[Cox3J - Discrete groups generated by reflections, Ann. Math. 35 (1934),
588-621 [4].
[Cox4J - The come,lete enumeration of finite groups of the form
R? = (R;R) ij = I, JLMS 10 (1935), 21-25 [4J.
[Cox5J - Wythojj's construction for uniform polytopes, PLMS 38 (1935),
327-339 = Chap. 3 of [Cox18J [21].
[Cox6J - Regular and semi-regular polytopes, Math. Z. 46 (1940) 380-407
[2ll.
[Cox7J - The polytope 221> whose twenty-seven vertices correspond to the lines
on the general cubic surface, AJM 62 (1940),457-486 [4, 2ll.
[Cox8J - Quaternions and reflections, AMM 53 (1946), 136-146 [2,41.
[Cox9J - Integral Cayley numbers, DMJ 13 (1946), 561-578 - Chap. 2 of
[Cox18J [41.
[CoxIOJ - Extreme forms, CJM 3 (1951), 391-441 [2,4,8,21,261.
[CoxlIJ - The product of the generators of a finite group generated by
reflections, DMJ 18 (1951),765-782.
[Cox12J - Twelve points in PG (5, 3) with 95040 self-transformations, PRS
A247 (1958),279-293 = Chap. 7 of [CoxI8J.
[Cox13J - Polytopes over GF(2) and their relevance for the cubic surface
group, CJM 11 (1959),646-650 [4J.
[Cox14J - Introduction to Geometry, Wiley, NY, 1961 [1,211.
[Cox15J - The problem of packing a number of equal nonoverlapping circles on
a sphere, Trans. N.Y. Acad. Sci. 24 (No.3, 1962),320-331 [ll.
[Cox16J - An upper bound for the number of equal nonoverlapping spheres
that can touch another of the same size, PSPM 7 (1963), 53-71 =
Chap. 9 of [Cox18J [I, 131.
[Cox17J - Geometry, in Lectures on Modern Mathematics, ed. T. L. Saaty,
Wiley, NY, Vol. III, 1965, pp. 58-94 = Chap. 12 of [Cox18J [2, 41.
[Cox18J - Twelve Geometric Essays, Southern Illinois Press, Carbondale IL,
1968 [1,2,4,21, BJ.
[Cox19J - Finite groups generated by unitary reflections, ASUH 31 (1967),
125-135.
[Cox20J - Regular Polytopes, Dover, NY, 3rd ed., 1973 [3,4,5,6,21,23]'
[Cox21J Regular Complex Polytopes, Camb. Univ. Press, 1974 [I, 2, 31.
[Cox2laJ - Non-Euclidean Geometry, Univ. Toronto Press, 5th ed., 1978 [261.
586 Bibliography

[Cox22] - Polytopes in the Netherlands, Nieuw Arch. Wisk. 26 (1978), 116-

141 [21].
[Cox23] - Regular and semi-regular polytopes II, Math. Z. 188 (1985), 559-
591 [21].
[Cox24] - A packing of 840 balls of radius 9 0' 19" on the 3 -sphere, preprint
0

[ I].
[Cox25] - Regular and semi-regular polytopes III, Math. Z., to appear [21l.
[Cox26] - L. Few and C. A. Rogers, Covering space with equal spheres, Math.
6 (1959), 147-157 [2].
[Cox27] - M. S. Longuet-Higgins and J. C. P. Miller, Uniform polyhedra, Phil
Trans. Royal Soc. 246 (1954),401-450 [21].
[Cox28] - and W. o. J. Moser, Generators and Relations fQr Discrete Groups,
Springer-Verlag, 4th ed., 1980 [3, 4, 7, 9, 10,21,27]'
[Cox29] - and J. A. Todd, An extreme duodenary form, CJM 5 (1953), 384-
392 [1, 4, 6].
[Cox30] - and G. J. Whitrow, World structure and non-Euclidean honeycombs,
PRS A201 (1950),417-437 [27].
[Cral] A. P. Cracknell, Irreducible representations of point groups and space
groups -the last 50 years and the next 10 years, Cornrn. Match in
Math. Chern. 9 (1980),227-241 [3].
[Cra2] M. Craig, A characterization of certain extreme forms, 11M 20 (1976),
706-717 [2, 3].
[Cra3] - Extreme forms and cyclotomy, Math. 25 (1978),44-56 [2,8].
[Cra4] - A cyclotomic construction for Leech's lattice, Math. 25 (1978),
236-241 [2, 8].
[Cra5] - Automorphisms of prime cyclotomic lattices, preprint [2, 8].
R. E. Crochiere: see J. L. Flanagan.
[Cunl] H. M. Cundy and A. P. Rollett, Mathematical Models, Oxford Univ.
Press, 2nd ed., 1961 [2, 21].
[CurO] M. P. J. Curran, Topics in Group Theory and Computation, Ac. Press,
NY, 1977 [B].
[Curl] C. W. Curtis, W. M. Kantor and G. M. Seitz, The 2-transitive
permutation representations of the finite Chevalley groups, TAMS 218
(1976),1-57 [9].
R. T. Curtis: see also J. H. Conway.
[Cur2] - On subgroups of·O, I: lattice stabilizers, J. Alg. 27 (1973), 549-573
[10, 11, 23].
[Cur3] - A new combinatorial approach to M 24 , PCPS 79 (1976), 25-42 [10,
11,14]'
[Cur4] - The maximal subgroups ofM 24 , PCPS 81 (1977), 185-192 [10].
[Cur5] - On subgroups of ·0, II: local structure, J. Alg. 63 (1980),413-434
[ 10].
[Cur6] - Eight octads suffice, JCT A36 (1984), 116-123 [11].
[Cur7] - The Steiner system S (5,6, 12), the Mathieu group M12 and the
"kitten", in [Atk2], 1984, pp. 352-358 [11].
[Cusl] E. L. Cusack, Error control codes for QAM signalling, Elect. Lett. 20
(1984),62-63.

[Dadl] E. C. Dade, The maximal finite groups of 4 x 4 integral matrices, 11M

9 (1965),99-122 [3].
R. M. Darnerell: see E. Bannai.
Bibliography 587

[DanI1 L. Danzer, Finite point-sets on S2 with minimum distance as large as

possible, Discr. Math. 60 (1986), 3-66 [11.
[DavI1 H. Davenport, The Higher Arithmetic, Hutchinson, London, 1952 [15].
[Dav2] - The covering of space by spheres, Rend. Circ. Mat. Palermo, 1
(1952),92-107 - Coli. Works II (1977), 609-624 [2, 211.
[Dav2a] and G. L. Watson, The minimal points of a positive definite
quadratic form, Math. 1 (1954), 14-17 = Coli. Works II (1977), 684-
688 [15].
[Dav3] e. Davis, B. Griinbaum and F. A. Sherk, editors, The Geometric Vein:
The Coxeter Festschrift, Springer-Verlag 1981 [B).
[Dav4] L. D. Davisson and R. M. Gray, editors, Data Compression, Dowden,
Stroudsberg PA, 1976 [2).
[DawI1 E. Dawson, Self-dual ternary codes and Hadamard matrices, Ars
Comb. A19 (1985), 303-308 [7].
M. Dekesel: see P. A. Devijver.
B. N. Delaunay: see B. N. Delone.
[DegI1 P. Deligne, Formes modulaires et representations I-adiques, Sem.
Bourbaki 21 (No. 335, 1969), 139-172 [2).
[Deg2] - La conjecture de Weil, Inst. Hautes Etudes Sci. Pub!. Math. 53
(1974),273-307 [2].
[DeIO] B. N. Delone, Neue Darstellung der geometrischen Kristallographie,
Z. Krist 84 (1933), 109-149 [2).
[Dell] - Sur la sphere vide, Bull. Acad. Sci. USSR, Classe Sci. Mat. Nat.
(1934), 793-800 [2).
[De12] - N. P. Dolbilin, S. S. Ryskov and M. I. Stogrin, A new construction in
the theory of lattice coverings of an n-dimensional space by equal
spheres, IANS 34 (1970),289-298 - Izv. 4 (1970),293-302 [2].
[Del3] - R. V. Galiulin and M. I. Stogrin, The types of Bravais lattices, in
Current Problems in Math. Vo!. 2, Moscow 1973, pp. 119-257. Math.
Rev. 54 #1068 [11.
[De14] - and S. S. Ryskov, Solution of the problem of least dense lattice
covering of a four-dimensional space by equal spheres, DAN 152
(1963),523-524 - SMD 4 (1963), 1333-1334 [2,4].
[Del5] - - Extremal problems in the theory of positive quadratic forms,
TMIS 112 (1971) 203-229 - PSIM 112 (1971), 21I-231.
[De16] - N. N. Sandakova and S. S. Ryskov, An optimal cubature lattice for
bilaterally smooth functions of two variables, DAN 162 (1965), 1230-
1233 - SMD 6 (1965), 836·839 [11.
[De17] - and M. I. Stogrin, Simplified proof of the Schonjiies theorem, DAN
219 (1974),95-98 - SOy. Phys. Dok!. 19 (1975),727-729 [4].
P. Delsarte: see also E. Bannai, A. E. Brouwer.
[De18] - Bounds for unrestricted codes, by linear programming, Philips Res.
Reports 27 (1972), 272·289 [9].
[De19] - An algebraic approach to the association schemes of coding theory,
Philips Res. Reports Supplements, No. 10 (1973) [9, 14).
[Dell 0] - Four fundamental parameters of a code and their combinatorial
significance, IC 23 (1973),407-438 [9).
[DellI] - Hahn polynomials, discrete harmonics, and t-designs, SIAJ 34
(1978), 157·166 [3, 9).
[Dell 2] - Bilinear forms over a finite field, with applications to coding theory,
JCT A25 (1978), 226·241 [9].
[Del13] - and J.·M. Goethals, Unrestricted codes with the Golay parameters
are unique, Discr. Math 12 (1975), 21I·224 [7, 14].
588 Bibliography

[De1l4] Alternating bilinear forms over GF(q), JCT AI9 (1975), 26-50
[9].
[Dell 5] and J. J. Seidel, Bounds for systems of lines, and Jacobi
polynomials, Philips Res. Reports 30 (1975),91*-105* [3,9].
[De116] - - - Spherical codes and designs, Geom. Dedic. 6 (1977), 363-
388 [I, 3,9,14]'
[Deml] A. P. Dempster, The minimum of a definite ternary quadratic form,
CJM 9 (1957), 232-234 [I].
[Denl] R. H. F. Denniston, Some new 5-designs, BLMS 8 (1976),263-267 [3].
N. Deo: see E. M. Reingold.
[Devil P. A. Devijver and M. Dekesel, Insert and delete algorithms for
maintaining dynamic Delaunay triangulations, Patt. Recog. Lett. 1
(1982),73-77 [2].
[Dev2] - - Computing multidimensional Delaunay tessellations, Patt.
Recog. Lett. I (1983), 311-316 [2].
P. Dewilde: see B. S. Atal, S. G. Wilson.
[Dia I] P. Diaconis, Lectures on the Use of Group Representations in
Probability and Statistics, Inst. Math. Statist. Lecture Note Series, to
appear [9].
[Dia2] - R. L. Graham and W. M. Kantor, The mathematics of perfect
shuffles, Adv. Appl. Math. 4 (1983),175-196 [II].
[Dia3] W. J. Diamond, Practical Experiment Designs, Wadsworth, Belmont
CA, 1981 [3].
J. C. Diaz: see P. Keast.
L. E. Dickson: see also G. A. Miller.
[Dicl] - Linear Groups with an Exposition of the Galois Field Theory,
Dover, NY, 1958 [4, 10].
[Dic2] - History of the Theory of Numbers, Carnegie Institution of
Washington Publications, No. 256, 1919, 1920, 1923, 3 vols; Chelsea,
NY, 1966 [4, 15].
[Dic3] - Studies in the Theory of Numbers, Univ. of Chicago Press, 1930;
Chelsea, NY (no date) [15].
T. J. Dickson: see also E. S. Barnes.
[Dic4] - An extreme covering of 4-space by spheres, JAMS 6 (1966), 179-
192 [2].
[Dic5] - The extreme coverings of 4-space by spheres, JAMS 7 (1967), 490-
496 [2].
[Dic6] - A sufficient condition for an extreme covering of n-space by spheres,
JAMS 8 (1968), 56-62 [2].
[Dic7] - On Voronoi reduction of positive definite quadratic forms, JNT 4
(1972), 330-341 [2].
[Diel] U. Dieter, How to calculate shortest vectors in a lattice, MTAC 29
(1975),827-833 [2].
[Dijl] E. W. Dijkstra, A Discipline of Programming, Prentice-Hall, Englewood
Cliffs NJ, 1976, Chap. 19 [15].
[Dirl] P. G. L. Dirichlet, Uber die Reduktion der positiven quadratische
Formen mit drei unbestimmten ganzen Zahlen, JRAM 40 (1850), 216-
219 [2].
[Dobl] D. Dobkin and R. J. Lipton, Multidimensional searching problems,
SIAM J. Comput. 5 (1976), 181-186 [I].
N. P. Dolbilin: see B. N. Delone.
[Doll] I. Dolgachev, Integral quadratic forms: applications to algebraic
geometry, Sem. Bourbaki 611, Asterisque 105 (1983),251-278 [2, 151.
Bibliography 589

[Donll J. L. Donaldson, Minkowski reduction of integral matrices, MTAC 33

(1978),201-216 [2].
E. de Doncker: see D. Roose.
C. P. Downey: see also J. K. Karlof.
[Dowll - and J. K. Karlof, On the existence of [M, n] group codes for the
Gaussian channel with M and n odd, PGIT 23 (1977), 500-503 [3].
[Dow2] - - Optimum [M,3] group codes for the Gaussian channel, PGIT 24
(1978),760-761 [3].
[Dow3] - - Computation methods for optimal [M,3] group codes for the
Gaussian channel, Util. Math. 18 (1980), 51-70 [3].
[Dow4] D. E. Downey and N. J. A. Sloane, The covering radius of cyclic codes
of length up to 31, PGIT 31 (1985),446-447 [3].
[Dull P. Du Val, Homographies, Quaternions and Rotations, Oxford Univ.
Press, 1964 [8].
[Dun] R. J. Duffin, Infinite programs, in Linear inequalities and related
systems, ed. H. W. Kuhn and A. W. Tucker, Princeton Univ. Press,
1956, pp. 157-170 [9].
[Dunll C. F. Dunld, A Krawtchouk polynomial addition theorem and wreath
products of symmetric groups, Indiana Univ. Math. J. 2S (1976), 335-
358 [9].
[Dun2] - Spherical functions on compact groups and applications to special
functions, Symposia Mathematica 22 (1977), 145-161 [9].
[Dun3] - An addition theorem for some q-Hahn polynomials, Monatsh. Math.
8S (1977), 5-37 [9].
[Dun4] - An addition theorem for Hahn polynomials: the spherical functions,
SIAM J. Math. Anal. 9 (1978), 627-637 [9].
[Dun5] - Discrete quadrature and bounds on t-design, Mich. Math. J. 26
(1979),81-102 [3,9].
[Dun6] - Orthogonal functions on some permutation groups, PSPM 34
(1979), 129-147 [9].
[Dun6a] - Orthogonal polynomials on the sphere with octahedral symmetry,
TAMS 282 (1984),555-575 [9].
[Dun6b] - Orthogonal polynomials with symmetry of order three, CJM 36
(1984),685-717 [9].
[Dun6c] - Reflection groups and orthogonal polynomials on the sphere, Math.
Z., submitted [9].
[Dun7] - and D. E. Ramirez, Topics in Harmonic AnalYSiS, Appleton-
Century-Crofts, NY, 1971 [9].
[Dun8] - - Krawtchouk polynomials and the symmetrization of hypergroups,
SIAM J. Math. Anal. S (1974), 351-366 [9].
[Dun9] - - A linear programming problem in harmonic analYSiS, Linear Aig.
Applic. 14 (1976), 107-116 [9].
[DunlO] J. G. Dunn, The performance of a class of n-dimensional quantizers for
a Gaussian source, in Proc. Symp. Signal Trans. Process., IEEE Press,
NY, 1965, pp. 76-81 [2].
J.-P. Dussault: see W. Brostow.
[Dyml] H. Dym and H. P. McKean, Fourier Series and Integrals, Ac. Press,
NY, 1972 [3,4,9].
E
[Earll A. G. Earnest, Spinor genera of unimodular Z-Iattices in quadratic
fields, PAMS 64 (1977), 189-195 [15].
[Ear2] - and J. S. Hsia, Spinor norms of local integral rotations II, PJM 61
590 Bibliography

(1975),71-86; 115 (1984),493-494 [15).

[Ear3] - - Spinor genera under field extensions I, Acta Arith. 32 (1977),
115-128 [15).
[Ear4] Spinor genera under field extensions II: 2 unramified in the
bottom field, AJM 100 (1978),523-538 [15).
[Ear5] - - Spinor genera under field extensions III: quadratic extensions,
in Number Theory and Algebra, ed. H. Zassenhaus, Ac. Press, NY
1977, pp. 43-62 [15).
[Edel] H. Edelsbrunner and R. Seidel, Voronoi diagrams and arrangements,
Discr. Comput. Geom. 1 (1986),25-44 [2).
[Edgl1 W. L. Edge, The geometry of an orthogonal group in six variables,
PLMS 8 (1958), 416-446 [4).
[Edg2] - The partitioning of an orthogonal group in six variables, PRS A247
(1958),539-549 [4).
[Edg3] - An orthogonal group of order 2 13 '3 5 '5 2 '7, Annali di Mat. 61
(1963), 1-95 [4).
[Edwl] H. M. Edwards, Fermat's Last Theorem: A Genetic Introduction to
Algebraic Number Theory, Springer-Verlag, 1977 [15].
[Eicl1 M. Eichler, Quadratische Formen und Orthogonal Gruppen, Springer-
Verlag, 1952 [15, 16).
[Einl1 G. Einarsson, Performance of polyphase signals on a Gaussian channel,
Ericsson Tech. 23 (No.4, 1967),411-438 [3].
[Ein2] - Polyphase coding for a Gaussian channel, Ericsson Tech. 24 (No.2,
1968),75-130 [3).
M. Elia: see E. Biglieri.
[Elil] P. Elias, Bounds and asymptotes for the performance of multivariate
quantizers, Ann. Math. Stat. 41 (1970), 1249-1259 [2).
[Elsl] V. Elser, The diffraction pattern of projected structures, Acta Cryst.
A42 (1986), 36-43 [11.
[Els2] - and N. J. A. Sloane, A highly symmetric four-dimensional
quasicrystal, J. Phys. A, (1987), to appear [I).
[Eltl] E. L. Elte, The semiregular polytopes of the hyperspaces, Doctor of
Math. and Science Dissertation, State Univ. of Gronigen, 1912 [2!l.
[Emdl] P. van Emde Boas, Another NP-complete partition problem and the
complexity of computing short vectors in lattices, Math. Dept. Report
81-04, Univ. Amsterdam, 1981 [2).
M. Enguehard: see M. Broue.
[Enrl] G. M. Enright, A description of the Fischer group F 22 , J. Alg. 46
(1977),334-343 [10].
[Enr2] - Subgroups generated by transpositions in F22 and F 23 , Comm. Alg. 6
(1978),823-837 (10).
[Erdl] A. Erdelyi, editor, Higher Transcendental Functions, McGraw-Hill, NY,
3 vols., 1953, especially Vol. II, Chap. XI [9].
[Erd2] P. ErdOs and C. A. Rogers, Covering space with convex bodies, Acta
Arith.7 (1962), 281-285 (2).
[Erol] V. A. Erokhin, Theta-series of even unimodular 24-dimensionallattices,
LOMI86 (1979),82-93,190 - JSM 17 (1981),1999-2008 (16).
[Ero2] - On theta-series of even unimodular lattices, LOMI 112 (1981), 59-
70, 199 - JSM 25 (1984), 1012-1020 [16).
[Ero3] - Automorphism groups of 24-dimensional even unimodular lattices,
LOMI 116 (1982), 68-73, 162 = JSM 26 (1984), 1876-1879 [16].
[Estl] D. Estes and G. Pall, Spinor genera of binary quadratic forms, JNT 5
(1973),421-432 [I5).
[Eva 11 D. M. Evans, The 7 -modular representations of Janko's smallest simple
Bibliography 591

[Evall D. M. Evans, The 7 -modular representations of Janko's smallest simple

group, 1. Alg. 96 (1985), 35-44 [101.
F
[Fall) D. D. Falconer and R. D. Gitlin, editors, Special Issue on Voiceband
Telephone Data Transmission, IEEE 1. Select. Areas Commun. 2
(No.5, 1984) [3, B).
H. M. Farkas: see H. E. Rauch.
[Feil) W. Feit, On integral representations of finite groups, PLMS 29 (1974),
633-683 [3, 81.
[Fei2) - Some lattices over Qh/::J), 1. Alg. 52 (1978),248-263 [2,4,7, 161.
[Fejl) G. Fejes Toth, New results in the theory of packing and covering, in
[Gru2), 1983, pp. 318-359 [1, 21.
[Fej2) - and L. Fejes Toth, Dictators on a planet, Studia Sci. Math. Hung.
15 (1980), 313-316 [I).
[Fej2a) P. Gritzmann and 1. M. Wills, Sausage-skin problems for finite
covering, Math. 31 (1984), 117-136 [11.
[Fej2b) - - - Finite sphere packing and sphere covering, Discrete Com put.
Geom. to appear [I I.
L. Fejes Toth: see also G. Fejes Toth.
[Fej3) - Uber einen geometrischen Satz, Math. Z. 46 (1940) 79-83 [II.
[Fej4] - Kreisausfiillungen der hyperbolischen Ebene, AMAH 4 (1953),
103-110 [11.
[Fej5) - Kreisiiberdeckungen der hyperbolischen Ebene, AMAH 4 (1953),
111-114 [II.
[Fej6) - On c1ose-packings of spheres in spaces of constant curvature, Pub!.
Math. Debrecen 3 (1953), 158-167 [11.
[Fej7) - Kugelunterdeckungen und Kugeliiberdeckungen in Riiumen
konstanter Kriimmung, Archiv Math. 10 (1959),307-313 [11.
[Fej8) - Sur la representation d'une population infinie par une nombre fini
d'elements, AMAH 10 (1959), 299-304 [21.
[Fej8a) - What the bees know and what they do not know, BAMS 70 (1964),
468-481 [4).
[Fej9) - Regular Figures, Pergamon, Oxford, 1964 [1, 2, 4, 211.
[Fejl0) - Lagerungen in der Ebene, auf der Kugel und in Raum, Springer-
Verlag, 2nd ed., 1972 [1, 2, 3, 4, 21, B).
[Fejll) - Solid packing of circles in the hyperbolic plane, Studia Sci. Math.
Hungar. 15 (1980), 299-302 [11.
[FejI2) - Stable packings of circles on the sphere, Structural Topology 11
(1985),9-14.
[Ferll G. Ferraris and G. Ivaldi, A simple method for Bravais lattice
determination, Acta Cryst. A39 (1983), 595-596 [21.
L. Few: see also H. S. M. Coxeter.
[Fewl) - Covering space by spheres, Math. 3 (1956), 136-139 [2,41.
[Fiel) H. Fiedler, W. lurkat and O. Korner, Asymptotic expansions of finite
theta series, Acta Arith. 32 (1977), 723-745.
[Fie2) K. L. Fields, Stable, fragile and absolutely symmetric quadratic forms,
Math. 26 (1979),76-79.
[Finll U. Fincke and M. Pohst, Improved methods for calculating vectors of
short length in a lattice, including a complexity analysis, MTAC 44
(1985), 463-471 [21.
[Fin2) L. Finkelstein, The maximal subgroups of Conway's group C 3 and
McLaughlin's group, 1. Alg. 25 (1973), 58-89 [101.
592 Bibliography

[Fin3) and A. Rudvalis, Maximal subgroups of the Hall-Janko- Wales

group, J. Alg. 24 (1973),486-493 [101.
[Fin4) The maximal subgroups of Janko's simple group of order
50,232,960, J. Alg. 30 (1974), 122-143 [101.
[FinS) J. L. Finney, A procedure for the construction of Voronoi polyhedra, J.
Compo Phys. 32 (1979), 137-143 [21.
[Fisl) B. Fischer, Finite groups generated by 3-transpositions I, Invent. Math.
13 (1971), 232-246 [101.
W. Fisher: see E. Koch.
[Flal) J. L. Flanagan, M. R. Schroeder, B. S. Atal, R. E. Crochiere,
N. S. Jayant and J. M. Tribolet, Speech coding, COMM 27 (1979),
710-737 and 932 [21.
[Fla2) L. Flatto, Basic sets of invariants for finite reflection groups, BAMS 74
(1968),730-734 [7).
[Fla3) - Invariants of finite reflection groups and mean value problems II,
AJM 92 (1970), 552-561 [71.
[Fla4) Invariants of finite reflection groups, L'Enseignement Math. 24
(1978), 237-292 [71.
[Fla5) - and M. M. Wiener, Invariants of finite reflection groups and mean
value problems, AJM 91 (1969),591-598 [7).
[Fla6) - - Regular polytopes and harmonic polynomials, CJM 22 (1970),
7-21 [71.
[Fla7) - and D. J. Newman, Random coverings, Acta Math. 138 (1977),
241-264.
[Foil) J. H. Folkman and R. L. Graham, A packing inequality for compact
convex subsets of the plane, CMB 12 (1969), 745-752 [II.
P. Fong: see also A. O. L. Atkin.
[FonI) - Characters arising in the Monster-modular connection, PSPM 37
(1980), 557-559 [1, 301.
[Fori) G. D. Forney, Jr., personal communication [3).
[For2) - Coset codes, PGIT to appear [2, 3, 201.
[For3) - R. G. Gallager, G. R. Lang, F. M. Longstaff and S. U. Qureshi,
Efficient modulation for band-limited channels, in [Fall), 1984,
pp. 632-646 [31.
[Fosl) G. J. Foschini, R. D. Gitlin and S. B. Weinstein, Optimization of two-
dimensional signal constellations in the presence of Gaussian noise,
COMM 22 (1971),28-38 [31.
[Fos2) - - - On the selection of a two-dimensional signal constellation in
the presence of phase jitter and Gaussian noise, BSTJ 52 (1973), 927-
965 [3],
[Fowl) R. J. Fowler, M. S. Paterson and S. L. Tanimoto, Optimal packing and
covering in the plane are NP-complete, Info. Proc. Lett. 12 (1981),
133-137 [2).
B. L. Fox: see W. Brostow.
[Fral) W. Fraser and C. C. Gotlieb, A calculation of the number of lattice
points in the circle and sphere, MTAC 16 (1962), 282-290 [4).
P. O. Frederickson: see J. R. Baumgardner.
[Frel) I. B. Frenkel, J. Lepowsky and A. Meurman, A natural representation
of the Fischer-Griess Monster with the modular function J as character,
PNAS 81 (1984),3256-3260 [29, 30).
[Fre2) - - - A moonshine module for the Monster, in [Lepl), 1984,
pp. 231-273 [29, 301.
[Fre3) - - - An Eg-approach to Flo in [McK3), 1985, pp. 99-120 [29, 30).
Bibliography 593

[Fre4] - - - An introduction to the Monster, in Unified String Theories,

M. G. Green and D. Gross, eds., World Scientific, Singapore 1986,
pp. 533-546 [29, 301.
[Fre5] - - - Vertex Operators and the Monster, to appear [29, 30].
[Frill F. Fricker, Einfuhrung in die Gitterpunktlehre, Birkhiiuser, Boston, 1982
[4].
[Fri2] D. Fried and W. M. Goldman, Three-dimensional affine
crystallographic groups, Adv. in Math. 47 (1983), 1-49 [41.
A. Frohlich: see J. W. S. Cassels.
[Froll K. K. Frolov, On the connection between quadrature formulas and
sub lattices of the lattice of integral vectors, DAN 232 (No. I, 1977) =
SMD 18 (1977),37-41 [Il.
W. Fumy: see T. Beth.
G
[GabIl E. M. Gabidulin and V. R. Sidorenko, A general bound for code volume,
PPI 12 (No.4, 1976),31-35 - PIT 12 (1976),226-269 [9].
[Gagl] T. M. Gagen, A characterization of Janko's simple group, PAMS 19
(1968), 1393-1395 [101.
[Gag2] - editor, Finite Groups 72, North-Holland, Amsterdam, 1973 [B1.
R. V. Galiulin: see also B. N. Delone.
[GaIO] - Delaunay systems, Kristallografiya 25 (1980), 901-907 - SOy. Phys.
Cryst. 25 (1980), 517-520.
R. G. Gallager: see also G. D. Forney, Jr., C. E. Shannon.
[Gall] - Information Theory and Reliable Communication, Wiley, NY 1968
[31.
[GaI2] N. C. Gallagher, Jr. and J. A. Bucklew, Some recent developments in
quantization theory, in Proc. 12th Ann. Sympos. System Theory,
Virginia Beach VA, May 19-20, 1980 [21.
[Gamll A. F. Gameckii, On the theory of covering an n-dimensional Euclidean
space with equal spheres, DAN 146 (1962), 991-994 - SMD 3 (1962),
1410-1414 [2,41.
[Gam2] - The optimality of the principal lattice of Voronoi of the first type
among the lattices of the first type of any number of dimensions, DAN
151 (1963),482-484 = SMD 4 (1963), 1014-1016 [2,41.
[Ganl] R. Gangolli, Spherical functions on semisimple Lie groups, in [BooIl,
1972, pp. 41-92 [91.
[Gaol] J. Gao, L. D. Rudolph and C. R. P. Hartmann, Iteratively maximum
likelihood decodable spherical codes and a method for their
construction, PGIT to appear [31.
R. A. Gaskins: see I. J. Good.
Z. Gaspar: see T. Tarnai.
[Gasl] G. Gasper, Positivity and special functions, in [Ask2a], 1975, pp. 375-
433 [91.
[Gaul] C. F. Gauss, Disquisitiones Arithmeticae, Fleischer, Leipzig, 1801;
English translation, Yale University Press, 1966 [15].
[Gau2] - Besprechung des Buchs von L. A. Seeber: Untersuchungen uber die
Eigenschaften der positiven terniiren quadratischen Formen usw.,
Gottingsche Gelehrte Anzeigen (1831, July 9) - Werke, II (1876),
188-196 [11.
[Gell] I. M. Gelfand, Spherical functions on symmetric Riemannian spaces
DAN 70 (No.1, 1970) 5-8 = Amer. Math. Soc. Trans!. (2) 37 (1964),
39-43 [91.
594 Bibliography

[GeI2) M. Geller, Comment on "New Jacobian theta functions and the

evaluation of lattice sums" by l. J. Zucker, J. Math. Phys. 18 (1977),
187 [21.
A. Gersho: see also D.-Y. Cheng, C.-S. Wang.
[Gerl) - Principles of quantization, IEEE Trans. Circuit Syst. 25 (1978),
427-436 [21.
[Ger2) - Asymptotically optimal block quantization, PGIT 25 (1979), 373-
380 [21.
[Ger3) - On the structure of vector quantizers, PGIT 28 (1982), 157-166 [2,
201.
[Ger4) - and V. B. Lawrence, Multidimensional signal constellations for
voiceband data transmission, in [Fall), 1984, pp. 687-702 [31.
[Ger5) L. J. Gerstein, Classes of definite hermitian forms, AJM 100 (1978),
81-97 [21.
[Gibll I. B. Gibson and I. F. Blake, Decoding the binary Golay code with
miracle octad generators, PGIT 24 (1978), 261-264 [11 I.
[Gill) E. N. Gilbert, A comparison of signalling alphabets, BSTJ 31 (1952),
504-522 [3, 9).
[Ginll V. V. Ginzburg, Multidimensional signals for a continuous channel, PPI
20 (1984) 28-46 = PIT 20 (1984) 20-28 [31.
[Gisl) H. Gish and J. N. Pierce, Asymptotically efficient quantizing, PGIT 14
(1968), 676-683 [21.
R. D. Gitlin: see D. D. Falconer, G. J. Foschini.
[Glal) J. W. L. Glaisher, On the representations of a number as a sum of four
squares and on some allied arithmetical functions, Quart. J. Math. 36
(1905),305-358 [41.
[Gla2) - The arithmetical functions P(m), Q (m), n (m), Quart. J. Math. 37
(1906),36-48 [41.
[Gla3) - On the representations of a number as the sum of two, four, six,
eight, ten, and twelve squares, Quart. J. Math. 38 (1907), 1-62 [41.
[Gla4) - On the representations of a number as the sum of fourteen and
sixteen squares, Quart. J. Math. 38 (1907),178-236 [41.
[Gla5) - On the representations of a number as the sum of eighteen squares,
Quart. J. Math. 38 (1907),289-351 [41.
[Gla6) - On the numbers of representations of a number as a sum of 2r
squares, where 2r does not exceed eighteen, PLMS 5 (1907), 479-490
[41.
[Gla7) M. L. Glasser, The evaluation of lattice sums, J. Math. Phys. 14
(1973), 409-413, 701-703; 15 (1974), 188-189, 520; 16 (1975), 1237-
1238 [2).
[Gla8) - and I. J. Zucker, Lattice sums, Theoret. Chern.: Advances and
Perspectives 5 (1980),67-139 [21.
B. G. Glazer: see C. N. Campopiano.
[Glell A. M. Gleason, Weight polynomials of self-dual codes and the
Mac Williams identities, Actes, Congres Intern. Math., Gauthier-Villars,
Paris, 1971, 3, pp. 211-215 [71.
[God 11 P. Goddard and D. Olive, Algebras, lattices and strings, in [Lep1), 1985,
pp. 51-96 [J).
J.-M. Goethals: see also P. DeIsarte, F. J. MacWilliams.
[Goel) - Two dual families of nonlinear binary codes, Elect. Lett. 10 (1974),
471-472 [51.
[Goe2) - Nonlinear codes defined by quadratic forms over GF(2), IC 31
(1976),43-74 [51.
Bibliography 595

[Goe31 - The extended Nadler code is unique, PGIT 23 (1977), 132-135 [91.
[Goe41 - Association schemes, in Algebraic Coding Theory and Applications,
ed. G. Longo, CISM Courses and Lectures No. 258, Springer-Verlag,
1979, pp. 243-283 [9, 141.
[Goe51 - and J. J. Seidel, Cubature formulae, polytopes and spherical designs,
pp. 203-218 of [Dav31 [1, 31.
[Goe61 - - Spherical designs, PSPM 34 (1979), 255-272 131.
WoBI M. J. E. Golay, Notes on digital coding, PIEEE 37 (1949), 637 [31.
Wol21 - Binary coding, PGIT 4 (1954), 23-28 [3, 51.
[Gol31 M. Goldberg, Packing of 18 equal circles on a sphere, Elem. Math. 20
(1965), 59-61 [11.
[Go141 - Packing of 19 equal circles on a sphere, Elem. Math. 22 (1967),
108-110 [11.
[Go151 - An improved packing of 33 equal circles on a sphere, Elem. Math.
22 (1967), 110-111 [11.
[Go161 - AXially symmetric packing of equal circles on a sphere, Ann. Univ.
Sci. Budapest 10 (1967), 37-48; 12 (1969), 137-142 [11.
[Go171 - Stability configurations of electrons on a sphere, MTAC 23 (1969),
785-786 Ill.
[Go181 - On the densest packing of equal spheres in a cube, Math. Mag. 44
(1971), 199-208 [11.
W. M. Goldman: see D. Fried.
D. Golke: see H. E. Ong.
[Go191 E. S. Golod and I. R. Shafarevich, On class field towers, IANS 28
(1964),261-272 - Amer. Math. Soc. Trans!. (2) 48 (1965),91-102 [I,
81.
[Gooll I. J. Good and R. A. Gaskins, The centroid method of numerical
integration, Num. Math. 16 (1971), 343-359 [211.
[Gopl1 V. D. Goppa, A new class of linear error-correcting codes, PPI 6
(No.3, 1970) = PIT 6 (1970), 207-212 131.
[Gop21 - Codes associated with divisors, PPI 13 (1977), 33-39 = PIT 13
(1977) 22-27 [31.
Wop31 - Codes on algebraic curves, DAN 259 (1981), 1289-1290 = SMD 24
(1981),170-172 [31.
[Gop41 - Algebraic-geometric codes, IANS 46 (1982), 762-781 = Izv. 21
(1983),75-91 131.
Wop51 - Codes and information, Usp. Mat. Nauk 39 (1984), 77-120 - Russ.
Math. Surv. 39 (1984), 87-141 131.
[GorOI D. M. Gordon, Minimal permutation sets for decoding the binary Golay
codes, PGIT 28 (1982), 541-543 [201.
[Gorll L. M. Gordon and R. Levingston, The construction of some
automorphic graphs, Geom. Dedic. 10 (1981), 261-267 [91.
[Gor21 D. Gorenstein, The classification of finite simple groups I. Simple
groups and local analysis, BAMS 1 (1979),43-199 [3, 101.
[Gor31 - Finite Simple Groups, Plenum, NY, 1982 [3, 101.
[Gor41 - The Classification of Finite Simple Groups, Plenum, NY, 1983,
Vo!. 1 [3, 10].
[Gor51 - Classifying the finite simple groups, BAMS 14 (1986),1-98 [3,101.
[Gor61 - and K. Harada, A characterization of Janko's two new simple
groups, J. Fac. Sci. Univ. Tokyo 16 (1970), 331-406 [101.
Wosl1 T. Gosset, On the regular and semi-regular figures in space of n
dimensions, Messenger Math. 29 (1900),43-48 [4].
C. C. Gotlieb: see W. Fraser.
596 Bibliography

R. L. Graham: see also P. Diaconis, J. H. Folkman.

[Gral) - and N. J. A. Sloane, On constant weight codes and harmonious
graphs, in Proc. West coast conference on combinatorics, graph theory
and computing, Utilitas Math. Publ. Co., Winnipeg, Canada, 1979,
pp. 25-40 [91.
[Gra2) - - Lower bounds for constant weight codes, PGIT 26 (J980), 37-43
[3,91.
[Gra3) On additive bases and harmonic graphs, SIAD 1 (J 980), 382-404
[91.
[Gra4) On the covering radius of codes, PGIT 31 (J 985), 385-401 [2, 31.
[Gra4a) Penny-packing and two-dimensional codes, Discrete Comput.
Geom. to appear [11.
[Gra5) - H. S. Witsenhausen and H. J. Zassenhaus, On tightest packings in
the Minkowski plane, PJM 41 (1972),699-715 [II.
D. Gratias: see also D. Shechtman.
[Gra6) and L. Michel, editors, International Workshop on Aperiodic
Crystals, J. de Physique 47 (Colloq. C3, supp!. to No.7, July 1986) [11.
R. M. Gray: see also L. D. Davisson, Y. Linde, Y. Yamada.
[Gra7) - editor, Special Issue on Quantization, PGIT 28 (J982), 127-262 [2).
M. B. Green: see also I. B. Frenkel.
[Grel) - Unification of forces and particles in superstring theories, Nature
314 (J985), 409-414 [11.
[Gre2) R. R. Green, A serial orthogonal decoder, JPL Space Programs
Summary, Vol. 37-39-IV (J966), 247-252 [201.
[Gre3) - Analysis of a serial orthogonal decoder, JPL Space Programs
Summary, Vol. 37-53-III (J968), 185-187 [201.
[Grill R. L. Griess, Jr., Schur multipliers of the known finite simple groups I,
BAMS 78 (J972), 68-71 [101.
[Gri2) - The structure of the "Monster" simple group, in [Sco3), 1976,
pp. 113-118 [29).
[Gri3) - Schur multipliers of the known finite simple groups II, PSPM 37
(1980),279-282 [10).
[Gri4) - A construction of F, as automorphisms of a 196,883-dimensional
algebra, PNAS 78 (J981), 689-691 [3, 10,291.
[Gri5) - The friendly giant, Invent. Math. 69 (J982), 1-102 [3, 10,291.
[Gri6) - The sporadic simple groups and construction of the Monster, in
Proc. Intern. Congr. Math. Warsaw 1983, North-Holland, Amsterdam,
1984, pp. 369-384 [10, 291.
[Gri7) - Quotients of infinite reflection groups, Math. Ann. 263 (1983), 267-
278 [10, 291.
[Gri8) - The Monster and its nonassociative algebra, in [McK31, 1985,
pp. 121-157 [10, 291.
[Gri9) - Schur multipliers of the known finite simple groups III, in Proc.
Rutgers Group Theory Year 1983-1984, ed. M. Aschbacher et aI.,
Camb. Univ. Press, 1984, pp. 69-80 [101.
[GrilO) - Code loops, J. Alg. 100 (1986),224-234 [29).
[GrilOa) - Sporadic groups, code loops and nonvanishing cohomology, J. Pure
Appl. Alg. 44 (J978), 191-214 [29).
P. Gritzmann: see also G. Fejes T6th.
[Grill) - Lottice covering of space with symmetric convex bodies, Math. 32
(J985), 311-315 [2).
[GriI2) - Finite packing of equal balls, JLMS 33 (J 986), 543-553 [II.
[GriI3) - and J. M. Wills, An upper estimate for the lattice point enumerator,
Math. 33 (J 986), 196-202 [II.
Bibliography 597

[Gri14] Finite packing and covering, Studia Sci. Math. Hungar 21

(1986),149-162 [11.
[GroO] H. Groemer, Some basic properties of packing and covering constants,
Discr. Comput. Geom 1 (1986), 183-193 [I].
D. Gross: see I. B. Frenkel.
F. Gross: see W. R. Scott.
H. Gross: see B. L. van der Waerden.
[Groll K. I. Gross, On the evolution of noncommutative harmonic analysis,
AMM 85 (1978), 525-548 [9].
[Gr02] E. Grosswald, Representations of Integers as Sums of Squares,
Springer-Verlag, 1985 [41.
H. Grothe: see L. Afflerbach.
[Gr03] L. C. Grove and C. T. Benson, Finite Reflection Groups, Springer-
Verlag, 2nd ed., 1985 [4,211.
[Grull P. M. Gruber, Geometry of numbers, in [Toll], 1979, pp. 186-225 [1, 2,
15].
[Grula] - and C. G. Lekkerkerker, Geometry of Numbers, 2nd ed., to appear
[I, B].
[Gru2] - and J. M. Wills, editors, Convexity and Its Applications, Birkhauser,
Boston, 1983 [B].
B. Griinbaum: see also C. Davis.
[Grull - Convex Polytopes, Wiley, NY, 1967 [211.
[Gru2] - and G. C. Shephard, Patterns on the 2-sphere, Math. 28 (1981), 1-
35 [11.
[Gru3] - - Tilings and Patterns, Freeman, NY, 1987 [11.
[GudI] P. M. Gudivok, A. A. Kirilyuk, V. P. Rud'ko and A. I. Tsitkin, Finite
subgroups of the group GL(n,Z), Kibernetika 18 (No.6, 1982) 71-82 -
Cybernetics 18 (1982), 788-803 [31.
[Gunl] R. C. Gunning, Lectures on Modular Forms, Princeton Univ. Press,
1962 [2, 71.
[GunI] S. Giinther, Ein stereometrisches Problem, Archiv Math. Physik
(Grunert) 57 (1875), 209-215 [II.
[Gus 11 W. H. Gustafson, Remarks on the history and applications of integral
representations, in [RogIO], 1981, pp. 1-36 [31.
[Guyl] R. K. Guy, editor, Reviews in Number Theory 1973-83, Amer. Math.
Soc., Providence RI, 6 vols., 1984 [B1.
H

[HabI] W. Habicht and B. L. van der Waerden, Lagerung von Punkten aUf der
Kugel, Math. Ann. 123 (1951), 223-234 [11.
[HadI] H. Hadwiger, Uber ausgezeichnete Vektorsterne und reguliire Polytope,
Comment. Math. Helvet. 13 (1940),90-107 [61.
[HahI] T. Hahn, editor, International Tables for Crystallography, Reidel,
Dordrecht, 1983 [1, B1.
J. L. Hall: see M. R. Schroeder.
M. Hall, Jr.: see also N. J. A. Sloane.
[Hall] - Hadamard matrices of order 16, Jet Propulsion Lab. Research
Summary 36-10, Vol. 1 (Sept. 1, 1961), pp. 21-26 [31.
[HaI2] - Hadamard matrices of order 20, Jet Propulsion Lab. Technical
Report 32-761 Jet Propulsion Lab., Pasadena CA, 1965 [3].
[Hal3] - Combinatorial Theory, Blaisdell, Waltham MA, 1967 [31.
[HaI4] - and D. B. Wales, The simple group of order 604,800 J. Alg. 9
(1968),417-450 [10].
598 Bibliography

[Ha15] S. R. Hall and T. Ashida, editors, Methods and Applications in

Crystallographic Computing, Oxford Univ. Press, 1984 [21
[Ha16] P. R. Halmos, Measure Theory, Van Nostrand, Princeton NJ, 1965 [91
B. I. Halperin: see D. R. Nelson.
H. Hiimiiliiinen: see I. Honkala.
I. Hambleton: see C. R. Riehm.
[Haml] C. M. Hamill, On a finite group of order 6,531,840, PLMS 52 (1951),
401-454 [41
[Ham2] J. Hammer, Unsolved Problems Concerning Lattice Points, Pitman, San
Francisco CA, 1977 [11
[Ham3] P. Hammond, q-Coverings and completely regular codes in distance-
regular graphs, Quart. 1. Math. Oxford 29 (1978),23-30 [91
[Ham4] - and D. H. Smith, Perfect codes in the graphs Ob JCT B19 (1975),
239-255 [9].
[Hanl1 H. Hanani, On quadruple systems, CJM 12 (1960), 145-157 [3,51
[Han2] - Balanced incomplete block deSigns and related designs, Discr. Math.
11 (1975), 255-369 [3].
[Han3] H. Hancock, Development of the Minkowski Geometry of Numbers,
Dover, NY, 1964 [11
K. Hanes: see 1. D. Berman.
W. Hanrath: see W. Plesken.
K. Harada: see also D. Gorenstein.
[HarO] - On the simple group F of order 2 14 . 36. 56. 7· 11 '19, in [Sco3], 1976,
119-276 [101
[Harl] - The automorphism group and the Schur multiplier of the simple
group of order 214 '3 6'5 6'7'11'19, Osaka J. Math. 15 (1978),633-635
[101
[Har2] F. Harary, Graph Theory, Addison-Wesley, Reading MA, 1969 [281
[Har3] G. H. Hardy, On the representation of a number as the sum of any
number of squares, and in particular of five, TAMS 21 (1920), 255-284
- Coli. Papers, I, 345-374 [41
[Har4] - Ramanujan, Camb. Univ. Press, 1940; Chelsea, NY, 1959 [2, 4, 71
[Har5] - and E. M. Wright, An Introduction to the Theory of Numbers,
Oxford Univ. Press, 5th ed., 1980 [2, 4].
[Har6] E. M. Hartley, A sextic primal in five dimensions, PCPS 46 (1950),
91-105 [41
[Har7] - Two maximal subgroups of a collineation group in five dimensions,
PCPS 46 (1950), 555-569 [41
C. R. P. Hartmann: see J. Gao.
[Har8] M. Harwit and N. J. A. Sloane, Hadamard Transform Optics, Ac.
Press, NY, 1979 [3,201
[HasO] H. Hasse, Uber die Klassenzahl Abelscher Zahlkorper, Akademie-
Verlag, Berlin, 1952 [8].
[Hasl1 - Number Theory, Springer-Verlag, 1980 [8].
[Hazl1 M. Hazewinkel, W. Hesselink, D. Siersma and F. D. Veldkamp, The
ubiquity of Coxeter-Dynkin diagrams (an introduction to the A-D-E
problem), Nieuw Arch. Wisk. 25 (1977),257-307 [1, 2,4,211
[HecO] E. Heeke, Vorlesungen iiber die Theorie der Algebraischen Zahlen,
Chelsea, NY 1970; English translation, Springer-Verlag, 1981 [8].
[Heel1 - Uber Modulfunktionen und die Dirichletschen Reihen mit
Eulerscher Produktentwickling, Math. Ann. 114 (1937), 1-28 and 316-
351 = [Hec3], pp. 644-707 [7].
[Hec2] - Analytische Arithmetik der positiven quadratischen Formen, Kg!.
Bibliography 599

Danske Vid. Selskab. Mat.-fys. Medd. 13 (No. 12, 1940) - [Hec3],

pp. 789-918 [2,7, 181.
[Hec3] - Mathematische Werke, Vandenhoeck and Ruprecht, Gottingen, 1970
[2,7,8, BI.
[Hedl] A. Hedayat and W. D. Wallis, Hadamard matrices and their
applications, Ann. Stat. 6 (1978), 1184-1238 [31.
[Hell] D. Held, The simple groups related to M 24 , J. Alg. 13 (1969), 253-296
[101.
[HeI2] - The simple groups related to M 24, II, JAMS 16 (1973),24-28 HoI.
[Hel3] B. Helfrich, An algorithm to construct Minkowski-reduced lattice-
bases, Lect. Notes Compo Sci. 182 (1985), 173-179 [21.
[HeI4] - Algorithms to construct Minkowski reduced and Hermite reduced
lattice bases, Theoretical Computer Science 41 (No. 2-3, 1986), 125-139
[2].
[HeI5] S. Helgason, Differential Geometry and Symmetric Spaces, Ac. Press,
NY, 1962 [9].
[HeI6] - Differential Geometry, Lie Groups, and Symmetric Spaces, Ac.
Press, NY, 1978 [91.
[HeI7] H. J. Helgert and R. D. Stinaff, Minimum-distance bounds for binary
linear codes, PGIT 19 (1973), 344-356 [3].
[Hemll C. Hermann, Kristallographie in Riiumen beliebiger Dimensionszahl I,
Die Symmetrieoperationen, Acta Cryst. 2 (1949), 139-145 [31.
[HerO] C. Hermite, Second letter to Jacobi on number theory, JRAM 40
(1850), 279-290 - Oeuvres I (1905), 122-130 [21.
[Herll I. N. Herstein, Topics in Algebra, Ginn, Waltham MA, 1964 [2].
W. Hesselink: see M. Hazewinkel.
T. Hickey: see J. Cohen.
[Higll J. R. Higgins, Five short stories about the cardinal series, BAMS 12
(1985), 45-89 [3].
[Hig2] D. G. Higman, Coherent configurations, Geom. Dedicata 4 (1975), 1-32;
5 (1976), 413-424 [9, 141.
[Hig3] - and C. C. Sims, A simple group of order 44,352,000, Math. Z. 105
(1968), 110-113 HoI.
G. Higman: see also M. B. Powell.
[Hig4] - On the simple group of D. G. Higman and C. C. Sims, 11M 13
(1969),74-80 [101.
[Hig5] - and J. McKay, On Janko's simple group of order 50,232,960, BLMS
1 (1969),89-94 [101.
[Hill] D. Hilbert, Mathematische Probleme, Archiv. Math. Phys. 1 (1901),
44-63 and 213-237 - Gesamm. Abh., III, 290-329. English translation
in BAMS 8 (1902),437-479 - PSPM 28 (1976), 1-34 [I].
[Hil2] - and S. Cohn-Vossen, Anschauliche Geometrie, Julius Springer,
Berlin, 1932. English translation, Geometry and the Imagination,
Chelsea, NY, 1952 [I, 21].
[Hil3] H. Hiller, Crystallography and cohomology of groups, AMM 93
(1986), 765-779 [41.
S. Hirasawa: see Y. Sugiyama.
[Hirl] J. W. P. Hirschfeld, Projective Geometries over Finite Fields, Oxford
Univ. Press, 1979.
[Hir2] - Linear codes and algebraic curves, in Geometrical Combinatorics,
Pitman, Boston, 1984, pp. 35-53 [31.
[Hir3] - Finite Projective Spaces of Three Dimensions, Oxford Univ. Press,
1986.
600 Bibliography

[Hir4] F. Hirzebruch, W. D. Neumann and S. S. Koh, Differentiable Manifolds

and Quadratic Forms, Dekker, NY, 1971 [2, 151.
[Hlal] E. Hlawka, Zur Geometrie der Zahlen, Math. Z. 49 (1944),285-
312 [11.
[Hla2] Zur angeniiherten Berechnung mehrfacher Integraler, Monatsh.
Math. 66 (1962) 140-151 [II.
[Hla3] - 90 Jahre Geometrie der Zahlen, Jahr. Uber. Math. (1980),9-41 [11.
[Hoal] M. R. Hoare, Structure and dynamics of simple microclusters, Adv.
Chern. Phys. 40 (1979),49-135 [11.
[Hoa2] - and J. A. Mcinnes, Morphology and statistical statics of simple
microclusters, Adv. Phys. 32 (1983), 791-821 [l I.
[Hob 11 E. W. Hobson, The Theory of Spherical and Ellipsoidal Harmonics,
Camb. Univ. Press, 1931; Chelsea, NY, 1955 [9].
[Hofl] N. Hofreiter, Uber Extremformen, Monatsh. Math. Phys. 40 (1933),
129-152 [21.
S. G. Hoggar: see also E. Bannai.
[Hogl] - Bounds for quaternionic line systems and reflection groups, Math.
Scand.43 (1978),241-249 [31.
[Hog2] - Parameters oft-designs in Fpd-t, Europ. J. Combin. 5 (1984), 29-36
[31.
[Hog3] - t-Designs in projective spaces, Europ. J. Combin. 3 (1982),233-254[31.
[HohIl B. von Hohenbalken, Finding Simplicial subdivisions of polytopes,
Math. Programming 21 (1981),233-234 [21].
[Holl] A. Holden, Shapes, Space, and Symmetry, Columbia Univ. Press, NY,
1971 [2, 211.
[HonIl Y. Hong, On spherical t-designs in R2, Europ. J. Combin. 3 (1982),
255-258 [31.
[Hon2] I. Honkala, Some optimal constant weight codes, Ars Combin. 820
(1985),43-47 [9].
[Hon3] - Some upper bounds for A (n,d) and A (n,d, w), PGIT to appear [9].
[Hon4] - H. Hamalainen and M. Kaikkonen, A modification of the Zinoviev
lower bound for constant weight codes, Discr. Appl. Math. 11 (1985),
307-310 [9].
[Hon5] - - - Some lower bounds for constant weight codes, preprint [91.
[HopI] R. Hoppe, Bemerkung der Redaction, Archiv Math. Physik (Grunert)
S6 (1874), 307-312 [1).
[How 11 D. J. Howarth and H. Jones, The cellular method of determining
electronic wave functions and eigenvalues in crystals, with applications
to sodium, PRS 6S (1952), 355-368 [2].
J. S. Hsia: see also J. W. Benham, P. J. Costello, A. G. Earnest.
[Hsil] - Spinor norms of local integral rotations I, PJM 57 (1975), 199-206
[l51.
[Hsi2] - Representations by spinor genera, PJM 63 (1976), 147-152 [151.
[Hsi3] - Arithmetic theory of integral quadratiC forms, in Proc. Queen's
Number Theory Conf. 1979, Queen's Papers Pure Appl. Math. 54,
Queen's Univ., Kingston, Ontario, 1980, pp. 173-204 [15].
[Hsi4] - On the classification of unimodular quadratic forms, JNT 12
(980),327-333 [151.
[Hsi5] - Regular positive ternary quadratic forms, Math. 28 (1981), 231-238
[2, 151.
[Hsi6] - Algebraic methods in integral quadratic forms, preprint [151.
[Hsi6a] Even positive definite unimodular quadratic forms over real
quadratic fields, preprint [2, 7, 151.
Bibliography 601

[Hsi7] - and D. C. Hung, Theta series of ternary and quaternary quadratic

forms, Invent. Math. 73 (1983), 151-156 [2, 151.
[Hsi8] - - Theta series of quaternary quadratic forms over Z and
z[(I +.Jj;)/21, Acta Arith. 45 (1985), 75-91 [2, 151.
[Hsi9] - Y. Kitaoka and M. Kneser, Representations of positive definite
quadratic forms, JRAM 301 (1978), 132-141 [151.
[HuaI1 L. K. Hua and Y. Wang, Applications of Number Theory to Numerical
Analysis, Springer-Verlag, 1981 1I1.
[HubI1 X. L. Hubaut, Strongly regular graphs, Discr. Math. 13 (1975), 357-
381 [91.
V. Huber-Dyson: see K. W. Roggenkamp.
[Hufl] W. C. Huffman, The biweight enumerator of self-orthogonal binary
codes, Discr. Math. 26 (1979), 129-143 [71.
[Huf2] - Polynomial invariants of finite linear groups of degree two, CJM 32
(1980),317-330 [7].
[Huf3] - Automorphisms of codes with applications to extremal doubly even
codes of length 48, PGIT 28 (1982), 511-521 [7].
[Huf4] - and V. I. Iorgov, A [72,36, 16] doubly even code does not have an
automorphism of order 11, PGIT to appear [71.
[Huf5] - and N. J. A. Sloane, Most primitive groups have messy invariants,
Adv. in Math. 32 (1979), 118-127 [3].
[Hug 11 D. R. Hughes, A combinatorial construction of the small Mathieu
designs and groups, Ann. Discr. Math. 15 (1982), 259-2641II1.
[Hug2] - and F. C. Piper, Projective Planes, Springer-Verlag, 1973 [3].
[Hug3] - - Design Theory, Camb. Univ. Press, 1985 [3].
K. Huml: see F. R. Ahmed.
[Huml] J. E. Humphreys, Introduction to Lie Algebras and Representation
Theory, Springer-Verlag, 1972 [4,23].
D. C. Hung: see J. S. Hsia.
[Hunl] D. C. Hunt, A characterization of the finite simple group M(23), J.
Alg. 26 (1972), 431-439 [101.
[HupI1 B. Huppert, Endliche Gruppen I, Springer-Verlag, 1967 [3,10].
[HurI1 A. Hurwitz, Uber die Zahlentheorie der Quaternionen, Nach.
Gesellschaft Wiss. Gottingen, Math.-Phys. Klasse (1896), 313-340
Math. Werke, Birkhiiuser, Basel, II, 1933, pp. 303-330 [2,41.
D. Husemoller: see J. Milnor.
B. G. Hyde: see M. O'Keeffe.

lIguI1 J. Igusa, Theta Functions, Springer-Verlag, 1972 [4].

lIhaI1 Y. Ihara, Some remarks on the number of rational points of algebraic
curves over finite fields, J. Fac. Sci. Univ. Tokyo, Sec. lA, 28 (1982),
721-724 [31.
lIncI1 E. L. Inee, Cycles of Reduced Ideals in QuadratiC Fields, Math. Tables
IV, British Assoc. Adv. Sci., Camb. Univ. Press, 1966 [151.
lIng 1] I. Ingemarsson, Commutative group codes for the Gaussian channel,
PGIT 19 (1973),215-219 [31.
lIntI] International Union of Crystallography, International Tables for X-Ray
Crystallography, Kynoch Press, Birmingham, England, 3 vols., 1959 [41.
O. Intrau: see H. Brandt.
V. I. Iorgov: see also W. C. Huffman.
602 Bibliography

Dorl] - Binary self-dual codes with automorphisms of odd order, PPI 19

(No.4, 1983), 11-24 = PIT 19 (1983),260-270 [71.
Dor2] A method for constructing inequivalent self-dual codes with
applications to length 56, PGIT 33 (1987), 77-82 [71.
Dor3] - Doubly-even extremal codes of length 64, PPI 22 (No.4, 1986),
35-42 - PIT 22 (1986), 277-284 [71.
[Ito I] N. Ito, Symmetry codes over GFO), JCT A29 (1980) 251-253 [31.
Dto2] - J. S. Leon and J. Q. Longyear, Classification of 3-(24, 12, 5) designs
and 24-dimensional Hadamard matrices, JCT A31 (1981),66-93 [31.
T. Ito: see E. Bannai.
G. Ivaldi: see G. Ferraris.
N. Iwahori: see J. G. Thompson.
DyaI] K. Iyanaga, Class numbers of definite Hermitian forms, J. Math. Soc.
Japan 21 (1969), 359-374 [7].
J
[Jacl] D. M. Jackson and S. A. Vanstone, editors, Enumeration and Design,
Ac. Press, 1984 [3, B].
[Jac2] M. Jacob, editor, Dual Theory, North-Holland, Amsterdam, 1974 [Il.
I. M. Jacobs: see J. M. Wozencraft.
[Janl] Z. Janko, A new finite simple group with Abelian Sylow 2-subgroups,
PNAS 53 (1965),657-658 [101.
[Jan2] - A new finite simple group with Abelian 2-Sylow subgroups and its
characterization, J. Aig. 3 (1966), 147-186 [101.
[Jan3] - A characterization of a new simple group, in Proc. First Internat.
Conf. Theory of Groups, ed. L. Kovacs and B. Neumann, Gordon and
Breach, NY, 1967, pp. 205-208 [101.
[Jan4] - Some new simple groups of finite order I, First Naz. Alta Math.
Symposia Math. 1 (1968),25-65 [101.
[Jan5] - Some new simple groups of finite order, in [Bra2], 1969, pp.63-64
[10].
[Jan6] - A new finite simple group of order 86,775,571,046,077,562,880
which possesses M24 and the full cover of M22 as subgroups, J. Aig. 42
(1976),564-596 [10].
[Jan7] G. J. Janusz, Algebraic Number Fields, Ac. Press, NY, 1973 [81.
N. S. Jayant: see also J. L. Flanagan.
[Jayl] - editor, Waveform Quantization and Coding, IEEE, NY, 1976 [21.
F. Jelinek: see T. Berger.
[Jeri] A. J. Jerri, The Shannon sampling theorem -its various extensions and
applications: a tutorial review, PIEEE 65 (1977), 1565-1596 [31.
[JodI] W. S. Jodrey and E. M. Tory, Computer simulation of close random
packing of equal spheres, Phys. Rev. A32 (1985), 2347-2351.
[Jonl] B. W. Jones, A table of Eisenstein-reduced positive ternary quadratic
forms of determinant ~ 200, Bull. Nat. Res. Council U.S.A. 97 (1935),
I-51 [151.
[Jon2] - A canonical rational form for the ring of 2-adic integers, DMJ 11
(1944),715-727 [15].
[Jon3] - The Arithmetic Theory of Quadratic Forms, Math. Assoc. America
and Wiley, NY, 1950 [15].
[Jon4] G. A. Jones, Geometric and asymptotic properties of Brillouin zones in
lattices, BLMS 16 (1984),241-263 [2].
H. Jones: see D. J. Howarth.
[Jon5] W. Jonsson, On the Mathieu groups M 22 , M 23 , M24 and the uniqueness
Bibliography 603

of the associated Steiner systems, Math. Z. 125 (1972), 193-214 [11,

141.
[Jull] D. Julin, Two improved block codes, PGIT 11 (1965), 459 [3, 51.
[Jus1] J. Justesen, A class of constructive asymptotically good algebraic codes,
PGIT 18 (1972),652-656 [3].
K

[Kabil G. A. Kabatiansky and V. I. Levenshtein, Bounds for packings on a

sphere and in space, PPI 14 (No.1, 1978), 3-25 - PIT 14 (No.1,
1978),1-17 [1,91.
[Kaci] V. G. Kac, Simple irreducible graded Lie algebras of finite growth,
IANS 32 (1968), 1323-1367 = Izv. 2 (1968), 1271-1311 [1,30].
[Kac2] Infinite-dimensional algebras, Dedekind's 1/ -function, classical
Mobius function and the very strange formula, Adv. in Math. 30
(1978),85-136 [11.
[Kac3] An elucidation of "Infinite-dimensional algebras ... and the very
strange formula." E JI) and the cube root of the modular invariant j,
Adv. in Math. 35 (1980),264-273 [1,301.
[Kac4] - A remark on the Conway-Norton conjecture about the "Monster"
simple group, PNAS 77 (1980) 5048-5049 [1, 30].
[Kac5] - Infinite Dimensional Lie Algebras, Birkhiiuser, Boston, 2nd ed. 1985
[301.
[Kac6] - and D. H. Peterson, Infinite-dimensional Lie algebras, theta
functions and modular forms, Adv. in Math. 53 (1984), 125-264 [11.
P. J. Kachoyan: see I. H. Sloan.
M. Kaikkonen: see I. Honkala.
[Kanil T. Kaneko and H. Miyakawa, A construction algorithm for linear codes
for the Gaussian channel, Elect. Commun. Japan A62 (No.4, 1979),
18-27 [31.
[Kan2] R. Kannan, Improved algorithms for integer programming and related
lattice problems, Proc. 15th ACM Symp. Theory of Computing, 1983,
pp. 193-206 [2].
W. M. Kantor: see also A. R. Calderbank, C. W. Curtis, P. Diaconis.
[Kan3] - Generation of linear groups, in [Dav3], 1981, pp. 497-509 [4].
[Kan4] - Spreads, translation planes and Kerdock sets I, SIAD 3 (J 982),
151-165 [3,91.
[Kan5] - and R. A. Liebler, The rank 3 permutation representations of the
finite classical groups, TAMS 271 (1982), 1-71 [91.
I. M. Kaplinskaja: see E. B. Vinberg.
[Karl] Z. A. Karian, The covering problem for the fields Q (- 1) 1/2 and
Q(-3)112, JNT 8 (1976),233-244 [21.
[Kar2] S. Karlin and J. L. McGregor, The Hahn polynomials, formulas and an
application, Scripta Math. 26 (1961), 33-46 [91.
J. K. Karlof: see also C. P. Downey.
[Kar3] - and C. P. Downey, Odd group codes for the Gaussian channel, SIAJ
34 (1978), 715-716 [31.
M. R. Karpovsky: see G. D. Cohen.
M. Kasahara: see Y. Sugiyama.
[Kas1] T. Kasami and N. Tokura, Some remarks on BCH bounds and
minimum weights of binary primitive BCH codes, PGIT 15 (1969),
408-413 [5].
W. Kassem: see also R. de Buda.
[Kas2] - Optimal Lattice Codes for the Gaussian Channel, Ph.D Dissertation,
604 Bibliography

McMaster Univ., Hamilton, Ontario, 1981 [3].

G. L. Katsman: see also M. A. Tsfasman.
[Katl] - Packing density for spheres of radius I in Hamming space, PPI 17
(No.2, 1981),52-56 = PIT 17 (1981), 114-117 131.
[Kat2] - M. A. Tsfasman and S. G. Vladuts, Modular curves and codes with
a polynomial construction, PGIT 30 (1984), 353-355 [31.
[Kat3] N. M. Katz, An overview of Deligne's proof of the Riemann hypothesis
for varieties over finite fields, PSPM 28 (1976),275-305 [2].
[Kaul] G. Kaur, Extreme quadratic forms for coverings in four variables, Proc.
Nat. Inst. Sci. India A32 (1966),414-417 [21.
[Keal] P. Keast and J. C. Diaz, Fully symmetric integration formulas for the
surface of the sphere in s dimensions, SIAM J. Num. Anal. 20 (1983),
406-419 [11.
K. Keating: see B. K. Teo.
[Kell] O. H. Keller, Geometrie der Zahlen, Springer-Verlag, 1954 [I, 21.
[Ken 11 D. G. Kendall and R. A. Rankin, On the number of points of a given
lattice in a random hypersphere, Quart. J. Math. Oxford 4 (1953),
178-189.
[Ken2] P. Kent, An efficient new way to represent multi-dimensional data,
Computer J. 28 (1985), 184-190.
D. L. Kepert: see B. W. Clarke.
[Kerll A. M. Kerdock, A class of low-rate nonlinear codes, IC 20 (1972),
182-187 [31.
[Ker2] B. W. Kernigham and Shen Lin, Heuristic solution of a signal design
optimization problem, BSTJ 52 (1973), 1145-1159 [31.
[Ker3] R. Kershner, The number of circles covering a set, AJM 61 (1939),
665-671 [21.
[Kibl] R. E. Kibler, Some new constant weight codes, PGIT 26 (1980), 364-
365 [91.
[Kill] K. E. Kilby and N. J. A. Sloane, On the covering radius problem for
codes I: bounds on normalized covering radius, SIAD to appear [31.
[Ki12] - - On the covering radius problem for codes II: codes of low
dimension; normal and abnormal codes, SIAD to appear [31.
[Kinl] K. Kinoshita and M. Kobayashi, Two-dimensionally arrayed optical-
fiber splicing with a CO 2 laser, Appl. Optics 21 (1982),3419-3422 [11.
A. A. Kirilyuk: see P. M. Gudivok.
Y. Kitaoka: see also J. S. Hsia.
[Kitl] - On the relation between the positive definite quadratic forms with
the same representation numbers, Proc. Jap. Acad. 47 (1971), 439-441
[2].
[Kit2] - Positive definite quadratic forms with the same representation
numbers, Archiv Math. 28 (1977), 495-497 [21.
[Kit3] - Representation of positive definite quadratic forms, in [Riel], 1983,
pp. 203-208 [2, 71.
[Kit3a] C. J. Kitrick, Geodesic domes, Structural Topology 11 (1985), 15-20.
[Kit4] C. Kittel, Introduction to Solid State Physics, Wiley, NY, 5th ed. 1976
[1,2,41.
[Klel] P. B. Kleidman and R. A. Wilson, The maximal subgroups of J 4,
PLMS to appear [10].
[KIt/> 11 T. Klt/>ve, A lower bound for A (n,4, w), PGIT 27 (1981), 257-258 [9].
[Knall W. Knapp and P. Schmid, Codes with prescribed permutation group, J.
Alg.67 (1980), 415-435 [3].
[Knell M. Knebusch and W. Scharlau, Algebraic Theory of Quadratic forms,
Bibliography 605

Birkhauser, Boston, 1980 [151.

M. Kneser: see also J. S. Hsia.
[Kne2) - Two remarks on extreme forms, CJM 7 (1955), 145-149 [2,151.
[Kne3) - Klassenzahlen indefinite quadratischer Formen in drei order mehr
Veriinderlichen, Archiv Math. 7 (1956), 323-332 [151.
[Kne4) - Klassenzahlen definiter quadratischer Formen, Archiv Math. 8
(1957),241-250 [2, 7,14,15,17,18,191.
[Kne5) - Lineare Relationen zwischen Darstellungsanzahlen quadratischer
Formen, Math. Ann. 168 (1967), 31-39 [2,4,15).
[Kne6) - Uber die Ausnahme-[somorphismem zwischen endlichen klassischen
Gruppen, ASUH 31 (1967), 136-140 [151.
[Kne7) - Witt's Satz uber quadratische Formen und die Erzeugung
orthogonaler Gruppen durch Spiegelungen, Math.-Phys. Semesterber.
17 (1970), 33-45 [151.
[Kne8) - Representations of integral quadratic forms, in [Riel), 1983,
pp. 159-172 [151.
[Kne9) - Erzeugung ganzzahliger orthogonaler Gruppen durch Spiegelungen,
Math. Ann. 255 (1981), 453-462 [151.
O. Knop: see T. W. Melnyk.
[Knol) M. I. Knopp, Modular Functions in Analytic Number Theory,
Markham, Chicago, 1970 [2, 71.
K. M. Knowles: see J. H. Conway.
[Knull D. E. Knuth, The Art of Computer Programming, Vol. 3, Sorting and
Searching, Addison-Wesley, Reading MA, 1973 [201.
[Kol) c. Ko, Determination of the class number of positive quadratic forms in
nine variables with determinant unity, JLMS 13 (1938), 102-110 [2).
[Ko2) - On the positive definite quadratic forms with determinant unity,
Acta Arith. 3 (1939),79-85 [161.
M. Kobayashi: see K. Kinoshita.
[Kobl) N. Koblitz, p-Adic Numbers, p-Adic Analysis, and Zeta Functions,
Springer-Verlag, 1977 [151.
[Kocll E. Koch, Wirkungsbereichspolyeder und Wirkungsbereichsteilungen zu
kubischen Gitterkomplexen mit weniger also drei Freiheitsgraden,
Z. Krist. 138 (1973), 196-215 [21.
[Koc2) - and W. Fischer, Types of sphere packings for crystallographic point
groups, rod groups and layer groups, Z. Krist. 148 (1978), 107-152 [2).
[Koc3) H. V. Koch, On self-dual, doubly-even codes of length 32, preprint [71.
S. S. Koh: see F. Hirzebruch.
[Koil) M. Koike, Mathieu group M24 and modular forms, Nagoya Math. J. 99
(1985),147-157 [II.
[Konll T. Kondo, The automorphism group of the Leech lattice and elliptiC
modular functions, J. Math. Soc. Japan 37 (1985), 337-361 [II.
[Kon2) - and T. Tasaka, The theta functions of sublattices of the Leech
lattice, Nagoya Math. J. 101 (1986),151-179 [II.
[Kool) T. Koornwinder, The addition formula for Jacobi polynomials and
spherical harmonics, SIAJ 25 (1973),236-246 [91.
[Koo2) - Explicit formulas for special functions related to symmetric spaces,
PSPM 26 (1973), 351-354 [91.
[Koo3) - Positivity proofs for linearization and connection coefficients or
orthogonal polynomials satisfying an addition formula, JLMS 18
(1978),101-114 [9).
[Koo4) - Krawtchouk polynomials, a unification of two different group
theoretic interpretations, SIAM J. Math. Anal. 13 (1982), 1011-1023 [9).
606 Bibliography

[Korl) A. Korkine and G. Zolotareff, Sur les formes quadratique positive

quaternaires, Math. Ann. 5 (1872), 581-583 [1,4].
[Kor2) - - Sur les formes quadratiques, Math. Ann. 6 (1873), 366-389 [4].
[Kor3) - - Sur les formes quadratique positives, Math. Ann. 11 (1877),
242-292 [I, 4].
L. Kovacs: see Z. Janko.
[Kral) E. S. Kramer, S. S. Magliveras and D. M. Mesner, t-Designs from the
large Mathieu groups, Discr. Math. 36 (1981),171-189 [Ill.
[Kra2) M. Kramer, Sphiirische Untergruppen in Kompakten Zusammen-
hiingenden Liegruppen, Compos. Math. 38 (1979), 129-153 [9].
[Kra3) A. Krazer, Lehrbuch der Thetafunktionen, Chelsea, NY, 1970 [4].
[Krell M. G. Krein, Hermitian-positive kernels on homogeneous spaces,
Ukranian Math. J. (No.4, 1949),64-98; (No. I, 1950), 10-59 = Amer.
Math. Soc. Trans!. (2) 34 (1963),69-108, 109-164 [9].
[Kril) H. A. Krieger and C. A. Schaffner, The global optimization of
incoherent phase signals, SIAJ 21 (1971), 573-589 [3].
J. B. Kruskal: see L. A. Shepp.
A. L. Krylov: see V. I. Arnold.
H. W. Kuhn: see R. J. Duffin.

[Lacl] G. Lachaud, Les codes geometriques de Goppa, Sem. Bourbaki, No.

641,1985 [3].
[Lagl) J. C. Lagarias, On the computational complexity of determining the
solvability or unsolvability of the equation X 2-Dy2=- L TAMS 260
(1980),485-508 [15].
[Lag2] Worst-case complexity bounds for algorithms in the theory of
integral quadratic forms, J. Algor. 1 (1980), 142-186 [15].
[Lag3) - H. W. Lenstra, Jr., and C. P. Schnorr, Korkine-Zolotarev bases and
successive minima of a lattice and its reciprocal lattice, Combinatorica
to appear [2].
[Lag4) and A. M. Odlyzko, Solving low-density subset sum problems,
JACM 32 (1985), 229-246 [2].
[Laml) T. Y. Lam, The Algebraic Theory of Quadratic Forms, Benjamin,
Reading MA, 1973 [15].
[Lam2) P. J. C. Lamont, The number of Cayley integers of given norm, Proc.
Edinburgh Math. Soc. 25 (1982), 101-103.
[LanO] E. Landau, Vorlesungen uber Zahlentheorie, Chelsea, NY, 1969 [2].
[Lanll H. J. Landau, Sampling, data transmission, and the Nyquist rate,
PIEEE 55 (1967), 1701-1706 [3].
[Lan2) - How does a porcupine separate its quills? PGIT 17 (1971), 157-
161 [2).
[Lan3] and H. O. Pollak, Prolate spheroidal wave functions, Fourier
analysis and uncertainty II, BSTJ 40 (1961),65-84 [3].
[Lan4] - - Prolate spheroided wave functions, Fourier analysis and
uncertainty III: the dimension of the space of essentially time- and
band-limited signals, BSTJ 41 (1962), 1295-1336 (3).
[Lan5] E. S. Lander, Symmetric Designs: An Algebraic Approach, Camb.
Univ. Press, 1983 [3].
G. R. Lang: see G. D. Forney, Jr.
[Lan7] S. Lang, Algebraic Number Theory, Addison-Wesley, Reading MA,
1970 [8].
Bibliography 607

[Lan8) Algebra, Addison-Wesley, Reading MA, revised 1971 [81.

[Lan9) Introduction to Modular Forms, Springer-Verlag, 1976 [2,71.
[Larl) J. Larmouth, The enumeration of perfect forms, in Computers in
Number Theory, ed. A.O.L. Atkin and B. J. Birch, Ac. Press, NY, 1971,
pp. 237-239 [1, 21.
V. B. Lawrence: see A. Gersho.
[Lazl) D. E. Lazic, Class of block codes for the Gaussian channel, Elect. Lett.
16 (1980), 185-186131.
D. W. Leavitt: see S. S. Magliveras.
A. Lebowitz: see H. E. Rauch.
[LeeO) A. J. Lee, A note on the Campbell sampling theorem, SIAJ 41 (1981),
553-557 [31.
[Leel) D. T. Lee and F. P. Preparata, Computational geometry - a survey,
IEEE Trans. Compo 33 (1984),1072-1101; 34 (1985), 584 [21.
R. C. T. Lee: see R. C. Chang.
S. L. Lee: see G. R. Welti.
T.-A. Lee: see A. R. Calderbank.
[Lee2) J. Leech, The problem of the thirteen spheres, Math. Gazette 40
(1956), 22-23 !ll.
[Lee3) - Equilibrium of sets of particles on a sphere, Math. Gazette 41
(1957), 81-90 [11.
[Lee4) - Some sphere packings in higher space, CJM 16 (1964), 657-682 [1,
5,61.
[Lee5) - Notes on sphere packings, CJM 19 (1967),251-267 [1,4,5,6,7).
[Lee6) - Five-dimensional nonlattice sphere packings, CMB 10 (1967), 387-
393 [5,6).
[Lee7) Six and seven dimensional nonlattice sphere pac kings, CMB 12
(1969), 151-155 [5,61.
[Lee8) and N. J. A. Sloane, New sphere packings in dimensions 9-15,
BAMS 76 (1970), 1006-1010 [51.
[Lee9) - - New sphere packings in more than 32 dimensions, Proc. 2nd
Chapel Hill Conf. on Comb. Math. Appl., Univ. North Carolina,
Chapel Hill, 1970, pp. 345-355 (5).
[Leel0) Sphere packing and error-correcting codes, CJM 23 (1971),
718-745 == Chap. 5 of this book [1, 5, 6,131.
[Leell) - and T. Tarnai, Arrangements of 22 circles on a sphere, Ann. Univ.
Sci. Budapest Ser. Math. to appear [11.
[Leg 11 A.-M. Legendre, Theorie des Nombres, 4-th edition, Blanchard, Paris,
1955,2 vols. !l51.
[Lehl) D. H. Lehmer, Guide to Tables in the Theory of Numbers, Bull. Nat.
Res. Council 105, Nat. Acad. Sci., Washington DC, 1941 [15].
[Leh2) - Ramanujan's function dn), DMJ 10 (1943), 483-492 [4].
C. G. Lekkerkerker: see also P. M. Gruber.
[Lekl) - Geometry of Numbers, Wolters-Noordhoff, Groningen, and North-
Holland, Amsterdam, 1969 (see also [Grula)) !l, 8, B).
A. Lempel: see J. Ziv.
[Leml) W. Lempken, Die Untergruppenstruktur der endlichen. einfachen
Gruppe J 4 , Dissertation, Univ. Mainz, 1985 [10).
[Lenl) A. K. Lenstra, H. W. Lenstra, Jr., and L. Lovasz, Factoring polynomials
with rational coefficients, Math. Ann. 261 (1982), 515-534 [2].
H. W. Lenstra, Jr.: see also J. C. Lagarias, A. K. Lenstra,
C. Pomerance.
[Len2) - Codes from algebraic number fields, in Fundamental Contributions
608 Bibliography

in the Netherlands since 1945, North-Holland, Amsterdam, II (1986),

pp. 95-104 [31.
J. S. Leon: see also N. Ito.
[Leo 11 An algorithm for computing the automorphism group of a
Hadamard matrix, JCT A27 (1979), 289-306 [31.
[Leo2] - On an algorithm for finding a base and a strong generating set for a
group given by generating permutations, MTAC 35 (1980), 941-974.
[Leo3] - Computing automorphism groups of error-correcting codes, PGIT
28 (1982), 496-511.
[Leo4] Computing automorphism groups of combinatorial objects, in
[Atk21, 1984, pp. 321-335.
[Leo5] - J. M. Masley and V. Pless, Duadic codes, PGIT 30 (1984), 709-714
[31.
[Leo6] - - - On weights in duadic codes, JCT A44 (1987),6-21 [31.
[Leo7] - V. Pless and N. J. A. Sloane, Self-dual codes over GF(5), JCT A32
(1982),178-194 [4,71.
[Leo8] - - - On ternary self-dual codes of length 24, PGIT 27 (1981),
176-180 [4,71.
[Leo9] - and C. C. Sims, The existence and uniqueness of a simple group
generated by {3,4}-transpositions, BAMS 83 (1977), 1039-1040 [101.
[Leol0] D. A. Leonard, Parameters of association schemes that are both P- and
Q-polynomial, JCT A36 (1984), 355-363 [91.
[Leoll] - Discrete orthogonal polynomials, duality, and association schemes,
SIAM J. Math Anal. 13 (1982), 656-663 [91.
J. I. Lepowsky: see also I. B. Frenkel.
[Lepl] S. Mandelstam and I. M. Singer, editors, Vertex Operators in
Mathematics and Physics, Springer-Verlag, 1985 [BI.
[Lep2] - and A. E. Meurman, An Es-approach to the Leech lattice and the
Conway group, J. Alg. 77 (1982), 484-504 [4, 81.
[Letl] G. Letac, Probleines c1assiques de probabilite'sur un couple de Gelfand,
LNM 861 (1981),93-120 [91.
[Let2] - Les fonctions sphlriques d'un couple de Gelfand symetrique et les
chaines de Markov, Adv. Appl. Prob. 14 (1982), 272-294 [91.
V. I. Levenshtein: see also G. A. Kabatiansky.
[Levi] The application of Hadamard matrices to a problem in coding,
Problemy Kibernetiki 5 (1961), 123-136 = Problems of Cybernetics 5
(1964),166-184 [3,91.
[Lev2] - On upper bounds for codes with codewords of a fixed weight, PPI 7
(1971, No.4), 3-12 = PIT 7 (1971),281-287 [91.
[Lev3] - On the minimal redundancy of binary error-correcting codes, PPI
10 (1974, No.2), 26-42 = IC 28 (1975), 268-291 [91.
[Lev4] Maximal packing density of equal spheres in n-dimensional
Euclidean space, Mat. Zametki 18 (1975, No.2), 301-311 - Math.
Notes Acad. Sci. USSR 18 (1975),765-771 [1,91.
[Lev5] - Methods for obtaining bounds in metric problems of coding theory,
in Proc. 1975 IEEE-USSR Joint Workshop on Information Theory,
IEEE, NY, 1976, pp. 126-144 [1,91.
[Lev6] - On choosing polynomials to obtain bounds in packing problems, in
Proc. Seventh AI/-Union Con! Coding Theory and Information
Transmission, Moscow, II, 1978, pp. 103-108 [91.
[Lev71 On bounds for packings in n-dimensional Euclidean space, DAN
245 (1979),1299-1303 = SMD 20 (1979),417-421 [1,91.
[Lev81 Bounds on the maximal cardinality of a code with bounded
Bibliography 609

modulus of the inner product, DAN 263 (1982) 1303-1308 = SMD 25

(1982),526-531 [9].
[Lev9] Bounds for packings in metric spaces and certain applications,
Problemy Kibernetiki, 40 (1983),44-110 [1,91.
[LevIO] D. Levine and P. J. Steinhardt, Quasicrystals: a new class of ordered
structures, Phys. Rev. Lett. 53 (1984),2477-2480 [11.
[Levll] Quasicrystals I: definition and structure, Phys. Rev. 834
(1986),596-616 [11.
[LeVll W. J. LeVeque, Topics in Number Theory, Addison-Wesley, Reading
MA, 2 vols., 1956 [8, 151.
[LeV2] - editor, Reviews in Number Theory, Amer. Math. Soc., Providence
RI, 6 vols., 1974 [BI.
R. Levingston: see L. M. Gordon.
[Lewll J. Lewittes, Construction of modular forms, Israel J. Math. 24 (1976),
145-183 [2,7].
[Lil] D. Li, Indefinite binary forms representing the same numbers, PCPS 92
(1982), 29-33 [21.
R. A. Liebler: see W. M. Kantor.
Shen Lin: see B. W. Kernighan.
[LinO] Shu Lin and D. J. Costello, Jr., Error Control Coding, Prentice-Hall,
Englewood Cliffs NJ, 1983 [31.
[Linll Y. C. Lin and D. E. Williams, Calculated geometry for ten-
coordination, Canad. J. Chern. 51 (1973),312-316 [11.
[Lin2] Y. Linde, A. Buzo and R. M. Gray, An algorithm for vector quantizer
design, COMM 28 (1980),84-95 [21.
[Lin3] C. C. Lindner and A. Rosa, editors, Topics on Steiner Systems, North-
Holland, Amsterdam, 1980 [3].
[Lin4] J. H. Lindsey II, On a projective representation of the Hall-Janko
group, BAMS 74 (1968), 1094 [101.
[Lin5] - Linear groups of degree 6 and the Hall-Janko group, in [Bra2],
1969, pp. 97-100 [101.
[Lin6] - A correlation between PSU 4 (3), the Suzuki group, and the Conway
group, TAMS 157 (1971), 189-204 [41.
[Lin7] Finite linear groups of degree 6, CJM 5 (1971), 771-790 [41.
[Lin8] On the Suzuki and Conway groups, PSPM 21 (1971), 107-109 [41.
[Lin9] Sphere-packing in R 3, Math. 33 (1986), 137-147 [I].
[Lin9a] Sphere-packing II, preprint [11.
[Lin9b] L. Lines, Solid Geometry with Chapters on Space-Lattices, Sphere-
Packs and Crystals, Macmillan, London, 1935; Dover, NY 1965.
J. H. van Lint: see also P. J. Cameron, N. J. A. Sloane.
[Linl0] Combinational Theory Seminar, Eindhoven University of
Technology, LNM 382, 1974 [31.
[Linll] - Introduction to Coding Theory, Springer-Verlag, 1982 [31.
[LinI2] and A. Schrijver, Construction of strongly regular graphs, two-
weight codes and partial geometries by finite fields, Combinatorica 1
(1981),63-73 [9].
[LinI3] - and R. M. Wilson, On the minimum distance of cyclic codes, PGIT
32 (1986), 23-40 [3].
R. J. Lipton: see D. Dobkin.
S. N. Litsyn: see also V. A. Zinoviev.
[Litl] - New non-linear codes with a minimum distance of 3, Problems of
Control and Inform. Theory 13 (1984), 13-15 [3, 51.
[Lit2] and O. I. Shekhovtsov, Fast decoding algorithms for first-order
610 Bibliography

Reed-Muller codes, PPI 19 (No.2, 1983), 3-7 = PIT 19 (1983), 87-91

[20).
[Lit3] and M. A. Tsfasman, Algebraic-geometric and number-theoretic
packings of spheres, Uspekhi Mat. Nauk 40 (1985), 185-186 [1, 8).
[Lit4] - - A note on lower bounds, PGIT 32 (1986), 705-706 [3].
[Lit5] - - Constructive high-dimensional sphere packings, DMJ 54 (1987),
147-161 [I].
[LjuI] W. Ljunggren, New solution of a problem proposed by E. Lucas, Norsk
Mat. Tid. 34 (1952),65-72 [26).
[LloI] S. P. Lloyd, Hamming association schemes and codes on spheres, SIAM
J. Math. Anal. 11 (1980),488-505 [9].
A. C. Lobstein: see G. D. Cohen.
[LoeI] A. L. Loeb, Space Structures: Their Harmony and Counterpoint,
Addison-Wesley, Reading MA, 1976 [2, 211.
G. Longo: see J.-M. Goethals.
F. M. Longstaff: see G. D. Forney, Jr.
M. S. Longuet-Higgins: see H. S. M. Coxeter.
J. Q. Longyear: see also N. Ito.
[Lonl] Criteria for a Hadamard matrix to be skew-equivalent, CJM 28
(1976), 1216-1223 [3,8].
[Lon2] There are nine different skew-symmetric Hadamard matrices of
order 24, Ars Combin. 10 (1980), 115-122 (3).
[LooI] L. H. Loomis, An Introduction to Abstract Harmonic Analysis, Van
Nostrand, Princeton NJ, 1953 [4).
L. Lovasz: see also A. K. Lenstra.
[LovI] - On the Shannon capacity of a graph, PGIT 2S (1979), 1-7 [9).
W. E. Love: see A. L. Patterson.
[Lubl] A. Lubotzky, R. Phillips and P. Sarnak, Hecke operators and
distributing points on the sphere I, Comm. Pure Appl. Math. 39 (1986),
S149-S186 (1).
[Lub2] - - - II: Comm. Pure Appl. Math. 40 (1987),401-420 [I).
[LiinI] H. Liineburg, Transitive Erweiterungen endlicher Permutationsgruppen
LNM 84, 1969 [12, 14).
[LyoI] R. Lyons, Evidence for a new finite simple group, J. Alg. 20 (1972),
540-569 [10).
M
[MacO] A. L. Mackay, The packing of three-dimensional spheres on the surface
of a four-dimensional hypersphere, J. Phys. A13 (1980), 3373-3379 [1,
9].
[MacI] G. W. Mackey, Unitary Group Representations in Physics, Probability
and Number Theory, Benjamin, Reading MA 1978 [9).
S. MacLane: see G. Birkhoff.
F. J. MacWilliams: see also E. R. Berlekamp, M. R. Best.
[Mac2] A theorem on the distribution of weights in a systematic code,
BSTJ 42 (1963),79-84 [3, 9).
[Mac3] - C. L. Mallows and N. J. A. Sloane, Generalizations of Gleason's
theorem on weight enumerators of self-dual codes, PGIT 18 (1972),
794-805 [7].
[Mac4] - A. M. Odlyzko, N. J. A. Sloane and H. N. Ward, Self-dual codes
over GF(4), JCT A2S (1978), 288-318 [7, 16, 18).
[Mac5] - and N. J. A. Sloane, Pseudo-random sequences and arrays, PIEEE
64 (1976),1715-1729.
[Mac6] The Theory of Error-Correcting Codes, North Holland,
Bibliography 611

[Mac6] The Theory of Error-Correcting Codes. North Holland,

Amsterdam, 2nd printing, 1978; Russian translation, Svyaz, Moscow,
1979 [2-5, 7-9, 11, 12, 14, 16,20, B].
[Mac7] - - and 1.-M. Goethals, The MacWilliams identities for nonlinear
codes. BSTJ 51 (1972),803-819 [3].
[Mac8] - - and 1. G. Thompson, Good self-dual codes exist. Discr. Math. 3
(1972), 153-162 [7, 16].
S. S. Magliveras: see also E. S. Kramer.
[Magi] and D. W. Leavitt, Simple six-designs exist. in Proc. 14th
Southeast. Con!. Boca Raton. Congr. Num. 40 (1983), 195-205 [3].
[Mag2] - - Simple 6-(33,8, 36) designs from prL 2 (32), in [Atk2], 1984,
pp. 337-352 [3].
[Mag3] W. Magnus, Uber die Anzahl der in einem Geschlecht enthaltenen
Klassen von positiv dejiniten quadratischen Formen. Math. Ann. 114
(1937), 465-475; 115 (1938), 643-644 [16].
[Mahll D. P. Maher, Self-Orthogonal Codes and Modular Forms. Ph.D.
Dissertation, Lehigh Univ., Bethlehem PA, 1976 [7].
[Mah2] - Lee polynomials of codes and theta functions of lattices. CIM 30
(1978),738-747 [7].
[Mah3] - Modular forms from codes. CIM 32 (1980),40-58 [7].
K. Mahler: see also B. Segre.
[Mah4] - p-Adic Numbers and their Functions. Camb. Univ. Press, 2nd ed.
1981 [IS].
C. L. Mallows: see also F. 1. MacWilliams.
[Mall] - A. M. Odlyzko and N. 1. A. Sloane, Upper bounds for modular
forms. lattices. and codes. 1. Alg. 36 (1975),68-76 [7, 19].
[MaI2] - V. Pless and N. 1. A. Sloane, Self-dual codes over GF(3). SIAl 31
(1976),649-666 [3, 5, 7, 16].
[Mal3] - and N. 1. A. Sloane, An upper bound for self-dual codes. IC 22
(1973), 188-200 [7].
[MaI4] Weight enumerators of self-orthogonal codes. Discr. Math. 9
(1974),391-400 [7].
[MaI5] - - Weight enumerators of self-orthogonal codes over GF(3). SIAD
2 (1981), 452-480 [7].
[MaI6] A. V. Malysev, On the representation of integers by positive quadratic
forms with four or more variables I. Izv. 23 (1959), 337-364 - Amer.
Math. Soc. Transl. (2) 46 (1965),17-47 [4].
S. Madelstam: see 1. I. Lepowsky.
Y. I. Manin: see also S. G. Vladuts.
[Manll - What is the maximum number of points on a curve over F2? 1. Fac.
Sci. Univ. Tokyo, Sec. lA, 28 (1982),715-720 [3].
H. B. Mann: see also E. C. Posner.
[Man2] - On the number of information symbols in Bose-Chaudhuri codes.
IC 5 (1962), 153-162 [5].
[Marl] R. P. Martineau, A characterization of Janko's simple group of order
175,560, PLMS 19 (1969),709-729 [10].
[Mar2] 1. Martinet, Tours de corps de classes et estimations de discriminants.
Invent. Math. 44 (1978), 65-73 [I, 8].
[Mar3] - Petits discriminants. Ann. Inst. Fourier 29 (1979), 159-170 [8].
1. M. Masley: see 1. S. Leon.
[Masll D. R. Mason, On the construction of the Steiner system S(5,8,24). 1.
Alg.47 (1977),77-79 [11].
[Mas2] G. Mason, Modular forms and the theory of Thompson series. in Proc.
612 Bibliography

Rutgers Croup Theory Year 1983-1984, ed. M. Aschbacher et a!.,

Camb. Univ. Press, 1984, pp. 391-407 [11.
[Mas3] - M24 and certain automorphic forms, in [McK31, 1985, 223-244 [11.
[Mati] E. Mathieu, Me'moire sur l'e'tude des fonctions de plusieurs quantite's, J.
Math. Pures Appl. 6 (1861), 241-243 [III.
[Mall] - Sur la fonction cinq fois transitive de 24 quantile's, J. Math. Pures
Appl. 18 (1873), 25-46 [111.
[Mat3] Mathlab Group, MACSYMA Reference Manual, Version 10, Laboratory
for Computer Science, MIT, Cambridge MA, 1983 [16, 19, 211.
H. F. Mattson, Jr.: see E. F. Assmus, Jr., G. D. Cohen.
[Maull A. Maus, Delaunay triangulation and the convex hull of n points in
expected linear time, BIT 24 (1984), 151-163 [2].
V. C. Mavron: see D. H. Smith.
[Maxi] J. Max, Quantizingfor minimum distortion, PGIT 6 (1960), 7-12 [21.
C. A. May: see B. S. Atal, S. G. Wilson.
J. E. Mazo: see also A. R. Calderbank.
[Mazl] - Some theoretical observations on spread-spectrum communications,
BSTJ 58 (1979),2013-2023 [11.
T. P. McDonough: see D. H. Smith.
R. J. McEliece: see also E. R. Berlekamp.
[McEI] The Theory of Information and Coding, Addison-Wesley, Reading
MA, 1977 [3,91.
[McE2] - The bounds of Delsarte and Lovdsz, and their application to coding
theory, in Algebraic Coding Theory and Applications, ed. G. Longo,
CISM Courses and Lectures No. 258, Springer-Verlag, 1979, pp. 107-
178 [91.
[McE3] E. R. Rodemich and H. C. Rumsey, Jr., The Lovdsz bound and
some generalizations, J. Combinatorics, Inform. Syst. Sci. 3 (1978),
134-152 [91.
[McE4] - - - and L. R. Welch, New upper bounds on the rate of a code
via the Delsarte-MacWilliams inequalities, PGIT 23 (J977), 157-166
[7,91.
J. L. McGregor: see S. Karlin.
J. A. McInnes: see M. R. Hoare.
[McKl] B. D. McKay, Transitive graphs with fewer than twenty vertices. MTAC
33 (1979), 1101-1121 [9].
J. McKay: see also G. Higman.
[McK2] - A setting for the Leech lattice, in [Gag2], 1973, pp. 117-118 [7,8].
[McK3] - editor, Finite Croups - Coming of Age, Contemp. Math. 45 (1985),
1-350 [BI.
H. P. McKean: see H. Dym.
[McLO] A. D. McLaren, Optimal numerical integration on a sphere, MTAC 17
(J 963), 361-383 [11.
[McLl] J. McLaughlin, A simple group of order 898,128,000, in [Bra21, 1969,
pp. 109-111 [101.
[McL2] A. M. McLoughlin, The complexity of computing the covering radius of
a code, PGIT 30 (1984),800-804 [21.
[McMl] P. McMullen, Convex bodies which tile space by translation, Math. 27
(1980),113-121; 28 (1980, 191 [211.
[McM2] and G. C. Shephard, Convex Polytopes and the Upper Bound
Conjecture, Camb. Univ. Press, 1971 [211.
[Mell] T. W. Melnyk, O. Knop and W. R. Smith, Extremal arrangements of
points and unit changes on a sphere: equilibrium configurations
Bibliography 613

revisited, Can. J. Chern. 55 (1977), 1745·1761 [11.

D. M. Mesner: see E. S. Kramer.
A. E. Meurman: see I. B. Frenkel, J. I. Lepowsky.
[Meyl] B. Meyer, On the symmetries of spherical harmonics, CJM 6 (1954),
135·157 [1].
[Mey2] J. Meyer, Priisentation der Einheitengruppe der quadratischen Form
FU() = - XJ + Xr + ... + Xrs, Archiv Math. 29 (1977), 261·266 [27,
281.
[Mey3] W. Meyer, W. Neutsch and R. A. Parker, The minimal 5·representation
of Lyons' sporadic group, Math. Ann. 272 (1985),29·39 [101.
L. Michel: see D. Gratias.
D. Middleton: see D. P. Petersen.
A. D. Mighell: see A. Santoro.
[Mill] E. Y. Miller and S. Washburn, A sublattice of the Leech lattice, JCT
A42 (1986),9·14 [41.
[Mil2] G. A. Miller, H. F. Blichfeldt and L. E. Dickson, Theory and
Applications of Finite Groups, Wiley, NY, 1916; Dover, NY, 1961 [7].
J. C. P. Miller: see H. S. M. Coxeter.
[Mil3] W. Miller, Jr., Symmetry Groups and Their Applications, Ac. Press,
NY, 1972 [91.
[MiI4] W. H. Mills, A new 5·design, Ars Combin. 6 (1978), 193·195 [31.
[MiI5] J. Milnor, Hilbert's Problem 18: on crystallographic groups,
fundamental domains, and on sphere packing, PSPM 28 (J976), 491·
506 [I).
[Mil6] - Hyperbolic geometry: the first 150 years, BAMS 6 (1982), 9·24 -
PSPM 39 (1983, Part I), 25·40 [26].
[MiI7] and D. Husemoller, Symmetric Bilinear Forms, Springer· Verlag,
1973 [1,7, 15, 16,261.
[MimI] Y. Mimura, On 2·lattices over real quadratic integers, Math. Sem.
Notes (Kobe Univ.) 7 (1959),327·342.
D. M. P. Mingos: see C. E. Briant.
[MinO] H. Minkowski, Grundlagen fur eine Theorie der quadratischen Formen
mit ganzzahligen KoefJizienten, Mem. pres. par divers savants a
I'Academie des Sci. Inst. nat. de France, 29 (No.2, 1884) - Ges. Abh.
1(1910,3·144 [4,15).
[MinI] - Sur la reduction des formes quadratiques positives quaternaires,
CR 96 (1883), 1205 - Ges. Abh. I (1910, 195 [2, 15].
[Min2] - Uber positive quadratische Formen, JRAM 99 (1886), 1·9 - Ges.
Abh. 1(1911),149·156 [15].
[Min3) Zur Theorie der positiven quadratischen Formen, JRAM 101
(1887), 196·202 = Ges. Abh. I (1910,212·218 [15].
[Min4] - Dichteste gitterformige Lagerung kongruenter Korper, Nachr. Ges.
Wiss. Gottingen (1904), 311·355 = Ges. Abh. II (1911),3·42 III.
[Min5] Geometrie der Zahlen, Teubner, Leipzig, 1910 [ll.
[Min6] - Diophantische Approximationen, Teubner, Leipzig, 1927 [ll.
[Mitl] H. H. Mitchell, Determination of the ordinary and modular ternary
linear groups, TAMS 12 (1910,207·242 [4).
[Mit2] Determination of the finite quaternary linear groups, TAMS 14
(J913), 123·142 [4].
[Mit3] - Determination of all primitive collineation groups in more than four
variables which contain homologies, AJM 36 (J914), 1·12 [4].
[Mit4] - The subgroups of the quaternary abelian linear groups, TAMS 15
(1914),379·396 [41.
614 Bibliography

[Mit5] W. C. Mitchell, The number of lattice points in a k-dimensional

hypersphere, MTAC 20 (1966), 300-310 [41.
[Mit6] D. S. Mitrinovic, Analytic Inequalities, Springer-Verlag, 1970 [3].
H. Miyakawa: see T. Kaneko.
[ModI] Modular Functions of One Variable, various editors, LNM 320 (1973),
349 (1973),350 (1973),476 (1975),601 (1977),627 (1977) [71.
[MohI] R. Mohr and R. Bajcsy, Packing volumes by spheres, IEEE Trans. Patt.
Anal. Mach. Int. 5 (1983), 111-116 [11.
J. Molk: see J. Tannery.
[Moo I] R. V. Moody, Euclidean Lie algebras, CJM 21 (1969), 1432-1454 [301.
[Moo2] - Root systems of hyperbolic type, Adv. in Math. 33 (1979), 144-160
[301.
[Moo3] E. H. Moore, Using the group of a code to compute its minimum
weight, preprint [71.
[MorI] L. J. Mordell, On Mr. Ramanujan's empirical expansions of modular
functions, PCPS 19 (1919),117-124 [71.
[Mor2] - On the representations of numbers as a sum of 2r squares, Quart.
J. Math. 48 (1920),93-104 [4].
[Mor3] The definite quadratic forms in eight variables with determinant
unity, J. Math. Pures Appl. 17 (1938),41-46 [21.
[Mor4] - Observation on the minimum of a positive definite quadratic form
in eight variables, JLMS 19 (1944), 3-6 [1, 6].
[Mor5] The minimum of a definite ternary quadratic form, JLMS 23
(1948),175-178 [11.
[Mor6] - Diophantine Equations, Ac. Press, NY, 1969 [2, 15,26].
[Mor7] C. J. Moreno, Algebraic Curves over Finite Fields with Applications to
Coding Theory, to appear [31.
[Mor8] M. A. Morrison and J. Brillhart, A method of factoring and the
factorization of F 7, MTAC 29 (1975), 183-205 [151.
W. O. J. Moser: see H. S. M. Coxeter.
[MudI] D. J. Muder, Putting the best face on a Voronoi polyhedron, PLMS,
submitted [ll
[Muel] E. L. Muetterties and C. M. Wright, Molecular polyhedra of high
coordination number, Quart. Rev. Chern. Soc. 21 (1967), 109-194 [ll
[Mull ] C. Miiller, Spherical Harmonics, LNM 17, 1966 [91.
R. C. Mullin: see I. F. Blake.
[MumI] D. Mumford, Tata Lectures on Theta, Birkhauser, Boston, 2 vols., 1983,
1984 [4].
N

T. Namekawa: see Y. Sugiyama.

[Narl] W. Narkiewicz, Elementary and Analytical Theory of Algebraic
Numbers, Panst. Wyd. Nauk., Warsaw, 1973 [8].
[Nell] D. R. Nelson and B. I. Halperin, Pentagonal and icosahedral order in
rapidly cooled metals, Science 229 (1985), 233-238 [I].
R. W. Nettleton: see G. R. Cooper.
J. Neubiiser: see H. Brown, R. Biilow.
[NeuI] A. Neumaier, Distances, graphs and designs, Europ. J. Combin. 1
(1980),163-174 [9].
[Neu2] - Combinational configurations in terms of distances, Memorandum
81-07, Math. Dept., Eindhoven Univ. Technology, 1981 [3].
[Neu3] - Lattices of simplex type, SIAD 4 (1983),145-160 [261.
Bibliography 615

[Neu4) - On norm three vectors in integral Euclidean lattices I, Math. Z. 183

(1983),565-574.
[Neu5) - and J. J. Seidel, Discrete hyperbolic geometry, Combinatorica 3
(1983),219-237 [26).
B. Neumann: see Z. Janko.
W. D. Neumann: see F. Hirzebruch.
W. Neutsch: see W. Meyer.
D. J. Newman: see also L. Flatto.
[New II - The hexagon theorem, PGIT 28 (1982), 137-139 [2).
[New2) M. Newman, A table of the coefficients of the power of." (r), KNA W
AS9 (1956),204-216 [41.
[New3) - Integral Matrices, Ac. Press, NY, 1972.
[Nie1) H. Niederreiter, Quasi-Monte Carlo methods and pseudo-random
numbers, BAMS 84 (1978),957-1041 [II.
[Nie2) H.-V. Niemeier, Definite quadratische Formen der Dimension 24 und
Diskriminante I, JNT S (1973), 142-178 [2,4,7,12,14-18,23).
J. Nievergelt: see E. M. Reingold.
[Nikl) V. V. Nikulin, On quotient groups of the automorphism groups of
hyperbolic forms modulo subgroups generated by 2-reflections, DAN
248 (1979), 1307-1309 - SMD 20 (1979), 1156-1158 [15).
[Nik2) - Integral symmetric bilinear forms and some of their applications,
IANS 43 (1979) 111-177 - Izv. 14 (1980),103-167 [2, 15).
[Nik3) On arithmetic groups generated by reflections in Lobachevsky
spaces, IANS 44 (1980),637-669 - Izv. 16 (1981), 573-601 U51.
[Nik4) - On classification of arithmetic groups generated by reflections in
Lobachevsky spaces, IANS 4S (1981) 113-142 - Izv. 18 (1982), 99-
123 [151.
[Nik5) - Quotient groups of groups of automorphisms of hyperbolic forms
with respect to subgroups generated by 2-reflections. Algebraic-
geometric applications, Itogi Nauk. Tekhniki, Ser. SOy. Prob. Mat. 18
(1981), 3-114 = JSM 22 (1983), 1401-1475 (5).
[Nik6] - Involutions of integral quadratic forms and their applicatiOns to
real algebraic geometry, IANS 47 (1983), 109-158 - Izv. 22 (1984),
99-172 U51.
[Norl) A. W. Nordstrom and J. P. Robinson, An optimum nonlinear code, IC
11 (1967), 613-616 [31.
S. P. Norton: see also J. H. Conway.
[Nor2] Construction of Harada's groups, Ph.D. Dissertation, Univ. of
Cambridge, 1976 [101.
[NorJ] - The construction of h PSPM 37 (1980), 271-277 UO,l11.
[Nor4) - A bound for the covering radius of the Leech lattice, PRS A380
(1982),259-260;;;; Chap. 22 of this book [2, 221.
[Nor5) - More on moonshine, in [Atk2), 1984, pp. 185-193 U, 29).
[Nor6) - The uniqueness of the Fischer-Griess monster, in [McK3), 1985, pp.
271-185 [291.
[Nor7) - personal communication [291.
[Nor8) and R. A. Wilson, Maximal subgroups of the Harada-Norton
group, J. Alg. 103 (1986), 362-376 [101.
[Novl] N. V. Novikova, Korkine-Zolotarev reduction domains for positive
quadratic forms in n ~ 8 variables and reduction algorithms for these
domains, DAN 270 (1983),48-51 - SMD 27 (1983),557-560 [2, 15).
[Nowl) W. Nowacki, Bibliography of Mathematical Crystallography, Inter.
Union Cryst., 1978.
616 Bibliography

[Nunl] R. de M. Nunes Mendes, Symmetries of spherical harmonics, TAMS

204 (1975),161-178 [ll.
o
[O'Co I] R. E. O'Connor and G. Pall, The construction of integral quadratic
forms of determinant I, DMJ 11 (1944),319-331 [4,61.
A. M. Odlyzko: see also M. R. Best, 1. H. Conway, J. C. Lagarias, F. 1.
MacWilliams, C. L. Mallows.
[Odlll Some analytic estimates of class numbers and discriminants,
Invent. Math. 29 (1975), 275-286 [81.
[OdI2] - Lower bounds for discriminants of number fields, Acta Arith. 29
(1976),275-297 [81.
[Od13] - Lower bounds for discriminants of number fields II, Tohoku Math.
J. 29 (1977), 209-216 [81.
[OdI4] - Cryptanalytic attacks on the mUltiplicative knapsack cryptosystem
and on Shamir's fast signature scheme, PGIT 30 (1984), 594-601 [21.
[OdI5] - and N. J. A. Sloane, New bounds on the number of unit spheres
that can touch a unit sphere in n dimensions, JCT A26 (1979), 210-214
== Chap. 13 of this book [I, 9, 13].
[OdI6] A theta-function identity for nonlattice packings, Studia Sci.
Math. Hung. 15 (1980), 461-465 [2,71.
[Oggl] A. Ogg, Modular Forms and Dirichlet Series, Benjamin, NY 1969 [2, 7,
8, 181.
[O'Kel] M.O'Keefl'e and B. G. Hyde, Plane nets in crystal chemistry, Phil.
Trans. Roy. Soc. London A295 (1980), 553-623 [11.
D. Olive: see P. Goddard.
[Olil] B. M. Oliver, J. R. Pierce and C. E. Shannon, The philosophy of PCM,
PIEEE 36 (1948),1324-1331 [31.
[Oli2] R. K. Oliver, On Bieberbach's analysis of discrete Euclidean groups,
PAMS 80 (1980),15-21 [41.
[Olvl] F. W. J. Olver, Bessel Functions III: Zeros and Associated Values,
Royal Soc. Math. Tables 7, Camb. Univ. Press, 1960 11 1.
[O'Mel] O. T. O'Meara, Introduction to Quadratic Forms, Springer-Verlag, 1971
[2, IS, 161.
[O'Me2] The construction of indecomposable positive definite quadratic
forms, JRAM 276 (1975),99-123 [IS].
[O'Me3] Hilbert's eleventh problem: the arithmetic theory of quadratic
forms, PSPM 28 (1976), 379-400 [151.
[O'Me4] and B. Pollak, Generation of local integral orthogonal groups,
Math. Zeit. 87 (1965), 385-400; 93 (1966), 171-188 [151.
1. K. O'Mura: see A. J. Viterbi.
[O'Nal] M. O'Nan, Some evidence for the existence of a new simple group,
PLMS 32 (1976),421-479 [101.
[Ongl] H. E. Ong, Perfect quadratic forms over real-quadratic number fields,
Geom. Dedic. 20 (1986),51-77 [21.
[Ong2] and D. Golke, An algorithm for the computation of perfect
polyhedral cones over real quadratic number fields, Lect. Notes Compo
Sci. 204 (1985), 487-488 [2].
[Oppl] A. Oppenheim, Remark on the minimum of quadratic forms, JLMS 21
(1946),251-252 [I, 61.
[OrzIl G. Orzech, editor, Conference on Quadratic Forms-1976, Queen's
Papers Pure Appl. Math. 46, Queen's Univ., Kingston, Ontario, 1977
[IS, B1.
Bibliography 617

[Otrl] G. Otremba, Zur Theorie der hermiteschen Formen in imaginiir-

quadratischen Zahlkorpern. JRAM 249 (1971), 1-19 [2].
[Ozel] M. Ozeki, On even unimodular positive definite quadratic lattices of
rank 32, Math. Z. 191 (1986),283-291 [7].
[Oze2] On the configurations of even unimodular lattices of rank 48,
Archiv Math. 46 (1986), 54-61 [7].
[Oze3] Hadamard matrices and doubly even self-dual error-correcting
codes. JCT A 44 (1987), 274-287 [7].
[Oze4] Examples of even unimodular extremal lattices of rank 40 and
their Siegel theta-series of degree 2, JNT to appear [7].

G. Pall: see also D. Estes, R. E. O'Connor.

[Pall] The arithmetical invariants of quadratic forms. BAMS 51 (1945),
185-197 [15, 16].
[PaI2] - The weight of a genus of positive n-ary quadratic forms. PSPM 8
(1965),95-105 [16].
R. A. Parker: see also J. H. Conway, W. Meyer.
[ParI] - The computer calculation of modular characters (the meat-axe). in
[Atk2], 1984, pp. 267-274 [3].
[Par2] - and J. H. Conway, A remarkable Moufang loop. with an application
to the Fischer group Fh4. in preparation [29].
[Par3] D. Parrott, Characterization of the Fischer groups. TAMS 269 (1981),
303-347 [10].
M. S. Paterson: see R. J. Fowler.
[Patl] A. L. Patterson and W. E. Love, Remarks on the Delaunay reduction.
Acta. Cryst.l0 (1957),111-116 [2].
S. E. Payne: see P. J. Cameron.
[Peel] D. E. Peek, The icosahedral puzzle group. Abstracts Amer. Math. Soc.,
#798-20-74, 3 (1982), 465 [I Il.
[Perl] M. Perkel, A characterization of J, in terms of its geometry. Geom.
Dedic. 9 (1980),291-298 [10].
[Petl] M. Peters, Definite unimodular 48-dimensional quadratic forms. BLMS
15 (1983), 18-20 [7].
[Pet2] D. P. Petersen and D. Middleton, Sampling and reconstruction of
wave-number-limited functions in N-dimensional Euclidean space. IC 5
(1962),279-323 [3].
D. H. Peterson: see V. G. Kac.
[Pet3] W. W. Peterson and E. J. Weldon, Jr., Error-Correcting Codes. MIT
Press, Cambridge, MA, 2nd ed., 1972 [3, 5, 81.
[Pet4] H. Petersson, Modulfunktionen und Quadratische Formen. Springer-
Verlag, 1982 [2, 71.
[Pfel] H. Pfeuffer, Bemerkung zur Berechnung dyadischer Darstellungsdichten
einer quadratischen Form iiber algebraischen Zahlkorpern. JRAM 236
(1969),219-220 [161.
[Pfe2] - Quadratsummen in totalreellen algebraischen Zahlkorpern. JRAM
249 (1971), 208-216 [16].
[Pfe3] Einklassige Geschlechter totalpositiver quadratischer Formen in
totalreellen algebraischen Zahlkopern. JNT 3 (1971), 371-411 [16].
[Phel] K. T. Phelps, Every finite group is the automorphism group of some
linear code. Congr. Numer. 49 (1985), 139-141 [3].
R. Phillips: see A. Lubotzky.
618 Bibliography

J. N. Pierce: see H. Gish, V. Pless.

J. R. Pierce: see B. M. Oliver.
F. C. Piper: see D. R. Hughes.
P. Piret: see also A. E. Brouwer.
[Pirl1 - Good linear codes of length 27 and 28, PGIT 26 (1980), 227 [9].
A. Pitre: see M. Bourdeau.
[Ple1] W. Plesken, Bravais groups in low dimensions, Comm. Match. in Math.
Chern. 10 (1981) 97-119 [I, 3].
[Ple2] - Applications of the theory of orders to crystallographic groups, in
[Rogl0], 1981, pp. 37-92 [3].
[Ple3] - Finite unimodular groups of prime degree and circulants, J. Alg. 97
(1985),286-312 [3].
[Ple4] - and W. Hanrath, The lattices of six-dimensional Euclidean space,
MTAC 43 (1984),573-587 [3].
[Ple5] and M. Pohst, On maximal finite irreducible subgroups of
GL(n, Z), MTAC 31 (1977),536-573; 34 (1980), 245-301 [3].
[Ple6] Constructing integral lattices with prescribed minimum I,
MTAC 4S (1985),209-221 [6].
V. Pless: see also J. H. Conway, J. S. Leon, C. L. Mallows.
[Ple7] - The number of isotropic subspaces in a finite geometry, Rend. CI.
Scienze fisiche, matematiche e naturali, Ace. Naz. Lincei 39 (1965),
418-421 [16].
[Ple8] - On the uniqueness of the Golay codes, JCT S (1968), 215-228 [7,
14, 16, 18].
[Ple9] - On a new family of symmetry codes and related new five-designs,
BAMS 7S (1969), 1339-1342 [3].
[Plel0] - A classification of self-orthogonal codes over GF(2), Discr. Math.
3 (1972),209-246 [71.
[Ple11] Symmetry codes over GF(3) and new five-designs, JCT AI2
(1972), 119-142 [3].
[Ple12] - The children of the (32,16) doubly even codes, PGIT 24 (1982),
738-746 [7, 15].
[Ple13] - 23 does not divide the order of the group of a (72,36,16) doubly-
even code, PGIT 28 (1982), 112-117 [7].
[Plel4] - On the existence of some extremal self-dual codes, in Dacl1, 1984,
pp. 245-250 [71.
[Plel5] - A decoding scheme for the ternary Golay code, in Proc. 20th
Allerton Conf. Comm. Control., Univ. of Ill., Urbana, 1982, pp. 682-
687 [J 11.
[Plel5a] - Decoding the Golay codes, PGIT 32 (1986), 561-567 [11].
[Plel6] - and J. N. Pierce, Self-dual codes over GF(q) satisfy a modified
Varshamov-Gilbert bound, IC 23 (1973), 35-40 [161.
[Plel7] - and N. J. A. Sloane, On the classification and enumeration of self-
dual codes, JCT AI8 (1975),313-335 [7, 14, 16].
[Plel8] - - and H. N. Ward, Ternary codes of minimum weight 6, and the
classification of self-dual codes of length 20, PGIT 26 (1980), 305-316
[4, 71.
[Plel9] - and J. G. Thompson, 17 does not divide the order of the group of a
(72,36,16) doubly-even code, PGIT 28 (1982),537-541 [7].
[Ple20] - and V. D. Tonchev, Self-dual codes over GF(7), PGIT to appear
[7].
M. Pohst: see also U. Fincke, W. Plesken.
[Pohl] - On the computation of lattice vectors of minimal length, successive
Bibliography 619

minima and reduced bases with applications, ACM SIGSAM Bull. 15

(1981), 37-44 [21.
[Poh21 On integral lattice constructions, Abstracts Amer. Math. Soc. 3
(1982), pp. 152, Abstract 793-12-14 [61.
B. Pollak: see O. T. O'Meara.
H. O. Pollak: see H. J. Landau, D. Slepian.
[PomIl C. Pomerance, Analysis and comparison of some integer factoring
algorithms, in Computational Methods in Number Theory, ed. H. W.
Lenstra, Jr. and R. Tijdeman, Math. Centre Tracts 154, Math. Centre,
Amsterdam, 1982, pp. 89-139 [151.
[Posl1 E. C. Posner, Combinatorial structures in planetary reconnaissance, in
Error Correcting Codes, ed. H. B. Mann, Wiley, NY, 1969, pp. 15-46
[201.
[Powl1 M. B. Powell and G. Higman, editors, Finite Simple Groups, Ac. Press,
NY, 1971 [BI.
F. P. Preparata: see also D. T. Lee.
[Prell - A class of optimum nonlinear double-error-correcting codes, IC 13
(1968), 378-400 [31.
[Prelal editor, Advances in Computing Research I: Computational
Geometry, Jai Press, Greenwich, CT 1983.
[Pre21 - and M. I. Shamos, Computational Geometry, Springer-Verlag, 1985
[21.
I. Prigogine: see G. Pry.
A. D. Pritchard: see J. H. Conway.
[Proll G. Promhouse and S. E. Tavares, The minimum distance of all binary
cyclic codes of odd lengths from 69 to 99, PGIT 24 (1978), 438-442
[3,51.
[Pryl1 G. Pry and I. Prigogine, Rayons et nombres de coordination de quelques
re'seaux simples, Acad. Roy. Belgique, Bull. Cl. Sci. 28 (1942),866-873 [41.
[Pugl1 A. Pugh, Polyhedra: A Visual Approach, Univ. of Calif., Berkeley CA,
1976 [211.

Q
[Quell H.-G. Quebbemann, Zur Klassifikation unimodularer Gitter mit
Isometrie von Primzahlordnung, JRAM 326 (1981), 158-170 [81.
[Que21 Gitter ohne Automorphismen der Determinante -1, Manuscripta
Math. 40 (1982), 245-253.
[Que31 - An application of Siegel's formula over quaternion orders to the
existence of extremal lattices, Math. 31 (1984), 12-16 [8, 161.
[Que41 - Definite lattices over real algebraic function domains, Math. Ann.
272 (1985),461-476 [21.
[Que51 A construction of integral lattices, Math. 31 (1984), 137-140
[1,7,81.
[Que61 - Lattices with theta-functions for G(Ji.) and linear codes, J. Alg.
lOS (1987),443-450 [1,2,7,81.
L. Queen: see also R. E. Borcherds, J. H. Conway.
[Que71 Some relations between finite groups, Lie groups and modular
functions, Ph.D. Dissertation, Univ. of Cambridge, 1980 [301.
[Que81 Modular functions and finite simple groups, PSPM 37 (1980),
561-566 [301.
[Que91 Modular functions arising from some finite groups, MTAC 37
(1981),547-580 [301.
S. U. Qureshi: see G. D. Forney, Jr.
620 Bibliography

[Radl) H. Rademacher, Topics in Analytic Number Theory, Springer-Verlag,

1973 [4, 71.
R. V. Raev: see V. D. Tonchev.
[Ragl) D. Raghavarao, Constructions and Combinational Problems in Design of
Experiments, Wiley, NY, 1971 [31.
B. Ramamurthi: see D.-Y. Cheng.
[Ramil S. Ramanujan, On certain arithmetical functions, Trans. Camb. Phil.
Soc. 22 (1916), 159-184 = Paper 18 of [Ram2) [71.
[Ram2) - Collected Papers, Camb. Univ. Press, 1927; Chelsea, NY, 1962 [BI.
D. E. Ramirez: see C. F. Dunk!.
R. A. Rankin: see also D. G. Kendal!.
[Ranil - On the closest packing of spheres in n dimensions, Ann. Math. 48
(1947),1062-1081 [91.
[Ran2) - A minimum problem for the Epstein zeta-function. Proc. Glasgow
Math. Assoc. 1 (1952-3), 149-158.
[Ran3) On positive definite quadratic forms, JLMS 28 (1953), 309-314
[15).
[Ran4) The closest packing of spherical caps in n dimensions, Proc.
Glasgow Math. Assoc. 2 (1955),139-144 [11.
[Ran4a) On packings of spheres in Hilbert space, Proc. Glasgow Math.
Assoc. 2 (1955), 145-146.
[Ran5) - On the minimal points of positive definite quadratic forms, Math. 3
(1956), 15-24 [151.
[Ran5a) On the minimal points of perfect quadratic forms, Math. Z. 84
(1964), 228-232 [151.
[Ran6) - Modular Forms and Functions, Camb. Univ. Press, 1977 [2, 4, 71.
[Ran7) - Fourier coefficients of cusp forms, PCPS 100 (1986), 5-29 [2).
[Raol) V. V. Rao and S. M. Reddy, A (48,31,8) linear code. PGIT 19 (1973),
709-711 [81.
[Rasl) R. Rasala, Split codes and the Mathieu groups, J. Alg. 42 (1976),422-
471 [181.
[Raul) H. E. Rauch and H. M. Farkas, Theta Functions with Applications to
Riemann Surfaces, Williams and Wilkins, Baltimore MD, 1974 [41.
[Rau2) - and A. Lebowitz, Elliptic Functions, Theta Functions, and Riemann
Surfaces, Williams and Wilkins, Baltimore MD, 1973 [4).
S. M. Reddy: see V. V. Rao.
[Reeil I. S. Reed and G. Solomon, Polynomial codes over certain finite fields,
SIAJ 8 (1960),300-304 [31.
[Reil) L. W. Reid, The Elements of the Theory of Algebraic Numbers,
Macmillan, NY, 1910 [151.
[Rei2) I. Reiner, A survey of integral representation theory. BAMS 76 (1970),
159-227 [31.
[Rei3) - Maximal Orders, Ac. Press, NY, 1975 [31.
[Rei4) - Topics in Integral Representation Theory. LNM 744, 1979 [31.
[Rei5) E. M. Reingold, J. Nievergelt and N. Deo, Combinatorial Algorithms:
Theory and Practice, Prentice-Hall, Englewood Cliffs NJ, 1977 [201.
H. P. Reiss: see T. Beth.
[Renil R. J. Renka, Interpolation of data on the surface of a sphere, ACM
Trans. Math. Software 10 (1984),417-436 and 437-439 [11.
[Ribl) P. Ribenboim, Algebraic Numbers, Wiley, NY, 1972 [81.
Bibliography 621

J. R. Rice: see T. J. Aird.

[Riell C. R. Riehm and I. Hambleton, editors, Quadratic and Hermitian
Forms, Canad. Math. Soc. Conf. Proc. 4, Amer. Math. Soc., Providence
RI, 1983 [15, BI.
[RioIl J. Riordan, An Introduction to Combinatorial Analysis, Wiley, NY,
1958 [21].
[Rio2] J. Riordan, Combinatorial Identities, Wiley, NY, 1968 [21l.
[RitIl D. Ritchie and S. F. Siovin, Protoachlya polysporus revisited, Mycologia
66 (1974), 362-364 [I].
E. F. Robertson: see R. A. Wilson.
J. P. Robinson: see A. W. Nordstrom.
[Robil R. M. Robinson, Arrangement of 24 points on a sphere, Math. Ann. 144
(1960, 17-48 [I I.
[Rob2] - Finite sets of points on a sphere with each nearest to five others,
Math. Ann. 179 (1969), 296-318 [II.
E. R. Rodemich: see R. J. McEliece.
C. A. Rogers: see also H. S. M. Coxeter, P. Erdos.
[Rogl] A note on coverings, Math. 4 (1957), 1-6 [21.
[Rog2] The packing of equal spheres, PLMS 8 (1958),609-620 [1,9].
[Rog3] Lattice coverings of space, Math. 6 (1959), 33-39 [21.
[Rog4] An asymptotic expansion for certain Schliifli functions, JLMS 36
(1960,78-80.
[Rog5] The chance that a point should be near the wrong lattice point,
JLMS 37 (1962),161-163.
[Rog6] Covering a sphere with spheres, Math. 10 (1963), 157-164 [21.
[Rog7] - Packing and Covering, Camb. Univ. Press, 1964 [I, 2, 91.
[Rog8] Probabilistic and combinatorial methods in the study of the
geometry of Euclidean spaces, in Proc. Intern. Congr. Math.,
Vancouver, 1974, pp. 497-500.
[Rog9] K. W. Roggenkamp, Lattices over Orders II, LNM 142, 1970 [21.
[RogIO] - editor, Integral Representations and Applications, LNM 882, 1981
[3, BI.
[RogIIl - and V. Huber-Dyson, Lattices over Orders I, LNM lIS, 1970 [2].
A. P. Rollett: see H. M. Cundy.
[Roml] A. M. Romanov, New binary codes of minimal distance 3, PPI 19
(1983), 101-102 [3, 5,91.
[RooI] C. Roos and C. de Vroedt, Upper bounds for A (n,4) and A (n,6)
derived from Delsarte's linear programming bound, Discr. Math. 40
(1982),261-276 [9].
[Roo2] D. Roose and E. de Doncker, Automatic integration over a sphere, J.
Comput. Appl. Math. 7 (1980,203-224 [ll.
[Roql] P. Roquette, On class field towers, in [Cas5], 1967, pp. 231-249 [8].
A. Rosa: see C. C. Lindner.
[Rudl] A. N. Rudakov and I. R. Shafarevich, Supersingular K3 surfaces over
fields of characteristic 2, IANS 42 (1978), 848-869 = Izv. 13 (1979),
147-165 [lSI.
[Rud2] - - Surfaces of type K3 over fields of finite characteristic, IANS 18
(1980,115-207 = JSM 22 (1983), 1476-1533 [lSI.
V. P. Rud'ko: see P. M. Gudivok.
L. D. Rudolph: see J. Gao.
A. Rudvalis: see also L. Finkelstein.
[Rud3] A rank 3 simple group of order 214 .3 3 .5 3 .7.13.29, J. Alg. 86
(1984), 181-258 [101.
622 Bibliography

H. C. Rumsey, Jr.: see R. J. McEliece.

[Rusl1 J. A. Rush and N. J. A. Sloane, An improvement to the Minkowski-
Hlawka bound for packing superballs, Math. 1987, to appear [II.
[Rybl1 A. J. E. Ryba, A new construction of the O'Nan simple group, J. Alg. to
appear [101.
S. S. Ryskov: see also E. P. Baranovskii, B. N. Delone.
[Rysll - Effectuation of a method of Davenport in the theory of coverings,
DAN 175 (1967) 303-305 - SMD 8 (1967), 865-867 [2, 151.
[Rys21 - On the reduction theory of positive quadratic forms, DAN 198
(1971), 1028-1031 - SMD 12 (1971),946-950 [2, 151.
[Rys31 On Hermite, Minkowski and Voronoi reduction of positive
quadratic forms in n variables, DAN 207 (1972) 1054-1056 - SMD 13
(1972), 1676-1679 [2, 151.
[Rys41 - On maximal finite groups of integer (n x n) -matrices, DAN 204
(1972, No.3) 561-564 = SMD 13 (1972), 720-724 [151.
[Rys51 - Maximal finite groups of integral (n x n) -matrices and full groups
of integral automorphisms of positive quadratic forms (Bravais
lattices), TMIS 128 (1972), 183-211,261 - PSIM 128 (1972) 217-250
[1, 151.
[Rys61 - On the question of the final f - optimality of lattices that yield the
densest packing of n-dimensional balls, Sibirsk. Mat. Z. 14 (1973),
1065-1075 and 1158 - Sib. Math. J. 14 (1973),743-750 [151.
[Rys71 The perfect form A;: the existence of lattices with a
nonfundamental division simplex, and the existence of perfect forms
which are not Minkowski-reducible to forms having identical diagonal
coefficients, LOMI 33 (1973),65-71 = JSM 6 (1976), 672-676 [4, 151.
[Rys81 The Hermite-Minkowski theory of reduction of positive definite
quadratic forms, LOMI 33 (1973), 37-64 = JSM 6 (1976), 651-671 [2,
151.
[Rys91 The geometry of positive quadratic forms, Proc. Int. Congress
Math., Vancouver, 1974, I, pp. 501-506 - Amer. Math. Soc. Transl.
(2) 109 (1977), 27-32 [151.
[Rys101 - On the problem of the determination of the perfect quadratic forms
in many variables, TMIS 142 (1976), 215-239 - PSIM 142 (1976),
233-259 [151.
[Rysll I editor, The Geometry of Positive Quadratic Forms, TMIS 152
(1980) = PSIM (1982, No.3) [151.
[Rys 121 - and E. P. Baranovskii, Solution of the problem of least dense lattice
covering of five-dimensional space by equal spheres, DAN 222 (1975),
39-42 = SMD 16 (1975) 586-590 [2,4, 151.
[Rys131 - - C-types of n-dimensional lattices and 5-dimensional primitive
parallelohedra (with application to the theory of coverings), TMIS 137
(1976, No.4) = PSIM (1978, No.4) [2,4, 151.
[Rys141 - - Classical methods in the theory of lattice packings, Uspekhi
Mat. Nauk 34 (No.4, 1979), 3-63 - Russian Math. Surveys 34 (No.4,
1979), 1-68 [2, 151.
[Rys151 - - Perfect lattices as admissable centerings, Itogi Nauki Tekh., Ser.
Probl. Geom. 17 (1985), 3-49 [21.
S

T. L. Saaty: see also H. S. M. Coxeter.

[Saall - and J. M. Alexander, Optimization and the geometry of numbers:
Bibliography 623

packing and covering, SIAM Rev. 17 (1975),475-519.

C.-H. Sah: see R. Brauer.
N. N. Sandakova: see B. N. Delone.
[Sanl] A. Santoro and A. D. Mighell, Determination of reduced cells. Acta
Cryst. A26 (1970), 124-127 [1].
P. Sarnak: see A. Lubotsky.
[SaUl J. Satterly, The moments of inertia of some polyhedra. Math. Gazette,
42 (1958), 11-13 [211.
[Sawl] K. Sawade, A Hadamard matrix of order 268, Graphs Comb. 1 (1985),
185-187 [31.
[Sayl] S. I. Sayegh, A condition for optimality of two-dimensional signal
constellations. COMM 33 (1985), 1220-1222 [31.
[Say2] D. Sayre, editor, Computational Crystallography, Oxford Univ. Press,
1982 [21.
c. A. Schaffner: see H. A. Krieger.
W. Scharlau: see also M. Knebusch.
[SchO] Quadratic Forms. Queen's Papers on Pure and Applied
Mathematics No. 22, Queen's Univ., Kingston Ontario, 1970 [151.
[Sch 1] - A historical introduction to the theory of integral quadratic forms.
in [Orzll, 1976, pp. 284-339 [I51.
[Sch2] - Quadratic and Hermitian Forms. Springer-Verlag, 1985 [151.
J. R. Schatz: see G. D. Cohen.
[Sch3] J. Scherk, An overview of supersymmetry and supergravity. Lecture
Notes Phys. 116 (1980), 343-357 [11.
P. Schmid: see W. Knapp.
P. H. Schmidt: see E. L. Allgower.
C. P. Schnorr: see also J. C. Lagarias, H. W. Lenstra, Jr.
[Sch4] A hierarchy of polynomial time basis reduction algorithms. in
Theory of Algorithms, North-Holland, 1985,375-386 [2].
[Sch5] I. J. Schoenberg, Positive definite functions on spheres. DMJ 9 (1942),
96-107 [9].
[Sch6] B. Schoeneberg, Das Verhalten von mehrfachen Thetareichen bei
Modulsubstitutionen. Math. Ann. 116 (1939), 511-523 [2, 7].
[Sch7] - Elliptic Modular Functions. Springer-Verlag, 1974 [2, 7].
A. Schrijver: see also J. H. van Lint.
[Sch8] editor, Packing and Covering in Combinatorics. Math. Centre
Tracts 106, Math. Centre, Amsterdam, 1979 [B].
[Sch9] - A comparison of the Delsarte and Lovdsz bounds. PGIT 25 (1979),
425-429 [9].
M. R. Schroeder: see also B. S. Atal, J. L. Flanagan.
[SchlO] - B. S. Atal and J. L. Hall, Optimizing digital speech encoders by
exploiting masking properties of the human ear. J. Acoust. Soc. Am.
66 (1979), 1647-1652 [2].
[Schill K. Schiitte and B. L. van der Waerden, Auf welcher Kugel haben 5. 6.
7. 8 oder 9 Punkte mit Mindesabstand Eins Platz?, Math. Ann. 123
(1950,96-124 [I).
[Sch12] - - Das Problem der dreizehn Zugeln. Math. Ann. 125 (1953),
325-334 [11.
[Sch13] R. Schulze-Pillot, Thetareihen positiv definiter quadratischer Formen.
Invent. Math. 75 (1984), 283-299 (15).
[Schl4] - Darstellungsmasse von Spinorgeschlechtern terniirer quadratischer
Formen. JRAM 352 (1984), 114-132 [151.
[Sch15] J. H. Schwarz, editor, Superstrings, World Scientific, Singapore, 1985 [I].
624 Bibliography

[SchI6] R. L. E. Schwarzenberger, N-Dimensional Crystallography, Pitman,

San Francisco CA, 1980 [1l.
[Sch 17] - Graphical representation of n-dimensional space groups, PLMS 44
(1982), 244-266.
[Scol] P. R. Scott, On perfect and extreme forms, JAMS 4 (1964), 56-77 [2].
[Sco2] The construction of perfect and extreme forms, CJM 18 (1966),
145-158 [21.
[Sco3] W. R. Scott and F. Gross, Proceedings of the Conference on Finite
Groups, Ac. Press, NY, 1976 [B].
B. Sedlacek: see F. R. Ahmed.
[Segl] B. Segre and K. Mahler, On the densest packing of circles, AMM 51
(1944),261-270 [ll.
J. J. Seidel: see also E. Bannai, P. Delsarte, J.-M. Goethals, A.
Neumaier, N. J. A. Sloane.
[Seill - Strongly regular graphs, in Progress in Combinatorics, edited by W.
T. Tutte, Ac. Press, NY, 1969, pp. 185-197 [9].
[Sei2] Strongly regular graphs, in Surveys in Combinatorics, ed. B.
Bollobas, Camb. Univ. Press, 1979, pp. 157-180 [9].
[Sei3] Delsarte's theory of association schemes, in Graphs and Other
Combinatorial Topics, Teubner, Leipzig, 1983, pp. 249-258 [9, 14].
R. Seidel: see H. Edelsbrunner.
[Sei4] Constructing higher-dimensional convex hulls at logarithmic cost
per face, in Proc. 18th Annual ACM Sympos. Theory of Computing,
Assoc. Computing Machinery, N.Y. 1986, pp. 404-413 [2].
F. Seitz: see E. Wigner.
G. M. Seitz: see C. W. Curtis.
[Semll N. V. Semakov and V. A. Zinoviev, Complete and quaSi-complete
balanced codes, PPI 5 (No.2, 1969), 28-36 = PIT 5 (No.2, 1969), 11-
13 [3].
[Serl] J.-P. Serre, Cours d'Arithme'tique, Presses Universitaires de France,
Paris, 1970; English translation, Springer-Verlag, 1973 [2, 4, 7, 15, 16,
18, 26].
[Ser2] - Corps Locaux, Hermann, Paris, 2nd ed., 1968; English translation,
Springer-Verlag, 1979 [2].
[Ser3] - Sur Ie nombre des points rationnels d'une courbe algeorique sur un
corpsjini, CR 296 (1983), 397-402 [3].
[Seyl] P. D. Seymour and T. Zaslavsky, Averaging sets: a generalization of
mean values and spherical designs, Adv. in Math. 52 (1984), 213-240
[3].
I. R. Shafarevich: see Z. I. Borevich, E. S. Golod, A. N. Rudakov.
M. I. Shamos: see F. P. Preparata.
[ShaO] E. A. Shamsiev and S. S. Shushbaev, Construction of cubature formulae
that are invariant with respect to groups of integral automorphisms of
perfect quadratic forms, Dokl. Akad. Nauk UzSSR (No.3, 1982), 6-8;
Math. Rev. 83 #41034 [ll.
C. E. Shannon: see also B. M. Oliver. All of the following except
[Sha6] are reprinted in [Sle5].
[Shall - A mathematical theory of communication, BSTJ 27 (1948), 379-
423 and 623-656 [3].
[Sha2] - Communication in the presence of noise, PIEEE 37 (1949), 10-21
[3].
[Sha3] - Prediction and entropy of printed English, BSTJ 30 (195 J), 50-64
[3].
Bibliography 625

[Sha4] The zero error capacity of a noisy channel, PGIT 2 (1956), 8-19 [3,
9].
[Sha5] Certain results in coding theory for noisy channels, IC 1 (1957),6-
25 [3].
[Sha6] - Probability of error for optimal codes in a Gaussian channel, BSTJ
38 (1959), 611-656 [I, 3].
[Sha7] - Coding theorems for a discrete source with a fidelity criterion, IRE
Nat. Cony. Record (Part 4, 1959), 142-163 [3].
[Sha8] R. G. Gallager and E. R. Berlekamp, Lower bounds to error
probability for coding on discrete memoryless channels, IC 10 (1967),
65-103 and 522-552 [3].
H. M. Shapiro: see A. R. Calderbank.
J. Shawe-Taylor: see N. L. Biggs.
[SheO] D. Shechtman, I. Blech, D. Gratias and J. W. Cahn, Metallic phase
with long-range orientation order and no translational symmetry, Phys.
Rev. Lett. 53 (1984), 1951-1953 [1].
O. I. Shekhovtsov: see S. N. Litsyn.
G. C. Shephard: see also B. Griinbaum, P. McMullen.
[Shell - Unitary groups generated by reflections, CJM 5 (1953), 364-383
[4].
[She2] - and J. A. Todd, Finite unitary reflection groups, CJM 6 (1954),
274-304 [3, 4, 71.
[She3] L. A. Shepp, Computerized tomography and nuclear magnetic
resonance, J. Computer Assisted Tomography 4 (1980),94-107 [1].
[She4] - and J. B. Kruskal, Computerized tomography: the new medical X-ray
technology, AMM 85 (1978),420-439 [11.
F. A. Sherk: see C. Davis.
[Shill G. Shimura, Introduction to the Arithmetic Theory of Automorphic
Functions, Princeton Univ. Press, 1971 [71.
Y. Shoham: see D.-Y. Cheng.
S. S. Shushbaev: see E. P. Baranovskii, E. A. Shamsiev.
[Shu 1] - An algorithm for the calculation of groups of integer automorphisms
of perfect forms in n ~ 2 variables, Voprosy Vychisl. Prikl. Mat.
(Tashkent) 22 (1973), 3-13; Math. Rev. 58 #16516 [3].
[Shu2] - The determination of the groups of integer automorphisms of perfect
quadratic forms, Dokl. Akad. Nauk UzSSR, (No. I, 1974), 16-18
13].
[Shu3] - The Dirichlet-Voronoi domain for the second perfect form, Voprosy
Vychisl. Prikl. Mat. (Tashkent) 45 (1977) , 3-11; Math. Rev. 58
#27811 [2].
[Shu4] - Voronoi neighborhood of the perfect form <PIS (XhX2, ••• , X1),
Voprosy Vychisl. Prikl. Mat. (Tashkent), 77 (1985),48-56 [2].
[Sidl] V. M. Sidel'nikov, On the density of sphere packings on the surface of
an n-dimensional Euclidean sphere and on the cardinality of a binary
code with a given minimum distance, DAN 213 (No.5, 1973), 1029-
1032 - SMD 14 (1973), 1851-1855 [1,9].
[Sid2] Upper bounds on the cardinality of a binary code with a given
minimum distance, PPI 10 (No.2, 1974), 43-51 - IC 28 (1975), 292-
303 [9].
[Sid3] New bounds for densest packings of spheres in n-dimensional
Euclidean space, Mat. Sb. 95 (No. 137, 1974), 148-158 - Sb. 24
(1974), 147-157 [1,9].
[Sid4] - On extremal polynomials used to estimate the size of codes, PPI 16
626 Bibliography

(No.3, 1980), 17-30 = PIT 16 (1980),174-186 [91.

V. R. Sidorenko: see E. M. Gabidulin.
[Siel] C. L. Siegel, Uber die analytische Theorie der quadratischen Formen,
Ann. Math. 36 (1935), 527-606 = Gesam. Abh. I, 326-405, Springer-
Verlag, 1966 [7, 161.
[Sie2] Uber die Fourierschen Koeffizienten der Eisensteinschen Reihen,
Danske Vid. Selskab. Mat.-fys. Meddelelser 34 (1964), Nr. 6 - Gesam.
Abh. III, 443-458 [7, 161.
[Sie3] - Berechnung von Zetafunktionen an ganzzahligen Stel/en, Gottingen
Nach. 10 (1969), 87-102 - Gesam. Abh. IV, 82-97 [7, 161.
D. Siersma: see M. Hazewinkel.
[Sigl] F. Sigrist, Sphere packing, Math. Intelligencer 5 (No.3, 1983), 34-
38 [11.
[SimO] M. Simonnard, Linear Programming, Prentice-Hall, Englewood Cliffs,
NJ, 1966 [9, 14].
C. C. Sims: see also D. G. Higman, J. S. Leon.
[SimI] - On the isomorphism of two groups of order 44,352,000, in [Bra21,
1969, pp. 101-108 [101.
[Sim2] The existence and uniqueness of Lyons' group, in [Gag21, 1973,
138-141 [101.
[Sim3] - How to construct a Baby Monster, in [Col21, 1980,339-345 [101.
I. M. Singer: see J. I. Lepowsky.
[Skul] B. F. Sku benko, Dense lattice packings of spheres in Euclidean spaces
of dimension n ~ 16, LOMI 82 (1979), 144-146 = JSM 18 (1982),
958-960 [4, 61.
[Sku2] - A remark on an upper bound on the Hermite constant for the
densest lattice packing of spheres, LOMI 82 (1979), 147-148 - JSM 18
(1982), 960-961.
H. A. Sleeper: see S. G. Wilson.
[Slel] D. Slepian, Bounds on communication, BSTJ 42 (1963), 681-707 [3].
[Sle2] - Permutation modulation, PIEEE (1965), 228-236 [31.
[Sle3] - Group codes for the Gaussian channel, BSTJ 47 (1968), 575-602
[31.
[S1e41 On neighbor distances and symmetry in group codes, PGIT 17
(1971),630-632 [31.
[Sle51 - editor, Key Papers in the Development of Information Theory, IEEE
Press, NY, 1974 [3, Bl.
[Sle61 - Some comments on Fourier analysis, uncertainty and modeling,
SIAM Rev. 25 (1983),379-393 [31.
[SIe71 - and H. O. Pollak, Prolate spheroidal wave functions, Fourier analysis
and uncertainty I, BSTJ 40 (1961),43-63 [31.
[Slnll I. H. Sloan, Lattice methods for multiple integration, J. Comput. Appl.
Math. 12 (1985), 131-143 [I1.
[SIn21 - and P. J. Kachoyan, Lattice methods for multiple integration: theory,
error analysis and examples, SIAM J. Num. Anal. 24 (1987), 116-128
[ 11.
N. J. A. Sloane: see also S. F. Assmann, R. P. Bambah, E. Bannai, E. S.
Barnes, E. R. Berlekamp, M. R. Best, R. E. Borcherds, A. Bos, A. R.
Calderbank, Y. Cheng, G. D. Cohen, J. H. Conway, D. E. Downey,
V. Elser, R. L. Graham, M. Harwit, W. C. Huffman, K. E. Kilby, J.
Leech, J. S. Leon, F. J. MacWilliams, C. L. Mallows, A. M. Odlyzko, V.
Pless, J. A. Rush, B. K. Teo.
[Slol] - Sphere packings constructed from BCH and Justesen codes, Math.
Bibliography 627

19 (J972), 183-190 [I, 51.

[SI02] - A Handbook of Integer Sequences. Ac. Press, NY, 1973 [211.
[SI03] - Is there a (72,36) d = 16 self-dual code?, PGIT 19 (1973), 251
[71.
[SI04] - A Short Course on Error-Correcting Codes. Springer-Verlag, 1975.
[SI05] An introduction to association schemes and coding theory. in
[Ask2a], pp. 225-260 [9, 141.
[SI06] - Weight enumerators of codes. in Combinatorics. ed. M. Hall Jr. and
J. H. van Lint, Reidel, Dordrecht, Holland, 1975, pp. 115-142 [71.
[SI07] - Error-correcting codes and invariant theory: new applications of a
nineteenth-century technique, AMM 84 (1977),82-107 [71.
[SI08] Binary codes. lattices and sphere packings. in Combinatorial
Surveys. ed. P. J. Cameron, Ac. Press, NY, 1977, pp. 117-164 [4, 71.
[SI09] - Codes over GF(4) and complex lattices. J. Alg. 52 (1978), 168-181
[71.
[SloIO] - Self-dual codes and lattices. PSPM 34 (J979), 273-308 [3,7, 161.
[Slo11] - A note on the Leech lattice as a code for the Gaussian channel. IC
46 (J980), 270-272 [31.
[SI012] Tables of sphere packings and spherical codes. PGIT 27 (J98J),
327-338 [1,41.
[SI013] Recent bounds for codes. sphere pac kings and related problems
obtained by linear programming and other methods. Con temp. Math. 9
(1982), 153-185 == Chap. 9 of this book [3, 9].
[SloI3a] - Encrypting by random rotations. Lecture Notes Computer Sci. 149
(1983),71-128 [11.
[SloI4] The packing of spheres. Scientific American 250 (No. I, 1984),
116-125 [11.
[SloI5] - A new approach to the covering radius of codes. JCT A42 (1986),
61-86 [31.
[SI016] - Unsolved problems related to the covering radius of codes. in Proc.
Conf. Specific Prob. Commun. (SPOC-85). Springer-Verlag, 1987, to
appear 131.
[SI017] Theta-series and magic numbers for diamond and certain ionic
crystal structures. J. Math. Phys. 28 (1987), 1653-1657 [1,2,4].
[SloI8] - and J. J. Seidel, A new family of nonlinear codes obtained from
conference matrices. Ann. New York Acad. Sci. 175 (1970), 363-365
[3, 51.
[SI019] - and B. K. Teo, Theta series and magic numbers for closed-packed
spherical clusters. J. Chern. Phys. 83 (1985),6520-6534 [I, 41.
[SI020] - and D. S. Whitehead, A new family of single-error correcting codes.
PGIT 16 (1970), 717-719 [3, 5].
S. F. Siovin: see D. Ritchie.
[Smal] I. Smalley, Simple regular sphere packings in three dimensions. Math.
Mag. 36 (1963), 295-299 [4].
D. H. Smith: see also N. L. Biggs, P. Hammond.
[Smil] - On tetravalent graphs. JLMS 6 (1973),659-662 [91.
[Smi2] - Distance-transitive graphs. in Combinatorics. ed. T. P. McDonough
and V. C. Mavron, Camb. Univ. Press, 1974, pp. 145-153 [91.
[Smi3] - Distance-transitive graphs of valency four. JLMS 8 (1974), 377-
384 [9].
[Smi4] F. L. Smith, A general characterization of the Janko simple group Jz,
Archiv Math. 25 (1974),17-22 [101.
[Smi5] H. J. S. Smith, Report on the theory of numbers. Report of the British
628 Bibliography

Association (1859) 228-267; (1860), 120-169; (1860, 292-340; (1862),

503-526; (1863), 768-786; (1865), 322-375 = Coll. Math. Papers I, 38-
364 [4, 15].
[Smi6] - On the orders and genera of quadratic forms containing more than
three indeterminates, PRS 16 (1867), 197-208 - Coll. Math. Papers I,
510-523 [2,4, 151.
[Smi7] K. T. Smith, D. C. Solmon and S. L. Wagner, Practical and
mathematical aspects of the problem of reconstructing objects from
radiographs, BAMS 83 (1977), 1227-1270 [11.
[Smi8] M. S. Smith, On rank 3 permutation groups, J. Alg. 33 (1975), 22-42
[101.
[Smi9] - On the isomorphism of two simple groups of order 44,352,000, J.
Alg.41 (1976), 172-174 [10].
[SmilO] - A combinatorial configuration associated with the Higman-Sims
group, J. Alg. 41 (1976), 175-195 [101.
N. K. Smith: see S. G. Wilson.
[Smill] R. F. Smith, The construction of definite indecomposable Hermitian
forms, AJM 100 (1978), 1021-1048 [2].
S. D. Smith: see also A. O. L. Atkin.
[Smil2] - Large extraspecial subgroups of widths 4 and 6, J. Alg. 58 (1979),
251-281 [29].
[Smil3] - On the Head characters of the Monster simple group, in [McK3],
1985, pp. 303-313 [1,29,301.
W. R. Smith: see T. W. Melnyk.
[Sobl] I. M. Sobol, On systematic search in a hypercube, SIAM J. Num. Anal.
16 (1979), 790-793 [11.
[Sob2] S. L. Sobolev, Cubature formulae on the sphere invariant under finite
groups of rotations, DAN 146 (No.2, 1962), 310-313 - SMD 3
(1962), 1307-1310 [1, 3].
[Sob3] - Introduction to the Theory of Cubature Formulae, Nauka, Moscow,
1974 [1, 31.
L. H. Soicher: see also J. H. Conway.
[Soil] Presentations for some groups related to COl, Proceedings of
"Computers in Algebra" Conference, Chicago, 1985, to appear 1101.
[Soi2] - From the Monster to the Bimonster, J. Alg. to appear [291.
[Soi3] - Presentations for Conway's group Co" pCPS 102 (1987), 1-3 [101.
D. C. Solmon: see K. T. Smith.
G. Solomon: see also I. S. Reed.
[Soil] - and M. M. Sweet, A Golay puzzle, PGIT 29 (1983),174-175 [111.
[Soml] D. M. Y. Sommerville, An Introduction to the Geometry of n
Dimensions, Dover, NY, 1958 [11.
[Som2] J. Sommer, Vorlesungen iiber Zahlentheorie, Teubner, Leipzig, 1907
[ 151.
[Spell T. Speevak, An efficient algorithm for obtaining the volume of a special
kind of pyramid and application to convex polyhedra, MTAC 46
(1986), 531-536 [2].
T. A. Springer: see F. van der Blij.
[Stall K. C. Stacey, The enumeration of perfect septenary forms, JLMS 10
(1975),97-104 [1, 21.
[Sta2] - The perfect septenary forms with ,14 = 2, JAMS A22 (1976), 144-
164 [1, 2].
[Sta3] R. P. Stanley, Invariants of finite groups and their application to
Bibliography 629

combinatorics, BAMS 1 (1979),475-511 [7].

[Sta4] D. Stanton, Some q-Krawtchouk polynomials on Chevalley groups,
AJM 102 (1980), 625-662 [9].
[Sta5] - Product formulas for q-Hahn polynomials, SIAM J. Math. Anal.
11 (1980), 100-107 [9].
[Sta6] - Another infinite family of perfect codes, preprint [9].
[Sta7] Three addition theorems for some q-Krawtchouk polynomials,
Geom. Dedic. 10 (1981),403-425 [9].
[Sta8] - A partially ordered set and q-Krawtchouk polynomials, JCT A30
(1981),276-284 [9].
[Sta9] - Generalized n-gons and Chebychev polynomials, JCT A34 (1983),
15-27 [9].
[StaIO] - Harmonics on posets, JCT A40 (1985), 136-149 [9].
[Stall] - Orthogonal polynomials and Chevalley groups, in [Ask2a], 1984,
pp. 87-128 [9].
[Stalla] - t-Designs in classical association schemes, Graphs Comb. 2 (1986),
283-286.
[StaI2] R. G. Stanton, The Mathieu groups, CJM 3 (1951),164-174 [11,14].
I. A. Stegun: see M. Abramowitz.
P. J. Steinhardt: see D. Levine.
[Stel] F. Stenger, Numerical methods based on Whittaker cardinal or sinc
functions, SIAM Rev. 23 (1981), 165-224 [1, 3].
R. D. Stinaff: see H. J. Helgert.
M. I. Stogrin: see also B. N. Delone.
[Stol] - The Voronoi, Venkov and Minkowski reduction domains, DAN 207
(1972), 1070-1073 - SMD 13 (1972), 1698-1702 [2].
[St02] - Locally quasidensest lattice packings of spheres, DAN 218 (1974),
62-65 - SMD 15 (1974), 1288-1292.
A. P. Street: see W. D. Wallis.
[Strl] R. L. Streit, Optimization of discrete arrays of arbitrary geometry, J.
Acoust. Soc. Am. 69 (1981), 199-212 [I].
[Str2] J. Strohmajer, Uber die Verteilung von Punkten auf der Kugel, Ann.
Univ. Sci. Budapest, Sect. Math. 6 (1963), 49-53 [I].
[Str3] A. H. Stroud, Approximate Calculation of Multiple Integrals,
Prentice-Hall, Englewood Cliffs NJ, 1971 [I].
[SugIl Y. Sugiyama, M. Kasahara, S. Hirasawa and T. Namekawa, A
modification of the constructive asympotically good codes of Justesen
for low rates, IC 25 (1974),341-350 [3].
[Sug2] - - - - A new class of asymptotically good codes beyond the
Zyablov bound, PGIT 24 (1978), 198-204 [3].
[Sug3] - - - - Superimposed concatenated codes, PGIT 26 (1980), 735-
736 [3].
[SuzIl M. Suzuki, A simple group of order 448,345,497,600, in [Bra2], 1969,
pp. 113-119 [10].
M. M. Sweet: see G. Solomon.
[Syll] J. J. Sylvester, Elementary researches in the analysis of combinatorial
aggregation, Phil. Mag. 24 (1844), 285-296 - ColI. Math. Papers I
(1904),91-102 [10, 23].
[Szell G. Szego, Orthogonal polynomials, Amer. Math. Soc., Providence RI,
4th ed., 1975 [9].
[Sze2] E. Szekely, Sur Ie proble"me de Tammes, Ann. Univ. Sci. Budapest.
Eotvos, Sect. Math. 17 (1974), 157-175 [1].
 630 Bibliography

T
[TamIl P. Tammela, The Hermite-Minkowski domain of reduction of positive
definite quadratic forms in six variables, LOMI 33 (1973), 72-89 =
1SM 6 (1976), 677-688 [2,151.
[Tam2] On the reduction theory of positive quadratic forms, DAN 209
(1973),1299-1302 = SMD 14 (1973),651-655 [2, IS].
[Tam3] On the reduction theory of positive quadratic forms, LOMI 50
(1975),6-96 = 1SM 11 (1979),197-277 [2, IS].
[Tam4] - Minkowski's fundamental reduction domain for positive quadratic
forms of seven variables, LOMI 67 (1977), 108-143, 226 = 1SM 16
(1981),836-857 [2, IS].
[TamS] P. M. L. Tammes, On the origin of number and arrangement of the
places of exit on the surface of pollen-grains, Recueil des travaux
botaniques neerlandais 27 (1930), 1-84 [I].
S. L. Tanimoto: see R. 1. Fowler.
[Tan 11 1. Tannery and 1. Molk, Ele'ments de la the'orie des Fonctions
Elliptiques, Chelsea, NY, 2nd ed., 4 vols., 1972 [4].
[Tarll T. Tarnai, Packing of 180 equal circles on a sphere, Elem. Math. 38
(1983),119-122; 39 (1984), 129 [I].
[Tar2] - Note on packing of 19 equal circles on a sphere, Elem. Math. 39
(1984),25-27 [I].
[Tar3] - Spherical circle-packing in nature, practice and theory, Structural
Topology 9 (1984), 39-58 [I].
[Tar4] - and Z. Gaspar, Improved packing of equal circles on a sphere and
rigidity of its graph, PCPS 93 (1983),191-218 [I].
T. Tasaka: see also T. Kondo.
[Tasl] - On even lattices of 2-squares type and self-dual codes, 1. Fac. Sci.
Tokyo Sect. lA Math. 28 (1981),701-714 [7].
[Tau Il O. Taussky, Introduction into connections between algebraic number
theory and integral matrices, Appendix 2 to [Coh6], 1978 [8].
[Tau2] O. Taussky, editor, Ternary Quadratic Forms and Norms, Dekker NY,
1982 [IS].
S. E. Tavares: see G. Promhouse.
K. F. Taylor: see D. Borwein.
S. Tazaki: see Y. Yamada.
[Teil] L. Teirlinck, Non-trivial t-designs without repeated blocks exist for all
t, Discr. Math. to appear [3].
B. K. Teo: see also N. J. A. Sloane.
[TeolJ - and K. Keating, Novel triicosahedral structure of the largest metal
alloy cluster: [(Ph3P)12Au13Ag12CI6]m+, 1. Am. Chern. Soc. 106 (1984),
2224-2226 [IJ.
[Teo2] - and N. 1. A. Sloane, Magic numbers in polygonal and polyhedral
atom counting, Inorg. Chern. 24 (1985),4545-4558 [1,4].
[Teo}] - - Atomic arrangements and electronic requirements for close-
packed circular and spherical clusters, Inorg. Chern. 25 (J 986), 2315-
2322 [1, 4].
[Terl] A. Terras, Harmonic Analysis on Symmetric Spaces and Applications I,
Springer-Verlag, 1985 [9].
1. Thas: see also P. J. Cameron.
[Tha 1] - Two infinite families of perfect codes in metrically regular graphs,
JCT B23 (1977),236-238 [9].
[Tha2l - Polar spaces, generalized hexagons, and perfect codes, JeT A29
(1980),87-93 [9].
Bibliography 631

B. R. C. Theobald: see C. E. Briant.

[Thil] J. Thierry-Mieg, Remarks concerning the E8 x E8 and D16 string
theories, Phys. Lett. B156 (1985), 199-202 [I].
[Thi2] Anomaly cancellation and Fermionisation in 10, 18 and 26
dimensional superstrings, Phys. Lett. B171 (1986), 163-169 [I].
J. G. Thompson: see also F. J. MacWilliams, V. Pless.
[Tholl Weighted averages associated to some codes, Scripta Math. 29
(1973),449-452 [7, 16].
[Tho2] - A simple subgroup of E8 (3), in Finite Groups Symposium, ed. N.
Iwahori, Japan Soc. Promotion Science, Tokyo, 1976, pp. 113-116 [10].
[Tho3] - Finite groups and even lattices, J. Aig. 38 (1976),523-524 [31.
[Tho4] - Uniqueness of the Fischer-Griess Monster, BLMS 11 (1979), 340-
346 [291.
[Tho5] - Finite groups and modular functions, BLMS 11 (1979),347-351
[1,29].
[Tho6] Some numerology between the Fischer-Griess Monster and the
elliptic modular function, BLMS 11 (1979),352-353 [1,29].
[Tho7] - personal communication [4, 5, 7, 8, 10, 171.
[Tho8] T. M. Thompson, From Error-Correcting Codes through Sphere
Packings to Simple Groups, Math. Assoc. Am., Washington, DC, 1983.
[Thul] A. Thue, Uber die dichteste Zusammenstellung von kongruenten Kreisen
in einer Ebene, Norske Vid. Selsk. Skr., No.1, 1910, pp. 1-9 [11.
[Tiel] A. Tietiiviiinen, Bounds for binary codes just outside the Plotkin range,
IC 47 (1980), 85-93 [9].
R. Tijdeman: see C. Pomerance.
H. C. A. van Tilborg: see E. R. Berlekamp.
[Titl] J. Tits, Sur certaines classes d'espaces homogenes de groupes de Lie,
Memoires Acad. Roy. Belg., Cl. Sci. 39 (1955), Case. 3 [9].
[Tit2] - Sur les systemes de Steiner associes aux trois "grands" groupes de
Mathieu, Rend. Math. e Appl. 23 (1964),166-184 [II].
[Tit3] - Classification of algebraic semisimple groups, PSPM 9 (1966), 33-
62 [10].
[Tit4] - Le groupe de Janko d'ordre 604,800, in [Bra2], 1969, pp. 91-95
[10].
[Tit5] - Groupes finis simples sporadiques, Sem. Bourbaki, 375, 1970 [10].
[Tit6] - Four presentations of Leech's lattice, in [CoI2], 1980, pp. 303-307
[4,8, 10, 24].
[Tit7] - Quaternions over Qhls], Leech's lattice and the sporadic group of
Hall-Janko, J. Aig. 63 (1980),56-75 [4,8, 10,24].
[Tit8] - Le Monstre (d'apres R. Griess, B. Fischer et aI.), Sem. Bourbaki
620, Asterisque 121 (1985), 105-122 [10,29].
[Tit9] - On R. Griess' "Friendly Giant", Invent. Math. 78 (1984), 491-499
[10,291.
J. A. Todd: see also H. S. M. Coxeter, G. C. Shephard.
[Todll The invariants of a finite collineation group in five dimensions,
PCPS 46 (1950), 73-90 [4].
[Tod2] The characters of a collineation group in five dimensions, PRS
A200 (1950), 320-336 [4].
[Tod3] - A representation of the Mathieu group M24 as a collineation group,
Ann. Mat. Pura Appl. 71 (1966), 199-238 [10].
N. Tokura: see T. Kasami.
[Toll] R. Tolimieri, The algebra of the finite Fourier transform and coding
theory, TAMS 287 (1985),253-273 [7].
632 Bibliography

[Toll] J. Tolke and J. M. Wills, editors, Contributions to Geometry.

Birkhauser, Boston, 1979 [B].
V. D. Tonchev: see also V. Pless.
[Toni] - Hadamard-type block designs and self-dual codes. PPJ 19 (No.4,
1983),25-30 = PIT 19 (1983),270-274 [3].
[Ton2] - Inequivalence of certain extremal self-dual codes. Compt. Rend.
Acad. Bulg. Sci. 36 (1983), 181-184 [71.
[Ton3] - A characterization of designs related to the Witt system S (5, 8, 24),
Math. Z. 191 (1986), 225-230 [II].
[Ton4] - and R. V. Raev, Cyclic 2-(17,8,7) designs and related doubly even
codes. Compt. Rend. Acad. Bulg. Sci. 35 (No. 10, 1982) [3].
E. M. Tory: see W. S. Jodrey.
[Tral] D. Travis, Spherical functions on finite groups. J. Alg. 29 (1974),65-76
[9].
D. W. Trenerry: see E. S. Barnes.
J. M. Tribolet: see J. L. Flanagan.
M. A. Tsfasman: see also G. L. Katsman, S. N. Litsyn, S. G. V!aduts.
[Tsfl] Goppa codes that are better than the Varshamov-Gilbert bound.
PPI 18 (No.3, 1982),3-6 = PIT 18 (1982), 163-165 [3].
[Tsf2] - S. G. Vladuts and T. Zink, Modular curves. Shimura curves. and
Goppa codes better than the Varshamov-Gilbert bound. Math. Nach.
104 (1982), 13-28 [3,9].
A. I. Tsitkin: see P. M. Gudivok.
A. W. Tucker: see R. J. Duffin.
[TurIl R. Turyn, Hadamard matrices. Baumert-Hall units. four-symbol
sequences. pulse compression. and surface wave encodings. JCT A16
(1974),313-333 [3].
W. T. Tutte: see J. J. Seide!.
u
[Ungl] G. Ungerboeck, Channel coding with multilevel/phase signals. PGIT 28
(1982),55-67 [3].
[UraIl H. Urakawa, On the least positive eigenvalue of the Laplacian for the
compact quotient of a certain Riemannian symmetric space. Nagoya
Math. J. 78 (1980), 137-152 [I, 9].
V
[Vall] A. Z. Val'fi~ On the representation of numbers by sums of squares
asymptotic formulae. Usp. Mat. Nauk 7 (No.6, 1952),97-178 = Amer.
Math. Soc. Trans!. (2) 3 (1956), 163-248 [4].
[VaI2] - Uber Gitterpunkte in mehrdimensionalen Ellipsoiden. Math. Z. 19
(1924),300-307 [4].
[VaI3] Gitterpunkte in Mehrdimensionalen Kugeln. Panst. Wyd. Nauk.,
Warsaw, 1957 [4].
S. A. Vanstone: see D. M. Jackson.
F. D. Veldkamp: see M. Hazewinke!.
[VenO] B. A. Venkov, On a class of Euclidean polytopes. Vestnik Leningrad.
Univ., Ser. Mat. Fiz. Him. 9 (1954),11-31 [21].
[Yen Il B. B. Venkov, The classification of integral even unimodular 24-
dimensional quadratic forms. TMIS 148 (1978),65-76 = PSJM (No.4,
1980),63-74;;;; Chap. 18 of this book [18].
[Ven2] On odd unimodular lattices. LOMI 86 (J 979), 40-48 = JSM 17
(J98I), 1967-1974 [17].
[Ven3] - On even unimodular Euclidean lattices of dimension 32, LOMJ 116
Bibliography 633

[Ven3] On even unimodular Euclidean lattices of dimension 32, LOMI 116

(1982), 44-45 and 161-162 - JSM 26 (1984), 1860-1867 [71.
[Ven4] Unimodular lattices and strongly regular graphs, LOMI 129
(1983), 30-38 - JSM 29 (1985), 1121-1127.
[Ven5] - Voronoi parallelohedra for certain unimodular lattices, LOMI 132
(1983),57-61 - JSM 30 (1985), 1833-1836 [2, 211.
[Ven6] - On even unimodular Euclidean lattices of dimension 32, II, LOMI
134 (1984), 34-58 - JSM 36 (1987),21-38 [7].
[Ven7] - Even unimodular extremal lattices, TMIS 165 (1984), 43-48 -
PSIM 165 (1984), 47-52 [71.
[VerIl T. Verhoeff, An updated table of minimum-distance bounds for binary
linear codes, PGIT 33 (1987), to appear [31.
[YetI] N. M. Vetchinkin, The packings of uniform n-dimensional balls that
are constructed from error-correcting codes, Ivanov. Gos. Univ. Ueen.
Zap. 89 (1974), 87-91 [5].
[Vet2] - Uniqueness of the classes of positive quadratic forms on which the
values of Hermite constants are attained for 6 " n " 8, TMIS 152
(1980) 34-86 - PSIM (1982, No.3), pp. 37-95 [1,4,61.
[VigIl M. F. Vigneras, Methodes analytiques, in [Orzl], 1977, pp. 340-360 [2].
[Vill] N. J. Vilenkin, Special Functions and the Theory of Group
Representations, Nauka, Nauka, Moscow, 1965; English translation,
Amer. Math. Soc., Providence RI, 1968 [9].
[Vinl] E. B. Vinberg, Discrete groups generated by reflections in Lobachevskii
spaces, Mat. Sb 71 (No.3, 1967), 471-488; 73 (1967), 303 - Sb. 1
(No.3, 1967),429-444 [4, 27, 28].
[Vin2] - Some examples of crystallographic groups in Lobachevskii spaces,
Mat. Sb. 78 (1969), 633-639 - Sb. 7 (1969), 617-622 [4, 27, 28].
[Vin3] - Discrete linear groups generated by reflections, IANS 35 (1970,
1072-1112 - Izv. 5 (1970, 1083-1119 [4,27,281.
[Vin4] - On the groups of units of certain quadratic forms, Mat. Sb. 87 (No.
1,1972),18-36 - Sb. 16 (No. I, 1972), 17-35 [4,27,28].
[Vin5] - On unimodular integral quadratic forms, Funkt. Analiz. Prilozen 6
(No.2, 1972),24-31 - Funct. Anal. Appl. 6 (1972),105-111 [4, 161.
[Vin6] On the SchOnflies-Bieberbach theorem, DAN 221 (1975), 1013-
1015 - SMD 16 (1975), 440-442 [41.
[Vin7] Some arithmetical discrete groups in Lobacevskii spaces, in
Discrete Subgroups of Lie Groups and Applications to Moduli, Oxford
Univ. Press, 1975, pp. 323-348 [4,27,28,301.
[Vin8] The nonexistence of crystallographic reflection groups in
Lobachevskii spaces of large dimension, Funkt. Analiz. Prilozen 15 (No.
2, 1980,67-68 - Funct. Anal. Applic. 15 (1980, 128-130 [4, 27].
[Vin9] Discrete reflection groups in Lobachevskii spaces of large
dimension, LOMI 132 (1982), 62-68 - JSM 30 (1985), 1837-1841 [4,
271.
[Vinll] - Discrete reflection groups in Lobachevskii spaces, in Proc. Intern.
Congr. Math. Warsaw 1983, North-Holland, Amsterdam, 1984, pp.
593-601 [4, 271.
[Vinl2] - On reflective hyperbolic lattices, DAN 271 (No.6, 1983), 1298-
1301 - SMD 28 (1983), 517-520 [4, 27].
[Vin13] - Absence of crystallographic reflection groups in Lobachevskii spaces
of high dimension, Trudy Moskov. Mat. Obshch. 47 (1984),68-102 and
246 [4, 271.
[Vinl4] - Hyperbolic reflection groups, Usp. Mat. Nauk. 40 (1985), 29-66
634 Bibliography

and 255 ~ Russ. Math. Surv. 40 (1985), 31-75 [4, 27].

[VinI5] - and I. M. Kaplinskaja, On the groups 018,1(Z) and 019.1(Z), DAN
238 (No.6, 1978) 1273-1275 = SMD 19 (No.1, 1978), 194-197 [4,27,
28].
[VitI] A. J. Viterbi and J. K. O'Mura, Principles of Digital Communication
and Coding, McGraw-HilI, NY, 1979 [31.
S. G. Vladuts: see also G. L. Katsman, M. A. Tsfasman.
[VIall On the exhaustion boundary for algebraic-geometric "modular"
codes, PPI, to appear [31.
[Vla2] - G. L. Katsman and M. A. Tsfasman, Modular curves and codes
with a polynomial complexity of construction, PPI 20 (No. I, 1984),
47-55 - PIT 20 (1984), 35-42 [3].
[Vla3] - and Y. I. Manin, Linear codes and modular curves, Itogi Nauk.
Tekh.2S (1984),209-257 - JSM 30 (No.6, 1985),2611-2643 [3].
[Vorl] G. F. Voronoi, Nouvelles applications des parametres continus d la
theorie des formes quadratiques, JRAM 133 (1908), 97-178; 134
(1908), 198-287; 136 (1909),67-181 [2,211.
[VosIl H, Voskuil, A special basis for the Leech lattice, KNAW A90 (1987),
73-86 [4].
C. de Vroedt: see C. Roos.
w
B. L. van der Waerden: see also W. Habicht, K. Schiitte.
[Wael] - Punkte auf der Kugel. Drei Zusatze, Math. Ann. 123 (1952), 213-
222 [11.
[Wae2] Moderne Algebra, Springer, Berlin, 2 vols., 1950; English
translation, Ungar, NY, 1953 [81.
[Wae3] - Die Reduktionstheorie der positiven quadratischen Formen, Acta
Math. 96 (1956), 263-309. Reprinted in part in [Wae5] [2, 151.
[Wae4] Pollen korner, Punktverteilungen auf der Kugel und
Injormationstheorie, Die Naturwissenschaften 7 (1961), 189-192.
[Wae5] - and H. Gross, editors, Studien zur Theorie der Quadratischen
Formen, Birkhaiiser, Basel, 1968 [2, 15, B1.
S. L. Wagner: see K. T. Smith.
D. B. Wales: see also A. R. Calderbank, J. H. Conway, M. Hall, Jr.
[Wall] - Uniqueness of the graph of a rank three group, PJM 30 (1969),
271-276 [101.
A. Z. Walfisz: see A. Z. Val'fi"&
G. E. Wall: see E. S. Barnes.
J. S. Wallis: see W. D. Wallis.
W. D. Wallis: see also A. Hedayat.
[WaI2] - A. P. Street and J. S. Wallis, Combinatorics: Room Squares, Sum-
Free Sets, Hadamard Matrices, LNM 292, 1972 [31.
[WanIl c.-S. Wang and A. Gersho, A bound on the number of nearest
neighbors of a Euclidean code, in Abstracts of Papers, IEEE Inter.
Symp. Info. Theory, St. Jovite 1983, IEEE Press, NY, 1983, p. 43 [2].
[Wan2] H.-C. Wang, Two-point homogeneous spaces, Ann. Math. SS (1952),
177-191 [91.
Y. Wang: see L. K. Hua.
H. N. Ward: see also F. J. MacWilliams, V. Pless.
[WarIl A form for M ll , J. Alg. 37 (1975), 340-351 [111.
[War2] - A restriction on the weight enumerator of a self-dual code, JCT
Bibliography 635

A21 (1976),253-255 [7, 191.

[War3) G. Warner, Harmonic Analysis on Semi-Simple Lie Groups, Springer-
Verlag, 2 vols., 1972 [91.
[War4) N. P. Warner, The symmetry groups of the regular tessellations of S2
and S3, PRS A383 (1982),379-398 [11.
S. Washburn: see E. Y. Miller.
[Wasl) A. J. Wasserman, The thirteen spheres problem, Eureka 39 (1978), 46-
49 [Il.
[Watl) D. F. Watson, Computing the n-dimensional Delaunay tessellation with
application to Voronoi polytopes, Computer J. 24 (1981), 167-172 [2).
G. L. Watson: see also H. Davenport.
[Wat2) The covering of space by spheres, Rend. Circ. Mat. Palermo 5
(1956),93-100 [2, 211.
[Wat3) - Integral Quadratic Forms, Camb. Univ. Press, 1960 [151.
[Wat4) Transformations of a quadratic form which do not increase the
class-number, PLMS 12 (1962), 577-587 [151.
[Wat5) - The class-number of a positive quadratic form, PLMS 13 (1963),
549-576 [IS).
[Wat6) Positive quadratic forms with small class-numbers, PLMS 13
(1963), 577-592 [151.
[Wat7) One-class genera of positive quadratic forms, JLMS 38 (1963),
387-392 [151.
[Wat8) - On the minimum of a positive quadratic form in n (~8) variables.
Verification of Blichfeldt's calculations, PCPS 62 (1966),719 [I, IS).
[Wat9) - On the minimal points of perfect septenary quadratic forms, Math.
16 (1969),170-177 [1,151.
[WatlO) The number of minimum points of a positive quadratic form,
Dissertationes Math. 84 (1971), 42 pages [1, 5, 151.
[Watll! On the minimum points of a positive quadratic form, Math. 18
(1971),60-70 [I, 151.
[Wat12) The least common denominator of the coefficients of a perfect
quadratic form, Acta Arith. 18 (1971),29-36 [15).
[WatI3) - The number of minimum points of a positive quadratic form having
no perfect binary section with the same minimum, PLMS 24 (1972),
625-646 [1, 151.
[WatI4) One-class genera of positive quaternary quadratic forms, Acta
Arith. 24 (1973),461-475 [151.
[Watl5) One-class genera of positive quadratic forms in at least five
variables, Acta Arith. 26 (1975), 309-327 [151.
[Watl6) One-class genera of positive ternary quadratic forms, Math. 19
(1972), 96-104; 22 (1975), 1-11 [151.
[Watl7) Transformations of a quadratic form which do not increase the
class number II, Acta Arith. 27 (1975), 171-189 [151.
[Watl8) The 2-adic density of a quadratic form, Math. 23 (1976), 94-
106 [151.
[WatI9) One-class genera of positive quadratic forms in nine and ten
variables, Math. 25 (1978), 57-67 [I, 151.
[Wat20) - Determination of a binary quadratic form by its values at integer
points, Math. 26 (1979), 72-75; 27 (1980),188 [2, 151.
[Wat21) One-class genera of positive quadratic forms in eight variables,
JLMS 26 (1982), 227-244 [151.
[Wat22) One-class genera of positive quadratic forms in seven variables,
PLMS 48 (1984),175-192 [151.
636 Bibliography

G. N. Watson: see also E. T. Whittaker.

[Wat23] The problem of the square pyramid, Messenger Math. 48 (1919),
1-22 [26].
[Wat24] G. S. Watson, Statistics on Spheres, Wiley, NY, 1983 [1].
[Wegll G. Wegner, Uber endliche Kreispackungen in der Ebene, Studia Sci.
Math. Hungar. to appear [!l.
[Weil] L.-F. Wei, Trellis-coded modulation with multi-dimensional
constellations, PGIT 33 (1987),483-501 [3].
V. K. Wei: see A. R. Calderbank.
S. B. Weinstein: see G. J. Foschini.
[Wei2] E. Weiss, Algebraic Number Theory, McGraw-Hili, NY, 1963 [8].
G. L. Weiss: see W. M. Boothby, R. R. Coifman.
[Wei3] R. Weiss, A geometric construction of Janko's group J], Math. Z. 179
(1982),91-95 [10].
[Wei4] On the geometry of Janko's group h Archiv Math. 38 (1982),
410-419 [10].
[Wei5] - A characterization and another construction of Janko's group J],
TAMS 298 (1986),621-633 [10].
A. J. WeIch: see C. E. Briant.
L. R. WeIch: see R. J. McEliece.
E. J. Weldon, Jr.: see also W. W. Peterson.
[Well] - Justesen's construction - the low-rate case, PGIT 19 (1973), 711-
713 [3].
[WeI2] A. F. Wells, Models in Structural Inorganic Chemistry, Oxford Univ.
Press, 1970 [J, 4].
[Wel3] Three-Dimensional Nets and Polyhedra, Wiley, NY, 1977 [1,4].
[WeI4] - Structural Inorganic Chemistry, Oxford Univ. Press, 5th ed., 1984
[1,2,4]'
[WeI5] - Survey of octahedral structures AXn and A 2X n, Phil. Trans. Roy.
Soc. London A312 (1980, 553-600 [I].
[WeI6] G. R. Welti, PCMIFDMA satellite telephony with 4-dimensionally-
coded quadrature amplitude modulation, COMSTAT Tech. Rev. 6
(1976),323-338 [3].
[Wel7] - and S. L. Lee, Digital transmission with coherent four-dimensional
modulation, PGIT 20 (1974), 497-502 [3].
[Wenl] M. J. Wenninger, Polyhedron Models, Camb. Univ. Press, 1971 [2,21].
[Wen2] - Spherical Models, Camb. Univ. Press, 1979 [2!l.
A. T. White: see N. L. Biggs.
J. W. White: see C. E. Briant.
D. S. Whitehead: see N. J. A. Sloane.
[Whill T. A. Whitelaw, On the Mathieu group of degree twelve, PCPS 62
(1966),351-364 [II].
[Whi2] Janko's group as a collineation group in PC (6,10, PCPS 63
(1967), 663-677 [10].
G. J. Whitrow: see H. S. M. Coxeter.
[WhtI] E. T. Whittaker and G. N. Watson, A Course of Modern Analysis,
Camb. Univ. Press, 4th ed., 1963 [4, 21].
[WhyI] L. L. Whyte, Unique arrangements of points on a sphere, AMM 59
(1952),606-611 [I].
M. M. Wiener: see L. Flatto.
[Wigll E. Wigner and F. Seitz, On the constitution of metallic sodium, Phys.
Rev. 43 (1933), 804-810 [2].
D. E. Williams: see Y. C. Lin.
Bibliography 637

J. M. Wills: see also G. Fejes T6th, P. Gritzmann, P. M. Gruber, J.

Tolke.
[WilO] - On the density offinite packings, AMAH 46 (1985),205-210 [II.
J. Wilson: see R. Askey.
R. A. Wilson: see also J. H. Conway, P. B. Kleidman, S. P. Norton.
[Will] - The quaternionic lattice for 2G 2 (4) and its maximal subgroups, J.
AIg.77 (1982),449-466 [8, 101.
[WiI2] The maximal subgroups of Conway's group '2, J. Alg. 84 (1983),
107-114 [10].
[Wil3] The complex Leech lattice and maximal subgroups of the Suzuki
group, J. Aig. 84 (1983), 151-188 [7, 101.
[WiI4] - The maximal subgroups of Conway's group COl, J. Aig. 85 (1983),
144-165 [101.
[WiI5] The geometry and maximal subgroups of the simple groups of A.
Rudvalis and J. Tits, PLMS 48 (1984), 533-563 [101.
[Wil6] - The subgroup structure of the Lyons group, PCPS 95 (1984), 403-
409 [101.
[Wil6a] - On lexicographic codes of minimal distance 4, Atti Sem. Mat. Fis.
Univ. Modena 33 (1984), 125-135 [111.
[WiI7] The maximal subgroups of the Lyons group, PCPS 97 (1985),
433-436 [10].
[Wil8] The maximal subgroups of the O'Nan group, J. Aig. 97 (1985),
467-473 [101.
[WiI9] - The geometry of the Hal/-Janko group as a quaternionic reflection
group, Geom. Dedic. 20 (1986), 157-173 [8, 101.
[WiIlO] - On maximal subgroups of the Fischer group Fi 22 , PCPS 95 (1984),
197-222 [10].
[Will 11 Maximal subgroups of automorphism groups of simple groups,
JLMS 32 (1985),460-466 [101.
[WiIl2] Is J I a subgroup of the Monster? BLMS 18 (1986), 349-350 [101.
[WilI3] Some subgroups of the Baby Monster, Invent. Math. to appear [101.
[WiI14] Some subgroups of the Thompson group, JAMS to appear [101.
[WiIl5] The 3-local subgroups of the Monster, JAMS to appear [101.
[WiIl6] The 3-local subgroups of the Fischer groups, JLMS to appear [101.
[WilI7] Maximal subgroups of sporadic groups, in Groups-St. Andrews
1985, ed. C. M. Campbell and E. F. Robertson, Camb. Univ. Press, to
appear [101.
R. M. Wilson: see also J. H. van Lint.
[WiIl8] - An existence theory for pairwise balanced designs III: proof of the
existence conjectures, JCT A18 (1975),71-79 [3].
[WiIl9] - On the theory oft-designs, in lJacl], 1984, pp. 19-49 [3].
[WiI20] S. G. Wilson, H. A. Sleeper and N. K. Smith, Four-dimensional
modulation and coding: an alternative to frequency-reuse, in Science,
Systems and Services for Communications, ed. P. Dewilde and C. A.
May, IEEE, NY, 1984, pp. 919-923 [31.
G. L. Wise: see J. A. Bucklew.
H. S. Witsenhausen: see R. L. Graham.
[wit11 E. Witt, Theorie der quadratischen Formen in beliebigen Korpern,
JRAM 176 (1937), 31-44 [15].
[Wit2] - Die 5-fach transitiven Gruppen von Mathieu, ASUH 12 (1938),
256-264 [3, 10, II].
[Wit3] - Uber Steinersche Systeme, ASUH 12 (1938), 265-275 [3, 10, II,
12, 141.
638 Bibliography

[Wit4] - Eine Identitiit zwischen Modulformen zweiten Grades, ASUH 14

(1941),323-337 [2,7].
[Wit5] - Spiegelungsgruppen und Aufziihlung halbeinfacher Liescher Ringe,
ASUH 14 (1941), 289-322 [4].
[WolO] A. Woldar, On the maximal subgroups of Lyons' group, Comm. in Alg.
15 (1987),1195-1203 [10].
[Woll] J. A. Wolf, Spaces of Constant Curvature, Publish or Perish, Boston
MA, 5th ed., 1985 [9].
J. K. Wolf: see T. Berger.
P. M. de Wolff: see H. Burzlaff.
[WoI2] J. Wolfmann, A permutation decoding of the (24, 12,8) Golay code,
PGIT 29 (1983), 748-750 [20].
[WoI3] J. Wolfowitz, The Coding Theorems of Information Theory, Prentice-
Hall, Englewood Cliffs NJ, 2nd ed., 1964 [3].
H. Wondratschek: see H. Brown, R. Biilow.
[Wooll A. C. Woods, Covering six-space with spheres, JNT 4 (1972), 330-341
[2].
[WorIl R. T. Worley, The Voronoi region of E~, JAMS A to appear [4, 211.
[Wor2] - The Voronoi region of E;, SIAD to appear [4, 211.
[Wozl] J. M. Wozencraft and I. M. Jacobs, Principles of Communication
Engineering, Wiley, NY, 1965 [3].
c. M. Wright: see E. L. Muetterties.
E. M. Wright: see G. H. Hardy.
[Wycl] R. W. G. Wyckoff, The Structure of Crystals, Chemical Catalog Co.,
NY, 1924 [4].
[Wyc2] - Crystal Structures, Wiley, NY, 2nd ed., 6 vols., 1965 [4].
[Wynl] A. D. Wyner, Capabilities of bounded discrepancy decoding, BSTJ 44
(1965),1061-1122 [II.
[Wyn2] Random packings and coverings of the unit n-sphere, BSTJ 46
(1967),2111-2118 [2].
[Wyn3] - Communication of analog data from a Gaussian source over a noisy
channel, BSTJ 47 (1968), 801-812 [3].
[Wyn4] - On coding and information theory, SIAM Rev. 11 (1969), 317-346
[3].
[Wyn5] - Fundamental limits in information theory, PIEEE 69 (1981), 239-
251 [3].
y
[Yagl] I. M. Yaglom, Some results concerning distributions in n-dimensional
space, Appendix to the Russian edition of [FejlO], Moscow, 1958 [9].
[Yaml] Y. Yamada, S. Tazaki and R. M. Gray, Asymptotic performance of
block quantizers with difference distortion measures, PGIT 26 (1980),
6-14 [2].
A. Yanushka: see E. Shult.
V. Y. Y orgov: see V. I. Iorgov.
[YosIl S. Yoshiara, The complex Leech lattice and the sporadic Suzuki group
(in Japanese), in Topics in Finite Group Theory, Kyoto Univ., 1982,
pp. 26-46 [7].
[Yos2] - The maximal subgroups of the sporadic simple group of O'Nan, J.
Fac. Sci. Univ. Tokyo 32 (1985),105-141 [10].
Z
[Zadl] P. L. Zador, Development and evaluation of procedures for quantizing
multivariate distributions, Ph.D. Dissertation, Stanford Univ., 1963 [2].
Bibliography 639

[Zad2] Asymptotic quantization error of continuous signals and their

quantization dimension, PGIT 28 (1982), 139-149 [2].
[Zagl] D. Zagier, Zetafunktionen und Quadratische Korper, Springer-Verlag,
1981 [15].
[Zar11 S. K. Zaremba, La methode des "bon treillis" pour Ie calcul des
integrales multiples, in Applications of Number Theory to Numerical
Analysis, ed. S. K. Zaremba, Ac. Press, NY, 1972, pp. 39-119 [11.
T. Zaslavsky: see P. D. Seymour.
H. J. Zassenhaus: see also H. Brown, A. G. Earnest, R. L. Graham.
[Zasl] - Neuer Beweis der Endlichkeit der Klassenzahl bei unimodularer
Jlquivalenz endlicher ganzzahliger Substitutionsgruppen, ASUH 12
(1938), 276-288 £14].
[Zas2] - On the sphere packing problem, Comm. Match in Math. Chern. 9
(1980), 127-135.
[ZeU] L. H. Zetterberg, A class of codes for polyphase signals on a
bandlimited Gaussian channel, PGIT 11 (1965), 385-395 [3].
[Zet2] - and H. Briindstrom, Codes for combined phase and amplitude
modulated signals in a four-dimensional space, COMM 25 (1977),
943-950 £3].
[Zim11 J. M. Ziman, Models of Disorder, Camb. Univ. Press, 1979 £1].
H. Zimmerman: see also H. Burzlaff.
[Zim2] and H. Burzlaff, DELOS -A computer program for the
determination of a unique conventional cell, Z. Krist. 170 (1985), 241-
246 [2].
T. Zink: see M. A. Tsfasman.
V. A. Zinoviev: see also L. A. Bassalygo, N. V. Semakov.
[Zinl] - A generalization of Johnson's bound for constant weight codes, PPI
20 (No.3, 1984), 105-108 [9].
[Zin2] - and S. N. Litsyn, Shortening of codes, PPI 20 (No.1, 1984), 3-11
- PIT 20 (1984), 1-7 [3,9].
[Zin3] - and S. N. Litsyn, Codes that exceed the Gilbert bound, PPI 21
(No.1, 1985) 109-111 [3].
[Zin4] - - A table of the best binary codes known, preprint £3].
[Zivl] J. Ziv, On universal quantization, PGIT 31 (1985), 344-347 [2].
[Ziv2] - and A. Lempel, A universe algorithm for sequential data-
compression, PGIT 23 (1977), 337-343 [2].
G. Zolotareff: see A. Korkine.
I. J. Zucker: see also M. L. Glasser.
[Zucl] - New Jacobian theta functions and the evaluation of lattice sums, J.
Math. Phys. 16 (1975), 2189-2191; 17 (1976), 853 [2].
 Index

The extensive bibliography on pages 572-639 is intended to serve as a more

complete author index; only a selection of names appears here. See also the list of
symbols on pages xiv-xvi.

A2 lattice, see hexagonal lattice A (n,d), 76, 245ff., 260jJ.

A 3 lattice, see face-centered cubic A (n,d,w), 246ff.
lattice A (n,<I», 25, 27, 248, 263jJ.
A; lattice, see body-centered cubic bounds for, 27jJ., 263ff.
lattice abbreviations, xiv, 92, 101
An group, see group, alternating Abel identity, 473
An lattice, 1Off., 94, 108ff., 406ff., Afflerbach, L., 41, 397
427jJ., 432, 443jJ., 449, 479jJ., algebraic number theory, 224
507 algorithm, 40, 402jJ., 443jJ.
decoding, 443ff. crystallographic, 41
definition, 108 decoding, 443jJ.
diagram, 109,456,481, 484jJ., 507 encoding, 443ff.
finding closest lattice point, 443jJ. LLL,41
fundamental simplex, 460 reduction, 41, 352, 396
glue vectors, 109, 507 Vinberg, 530, 547jJ.
Gram matrix, 109 ALTRAN,420
summary of properties, 108 analog-to-digital conversion, v, 31, 56,
theta series, 110 63,443ff.
Voronoi cell, 459jJ. Anderson, J. B., 74
An [s I lattice, 116 Andrilli, S., 298
A: lattice, 36, 60, 115, 449, 472jJ. antenna design, 11
decoding, 447 antipodal, 24, 343
Gram matrix, 115 antisquare, p-adic, 370
is good covering, 36, 115, 116 antisquare term, 383
sublattices, 116 approximation problem, 12
summary of properties, 115 argon, 2
thickness, 36, 115 arithmetic progression loop, 327
Voronoi cell, 115, 472jJ. Artin, E., 267
see also hexagonal lattice A2 Askey, R., 256, 266
see also body-centered cubic lattice Assmus, E. F. Jr., 76, 196
An-tree, 479, 488, 489 association scheme, 246, 251, 266, 342
642 Index

Astola, J. T., 28, 264 Beenker, G. F. M., 205

Atkin, A. O. L., 568 Benson, D. J., 300
autochronous, 527 Berlekamp, E. R., 76, 183
automorphism, affine, 91, 98 Best, M. R., viii, 76, 140, 185
autochronous, 412 binary symmetric channel, 66,
diagonal, 556 75
fixed-point-free, 54 binary weight, 83
group, 77, 90 binomial coefficient, 355
integral, 390 biology, 11,25
monomial, 77, 561 bipyramid, 36
outer, 269./J., 323 Blahut, R. E., 76
standard, 556 Blichfeldt, H. F., 12, 162,247
axis, 495 blob trick, 310
block design, 88
Bochner, S., 254, 257
Bn root system, 99 body-centered cubic (bcc) lattice
Babai, L., 41 A; ;;;; D;, II, 32, 34, 36, 60,
ball bearings, 1 116, 471
Bambah, R. P., 32, 36 covering radius, 34, 116
bandwidth, 67 definition, 116
Bannai, E., xi, xii, 90, 205, 266, 340 Delaunay cell, 36
Baranovskii, E. P., 11,36,397 holes in, 34
Barnes, E. S., 13, 36, 62, 129, 158, is good covering, 32
159,232,244,475 is good quantizer, 60
Barnes-Wall lattice A 16 " BW16 , 12, packing radius, 34, 116
13,94, 129./J., 219, 241 summary of properties, 116
as laminated lattice, 159, 173 theta series, 116
covering radius, 130, 175 Voronoi cell, 34, 116,471
deep holes, 175 Borcherds, R. E., xii, 49, 421, 510,
definition, 129./J. 513, 550, 568
from Construction C or D, 131 BorOczky, K., 28, 62, 339
from Leech lattice, 131 Bos, A., 234, 241
from Reed-Muller code, 129, 142, Bos, R. C., 81
191,234 bound, 14./J., 23./J., 37./J., 58jJ., 245./J.
generator matrix, 130 Astola, 28, 264
A 16 /2A 16, 174 Blichfeldt, 247
sections, 166 Coxeter, Few, Rogers, 40, 62
summary of properties, 129 Coxeter-BorOczky, 28, 62, 339
theta series, 130, 131 Delsarte, Goethals, Seidel, 28
Barnes-Wall lattices BWn , 16, 24, 152, for A (n,d), 260./J.
206,234 for A (n,d, w), 263, 265
definition, 234 for A (n,cf», 27./J., 263./J.
kissing number, 24, 151 for covering, 37./J.
see also Barnes-Wall lattice A16 for kissing number, 23./J., 337./J.
basis, 4, 42, 355 for packing, 14./J.
integral, 355 for quantizers, 58./J.
reduced, 41, 396 Gilbert, 155, 196,246,247
simplicial, 222 JPL, 156,246,247,263
Bateman, P. T., 108 Kabatiansky-Levenshtein, 19, 28,
Bayer-Fluckiger, E., 135, 194,229 190, 245, 247./J., 264./J.
bec lattice, see body-centered cubic Levenshtein, 19,247
lattice Lindsey, 3, 13, 15
Index 643

bound (cont.) Choi, C., 279

linear programming, 28, 246jJ., circumradius, 33
257jJ., 337, 341jJ. class, 355
Lovasz, 265 class field tower, 18, 206, 227
McEliece, Rodemich, Rumsey, class number, 403
Welch, 156,246,247,263 classification of, extremal Type I
Minkowski, 14, 19, 196,229,247 codes, 190, 440
Plotkin, 246, 262 extremal Type I lattices, 190, 439jJ.
Rankin, 27, 247, 264 extremal Type II codes, 194
Rogers, 3, 13jJ., 19, 39, 62 extremal Type II lattices, 194
Sidel'nikov, 247 extremal Type III codes, 205
Singleton, 81 extremal Type IV codes, 205
Tietiiviiinen, 262, 264 forms of determinant 2 or 3, 385,
Tsfasman, Vladuts, Zink, 87, 266 387, 399jJ.
union, 70 genera of forms with small
Varshamov, see bound, Gilbert determinant, 385jJ.
Zador, 58jJ., 452 integral lattices, v, 48, 49, 352jJ.,
Brillouin zone, see Voronoi cell 402jJ., 406jJ., 421jJ., 427jJ.,
Broue, M., 182, 183, 193 439jJ.
Brouwer, A. E., 76 p-elementary forms, 386, 388
Buda, R. de, 68, 74 positive definite lattices, 48, 49,
burst, 74 396jJ., 406jJ., 421jJ., 439jJ.
Butler, G., 298 self-dual Eisenstein lattices, 204, 205
self-dual Gaussian lattices, 204, 205
Type I codes, 189
Coset, 131,304,333,480,556 Type I lattices, 189, 385, 406jJ.,
C. root system, 99, 458 421jJ.
Calderbank, A. R., 74, 93, 266 Type II codes, 193, 433
capacity, 67 Type II lattices, 193jJ., 406ff., 427jJ.
cardinal series, 64 Type III codes, 204, 433
Carmichael, R. D., 307 Type IV codes, 204
Cartan subalgebra, 570 cluster, spherical, 95
Carter, R. W., 298 cocode, 556
Cassels, J. W. S., 187, 355, 378, 388, code, vi, 24, 63jJ., 75, 213, 257,433
390,397 algebraic geometry, 87
Cayley number, 122,251 automorphism group of, 77
integral, 122 BCH, 81, 152,206,235
Cayley projective plane, 251, 256 best, 152,247,248
cell, Voronoi, see Voronoi cell Best, viii, 87, 140, 141, 185
center density ~, 13, 32, 101 binary, 75
centroid, 58, 449 block,75
channel,63 bounded energy, 95
bandlimited, 47, 67 conference matrix, 87
binary symmetric, 66, 75 constant energy, 69, 95
coding problem, vi, 69 constant weight, 76, 246, 249
decoder, 63, 443jJ. constructed from a group, 92
encoder, 63, 443jJ. convolutional, 74, 75
character, 186 covering, 76
characteristic vector, 88 cyclic, 79
Chaundy, T. W., 158, 159 dimension of, 76
chemistry, vi, 11 double circulant, 86
Chen, C. L., 76 doubly even, see code, Type II
644 Index

code (cont.) RM, see code, Reed-Muller

dual, 77, 183, 245 self-dual, 77, 181, 185ff.
weight enumerator of, 78 soft-decision decoding of, 447
equivalent, 77 spherical, 24jJ., 69, 95, 248ff., 337,
even weight, 79 340
extended, 79 construction of, 26jJ.
extended Hamming, 80jJ. Sugiyama et at., 27, 83
extremal, 182, 189 ternary, 75
for band-limited channel, vi, 67jJ., tetra code, see tetra code
443, 449 trellis, 74
generalized, 257 two-weight, 251
glue, 406, 407, 508 Type I, 77, 182, 185ff.
Goethals, 152 Type II, 77,181,182, 191jJ., 433
Golay, see Golay codes Type III, 77, 182,202,433
Golay nonlinear, 87, 139 extremal, 204
Goppa,87 Type IV, 77, 182,203
group, 74, 76, 93 extremal, 204
Hadamard, 87 universe, 79
Hamming, 79, 81, 86, 139, 186, 194, Weldon, 27, 83
233, 408, 435, 448 zero, 79
hexacode, see hexacode zero-sum, 79
Julin, 87, 139 zeros of, 80
Justesen, 27, 82, 155 codeword, 75
Kerdock, 87,191 coding gain, 20, 73, 74
lattice, 68, 71jJ. Cohen, G. D., 76
lexicographic, 162, 327 Cohn-Vossen, S., 11
line-, 317 col (or column), 321
linear, 76 compartment, 381
mass constants for, 409 complexity, 40, 402
maximal distance separable, 81, component, 100,406
239jJ. composition, 364
MDS, 81, 239jJ. computational complexity, 40, 402
[n,k), 76 computationally effective, 404
[n,k,d),76 conjugate, 52, 77, 208
(n,M,d),75 constituent, Jordan, 378, 381, 383
Nadler, 261 construction A, 136jJ., 181jJ., 206,
nonlinear, 87, 139, 195 211jJ., 233, 408, 422, 448
Nordstrom-Robinson, 87, 190, 195 decoding, 448
of length 40, 146, 194 theta series of, 183
orthogonal, 77 A 3, 148
perfect, 85 A" 197
permutation, 74 A r, 206, 211jJ., 215, 217
Pless, 86, 148 Aq" 217jJ.
point-, 317 B, 136, 141jJ., 181, 191ff., 233
Preparata, 87, 191 B3, 148
q-ary, 75 Be, 198
quadratic residue, 84, 149, 308, 436 C, 136, 150jJ., 206, 232, 235
Rao-Reddy, 234 D, 206, 232jJ., 239
Reed-Muller, 24, 83, 129, 139, 146, D',235
151,234,236,447 Davenport, 35, 37jJ., 40
Reed-Solomon, 81, 220 E, 206, 236ff.
repetition, 79 holy, 506ff., 522
Index 645

construction (cont.) 529, 541, 552, 570

Turyn, 132, 206, 316, 317 conventions for, 96, 98, 456-457,
Witt-Tits, 315 528-529
Wythoff, 463jJ. extended, 479, 481
contact number, see kissing number of a hole, 480
conventions for diagrams, 96, 98, 456- ordinary, 478, 481, 519
457,528-529 Coxeter-Todd lattice K I2 , 13jJ., 94,
conventions for groups, 92 127jJ., 139, 146, I 59jJ., 198,
Conway group Coo - ·0, ix, 91, 134, 241
267, 331,485 as Eisenstein lattice, 128, 162, 198,
definition, 286jJ. 200
invariants, 93 decoding, 447
involutions, 294 definition, 128, 198
order of, 91, 289, 335 dual lattice ,;;; K 12, 128
subgroups, 290jJ. from Leech lattice, 128
Conway group COl - ·1, 91, 290, 296, group, 127jJ.
297, 561 norm-doubler, 241
definition, 290 sections, 166
subgroups, 298 summary of properties, 127
Conway group CO 2 - ·2, 91, 290, 296, theta series, 129, 204
550 Coxeter-Vinberg diagram, 541, 552
definition, 290 Craig, M., 17,224,227,229,244
Conway group C0 3 - ·3, 91, 290, 293, cross-section, 136, 145, 163
296 crosspolytope, 452
definition, 290 crystallographic algorithms, 41
Conway group Co .. - ·00, 290, 485, crystallography, vi, 3, 41
528, 569 cube,450
coordinate array, 137 snub,25
coordinate frame, 169, 289, 480 truncated, 464
coordination number, see kissing cubic closest packing, see face-centered
number cubic lattice
Costello, D. J., 76 cubic lattice Z', 42, 46, 48, 51, 94,
couples, 300 106jJ., 184, 443jJ.
covering density, 31, 32, 69, 76 definition, 106
covering problem, v, 32, 36jJ. finding closest lattice point, 443jJ.
summary of results, 36jJ. quadratic form, 42
covering radius, 6, 33, 40, 53, 76, 101, summary of properties, 106
163 theta series, 46, 106, 107, 184
algorithms for, 41 see also one-dimensional lattice Z
coverings in 4 and 5 dimensions, 36, 37 see also square lattice Z2
coverings, bounds on, 36jJ. cuboctahedral arrangement, 21
Coxeter, H. S. M., 13, 18, 28, 40, 96, cuboctahedron, 21,464
119, 121, 127, 159, 298, 451, Curtis, R. T., 279, 298, 299, 303, 307,
475, 523, 527 479, 523
Coxeter diagram, see Coxeter-Dynkin cycle, 353
diagram of reduced forms, 356
Coxeter element, 98
Coxeter group, 95, 412, 527, 530, 569
Coxeter number, 99, 101, 407, 408, D3 lattice, see face-centered cubic
425, 429, 510, 514, 519 lattice
Coxeter-Dynkin diagram, vi, x, 11, 96, D; lattice, see body-centered cubic
98, 101, 456jJ., 478jJ., 507, 519, lattice
646 Index

D4 lattice, 9jJ., 26, 55, 118, 164, 215, Delone, B. N., 35, 36
216, 495jJ., 507 Delsarte, P., 28, 90, 194, 245, 258
as Gaussian lattice, 202, 215 ~n' 247
as Hurwitzian lattice, 55, 119 density, 2, 7, 10,31,46, 101
as laminated lattice, 158jJ. center, 13,32, 101
constructions, 118, 138 doubling, 145, 149
deep holes, 26, 119,462 from theta series, 46
definition, 9, 118 designed distance, 81
density, 10, 119 determinant, 4, 5, 53, 101, 356
determinant, 9 conditions, 382
diagram, 117,457,481, 495jJ., 507 Diaconis, P., 266, 323
kissing number, 9, 22 diagonalization, 369
D 4 /2D 4,169 diagram, Coxeter-Dynkin, see
sections, 166 Coxeter-Dynkin diagram
summary of properties, 118 diamond structure Dt, 8, 46, 113, 119
theta series, 119, 202 definition, 113
Voronoi cell, 119,462 theta series, 46, 113
see also Dn lattice Dickson, L. E., 36,62,267,403
Ds/2Ds, 169 dictators, 25
Dn lattice, 10jJ., 94, 117jJ., 184, 406jJ., dictionary, 201, 328, 560
415,432, 443jJ., 449 different, 230
checkerboard lattice, 117 digital communications, 11, 56
decoding, 443jJ. dimension, 3, 76 379, 380
definition, 10 direct sum, 10 1
Delaunay cell, 144 Dirichlet, P. G. L., 33, 372,451
diagram, 117,457,481, 495jJ., 507 Dirichlet region, see Voronoi cell
finding closest lattice point, 443jJ. Dirichlet integral, 451
fundamental simplex, 461 discriminant, 4, 225, 356
glue vectors, 117, 507 disjoined, 480
summary of properties, 117 distance distribution, 257, 341
theta series, 46, 118, 184 distance, Hamming, 75
Voronoi cell, 462 distance invariant, 45
see also face-centered cubic lattice divisor, 366
D3 dodecad, 279, 308, 480
see also D4 lattice counts, 298
Dn-tree, 479, 496 dodecahedron, 25, 450, 455
Dn+ lattice, 119 rhombic, 34, 450, 462
D: lattice, 120, 449, 463jJ. · 0, see Conway group Coo
decoding, 447 '1, see Conway group COl
Voronoi cell, 463jJ. · 2, see Conway group CO 2
· 00, see Conway group Co~
Damerell, R. M., 90
data compression, v, 56 Downey, D. E., 76
data storage, vi, 63 duad, 269, 279, 495
data transmission, vi, 63 dual code, 77, 245
Davenport, H., 37 dual lattice, 10, 53, 99
Dawson, E., 205 dual problem, 258, 341
decoding algorithms, 41, 443jJ. dual quotient, see group, dual quotient
decoding error, 67 dual theory, vii, 12
deep hole, see hole, deep Dunkl, C. F., 266
Delaunay, B. N., 35, 36 duum, 279
Delaunay cell, 35, 144, 513 Dynkin diagram, see Coxeter-Dynkin
Deligne, P., 51 diagram
Index 647

E6 lattice, 1Off., 55, 94, I 25ff., 144, diagram, 121, 457, 481, 502ff., 507
164, 200, 227, 406.fJ, 427ff., E8/2E8, 169, 170
432, 449, 461ff., 507 E 8/3E 8,214
as Eisenstein lattice, 55, 126, 200 even coordinates, 120
as laminated lattice, 144, I 58ff., finding closest lattice point, 443ff.
162, 164 from Construction B, 142
cyclotomic construction, 227 from Hamming code, 121, 139, 185,
decoding, 447 233
definition, 125, 126, 200 fundamental simplex, 461
diagram, 126,457,481, 502ff., 507 group, 121,294
E6/2E6, 169 holes, 121, 170
fundamental simplex, 461 identified with icosians, 208ff.
glue vectors, 126, 507 initial shells, 123
summary of properties, 125 minimal vectors, 121, 209
theta series, 127, 200 norm-doubler, 240-241
uniqueness, 13 odd coordinates, 120
Voronoi cell, 462 optimality of kissing number, 12, 22,
E; lattice, 127 337, 338, 342
as Eisenstein lattice, 127 orbits of vectors, 123
decoding, 447 other constructions, 143
E7 lattice, I Off., 94, 124, 125, 164, quadratic form, 44
406ff., 427ff., 432, 449, 46 Iff., sections, 166
507 summary of properties, 120
as laminated lattice, 144, I 58ff., theta series, 46, 122, 123, 185, 188,
162, 164 192
construction, 124, 138 uniqueness, 13, 121, 340ff.
decoding, 447 Voronoi cell, 121,462
definition, 124 Weyl group W(E 8), 121,294
diagram, 124,457,481, 502ff., 507 Earnest, A. G., 389
E7/2E7, 169 Edge, W. L., 121
from Hamming code, 138, 185 edge-length, 451
fundamental simplex, 461 Edwards, H. M., 356, 366
glue vectors, 125, 507 Eichler, M., 352, 353, 356, 388, 390
summary of properties, 124 8-cage, 549, 553
theta series, 125 Eisenstein integer, 52, 197ff.
uniqueness, 13 elementary p.d.k., 257
Voronoi cell, 462 II-design, spherical, 196, 346
E; lattice, 125
empty component, 414
endomorphism, 236
decoding, 447
Enguehard, M., 182, 183, 193
E8 lattice, v, vi, 1Off., 22, 48, 90, 93,
entropy, 197, 246
94, 119, 120ff., 142, 146, 164,
equation, Diophantine, 11, 44
188, 192, 206, 240, 338, 340ff.,
Laplace, 256
406ff., 427ff., 432, 439, 449,
error probability, 69
461ff., 507, 523
error-correcting code, see code
as Eisenstein lattice, 200
error-correcting code problem, 76
as Gaussian lattice, 202
Estes, D., 366
as Hurwitzian lattice, 55
Euclidean Lie algebra, 570
as laminated lattice, 144, 158ff.,
eutactic star, 176
162, 164
Evans, D. M., 298
cyclotomic construction, 229
even interpretation, 304
decoding, 443ff.
explicit construction, 18
definition, 120ff.
648 Index

extending node, 98, 458 bilinear, 207, 211, 252, 354

extraspecial, 284 binary quadratic, 354, 357ff., 373,
extremal, 182, 189, 193, 204 404
extreme, 44, 182,227 definite, 357ff.
equivalent, 357
indefinite, 359ff.
f -point, 326 numbers represented by, 374
F4 root system, 99 classically integral, 355
face, 451 composition of, 364ff.
face-centered cubic (fcc) lattice cusp, 187,428
A3 == D 3, 1, 2, 9, 21, 32, 34, 36, existence of with prescribed
112, 143, 164, 182,462 invariants, 372, 382ff.
as laminated lattice, 143, 158, 164 extremal, 49,182,189,193,204
covering radius, 34, 112 extreme, 44, 182, 227
definition, 112 Hermitian, 215, 252
Delaunay cell, 36 indefinite, 353, 378, 403
glue vectors, 112 integral, 47, 48, 53, 355, 522
holes, 34, 112, 144 isotropic, 375
is good packing, 2, 9 Lorentzian, see lattice, Lorentzian
A/2A,168 Minkowski reduced, 358, 396ff., 403
packing radius, 34, 112 modular, 50,181,186,187,331,428
theta series, 112, 113 numbers represented by, 402
Voronoi cell, 34, 112,462 p-elementary, 353, 386
see also An lattice positive definite, 43
Falconer, D. D., 74 properly primitive, 364
fcc lattice, see face-centered cubic quadratic, v, 31, 42, 207, 211, 354
lattice associated with lattice, 42, 352
feasible solution, 259 binary, 354, 357ff., 373ff., 404
Feit, W., 204, 229 diagonalization of, 369
Fejes T6th, G., II indefinite, 43
Fejes T6th, L., 8, 11, 60 integral, 31, 353ff., 355
Few, L., 40 integral invariants for, 353
field, algebraic number, 206 integrally equivalent, 43
cyclotomic, 227 Lorentzian, see lattice, Lorentzian
F9 , 217 over a ring, 356
Galois, 75 p-adic invariants for, 353, 373ff.
golden, 207 rational invariants for, 370ff.
of rational numbers, Q, 366 see also lattice
totally complex, 225, 227 reduced, 358
totally real, 225 cycle of, 356ff.
finite packings, 21 standard, 373
Finkelstein, L., 298 Type I, 380
Fischer, B., 295, 297 see also lattice, Type I
5-design, 196 Type II, 380
5-problem, 302 see also lattice, Type II
fixer, 310 unit, 378
Flatto, L., 186 Forney, G. D. Jr., 73, 74
Hipper, 3 10 48-dimensional lattices P 48p and P 48q'
Fong, P., 568 52, 149, 194, 195, 243
form, absolutely extreme, 44 cyclotomic construction, 229ff.
alternating bilinear, 252 definitions, 149, 195
ambiguous, 366 kissing numbers, 52, 195
Index 649

P 48p and P48q (cont.) Gitlin, R. D., 20

norm-doubling map, 243 Gleason, A. M., 78, 186, 192, 202
sections, 166jJ. glue, 95, 99, 353, 399, 406
summary of properties, 195 code, 406, 407, 508
theta series, 52, 195, 196 group, 100, 101
4-problem, 321 self-, 100
Fourier expansion, 253 vector, 99, 101, 406, 506, 524, 570
free, 53 gluing theory, 95, 99, 353, 399, 406
Frenkel, I. B., 568 Goethals, J.-M., 90, 152, 194, 195,
function, bent, 130 205, 262
Bessel, 19 Golay, M. J. E., 84, 87, 139
entropy, H 2 (x), 197,246 Golay cocode C·, 556
error, 70 Golay codes C II and C 12 (ternary), x,
modular, 513, 568 85jJ., 142, 200, 271, 294, 320jJ.,
positive definite, 254 408, 434, 511
Riemann zeta, 14 definition, 85, 271
Schliifti, 28, 62 group of, 85, 271jJ.
zonal spherical, 249jJ., 254jJ. weight enumerator, 85, 202
fundamental domain, 4, 97, 458, 528 Golay codes C 23 and C 24 (binary), ix,
parallelotope, 4 x, 27, 84, 86, 131, 146, 191,
region, 4, 97, 458, 528 194, 261, 286, 299, 326, 327,
root, 97, 528, 531, 532, 547 408, 434, 438, 443, 511, 554,
simplex, 456 556
vector, 506jJ., 523 definition, 84, 275, 308
group of, 85, 275
Turyn construction, 317
g-point, 326 weight enumerator, 84, 192,278
G 2 root system, 99 Golod, E. S., 18,206,227
Gn , 58 Goppa, V. D., 87
Gallager, R. G., 74 Gorenstein, D., 298
Galois, E., 75, 268 Gosset, T., 120
Galois field, 75 Graham, R. L., 76, 265, 323
Gauss, C. F., 8, 352, 353, 355, 356, Gram matrix, 4, 11, 42, 101
359, 366, 372, 404 graph, distance-transitive, 251
Gaussian channel, 47, 67 8-cage, 549, 553
Gaussian integer, 52 Higman-Sims, 549
Gelfand pair, 251 McLaughlin, 293
genera, see genus of best codes, 247
generalized Riemann hypothesis, 229 of covering thickness, 36
genus (pl. genera), 352, 366, 378 of Eisenstein integers, 199
invariants for, 378jJ., 403 of error probability of lattices, 72
one-class, 354, 393jJ. of inclusions among laminated
spinor, see spinor genus lattices, 159
symbol for, 384jJ. of laminated lattices, 14, 36, 159
geometry, Higman, 292 of neighbors of Type II lattices, 423
hyperbolic, see hyperbolic space of packing densities, 14
of numbers, vi, 11, 397 of second moment of quantizers, 60
orthogonal, 213, 251 Petersen, 549
symplectic, 251 strongly regular, 251
unitary, 251 Green machine, 447
Gersho, A., 446, 475 Gregory, D., 21
Gilbert, E. N., 246 Griess, R. L. Jr., 91, 296, 297, 554
650 Index

Grothe, H., 397 L 2 (9) ;;0 A6, 268, 272, 296

group, affine Weyl, 98 L 2 (P),267ff.
alternating, An' 268 L 4 (2) ;;;; As, 268, 270
autochronous, 527 Lyons, Ly, 298
automorphism, 49, 77, 90, 101 MIQ,272
baby Monster, B = F2+, 297, 562 massive, 253
compact, 250 Mathieu, see Mathieu groups
complex reflection, 192 maximal subgroups, 298
conventions for, 92 McLaughlin, MeL, 291ff., 296
Conway, see Conway groups modular, 331, 568
Coxeter, 95, 412, 527, 530, 569 monad,279
crystallographic, 97 monomial, 77
doubly transitive, 250 Monster, see Monster group
duad, 279 N, 268, 287, 557
dual quotient A' / A, 48, 100 O'Nan, O'N, 298
duum, 279 octad, 279, 300, 311
elementary abelian, 291 octern, 279, 308, 318
Euclidean reflection, 98 of lattice formed by gluing, 100
exceptional, vi, 267, 296 orthogonal, 388
extraspecial, 291 projective L 2 (23), 279, 308
finite reflection, 95 projective special linear,
classification of, 96, 98 Ln(q) = PSLn(q), 267ff.
Fischer, 295, 296, 299 quintuply transitive, 271, 275
Fischer-Griess, see Monster group reflection, 95, 203, 528
Friendly Giant, see Monster group Rudvalis, Ru, 298
Galois, 227 sextet, 279, 300, 309
general linear, GLn (q), 267 simple, 91, 267, 298
glue, 48, 100ff. special orthogonal, SO(n), 250, 388,
Hall-Janko, HJ = h 210, 293, 297, 390
298 spherical reflection, 95
Harada-Norton, HN - F5+, 297 sporadic, vi, 267, 296
Hecke, 220 Suzuki, Suz, 293, 297
Held, He, 297 symmetric, S n, 82, 90, 250
Hessian, 203 Thompson, Th, 297
Higman-Janko-McKay, J 3 , 298 triad, 279, 300, 314
Higman-Sims, HS, 93, 291ff. trio, 279, 300, 316
hyperbolic reflection, 529 U 6 (2),296
hyperoctahedral, 250 Vinberg, 532
icosian, 207 Weyl, 98, 408, 432, 456, 458
ideal class, 365 see also under An, D n , £6, £7, £s
isomorphisms of, 268 lattices
Janko, It. 267, 273ff., 298 Gruber, P. M., 11, 397
Janko, J 2, see group, Hall-Janko Gunning, R. C., 187
Janko, J 3, see group, Higman- Guy, M. J. T., 327
Janko-McKay
Janko, J 4, 298ff.
Klein, V 4 , 118 H 2(X), 197, 246
linear fractional, 267ff. Hadamard matrix, 85, 87, 138, 139,
L 2 (3) ;;;; A4, 268, 269 221
L 2 (4) .. L 2 (5) ;;0 As, 268, 269 Hadwiger, H., 176
L 2 (7) ;;0 L 3 (2), 80, 268, 269ff., 279, Hahn, T., 11
488 half-cube, 144
Index 651

half-space, 451, 475 definition, 5, 110

Hall, M. Jr., 298 dual quotient group, 48, 110
Hammer, J., 11 glue vectors, 110
Hamming, R. W., 75, 79, 81 group of, 91
Hamming distance, see distance, quadratic form, 42, 43
Hamming summary of properties, 110
Hamming weight, see weight, theta series, 110Lr.
Hamming Voronoi cell, 6, 33
Harary, F., 549 Higman, D. G., 292
Harm, 255 Higman, G., 292, 298
harmonic analysis, 245 Hilbert, D., II
harmonic sum, 62 Hilbert norm residue symbol, 352, 353,
Hasse, H., 352, 370 370, 373
Hasse principle, strong, 372 Hirasawa, S., 27, 82
weak, 372 Hlawka, E., 15
Hasse-Minkowski invariant, 373 Hocquenghem, A., 81
Hauptmodul, 568 hole, 33, 163
Hazewinkel, M., II deep, 6, 26, 33, 142, 163, 407, 478,
hcp, see hexagonal close packing 510,512,525, 569
head character, 568 environs of, 510
Hecke, E., 187, 192 group of, 512
height, 529, 548, 569 diagram, x, 481, 511, 524
Held, D., 297 shallow, 33, 570, 571
Helgert, H. J., 76 see also under individual lattices
helium, 8 holy construction, 506
Hermite's constant, 20, 74 homogeneous minimum, 44
Hesselink, W., 11 honeycomb, 119
hexacode C 6, ix, 82, 128, 198, 203, Hsia, J. S., 194,389
300Lr., 327, 408, 436 Huffman, W. c., 53, 186
completion of, 302 Hurwitz integral quaternion, 52, 119
definition, 82, 300Lr. Husemoller, D., 11
justification of, 302 hyperbolic coordinates, 532
weight distribution, 82, 203, 301 hyperbolic space, vi, x, 11, 12, 29, 43,
hexacodeword, 82, 300 91,326, 522Lr., 527Lr.
hexad, 321 hyperconic, 315
heavy, 325 hypercube {4,3,3}, 451
light, 325 hyperplane, 451,475
signed, 321
hexagonal close-packing (hcp), 8, 113,
185 icosahedral arrangement, 21
built by layers, 143 icosahedron, 21, 25, 329, 450, 455
definition, 114 icosian, 122, 206, 207, 209
holes, 114 ideal, 364
theta series, 114 class, 364, 365
hexagonal lattice A 2, xiii, 6, 8, 31, 32, complementary, 230
110, 164,460 integral, 225
as Eisenstein lattice, 55 norm of, 364
as optimal code, 71 normed,364
as optimal covering, 31, 32 oriented, 364
as optimal packing, 5Lr. principal, 226, 364
as optimal quantizer, 60 normed,364
deep holes, 6 identity, Abel, 473
652 Index

identity (cont,) 04(z),103ff.

Jacobi, 103, 186, 440 Janko, Z., 267, 273, 298
MacWilliams, 78, 103, 186, 245, Johnson, S. M., 249
258 Jones, B. W., 378
Newton, 224 Jordan constituent, 378, 381, 383
Poisson summation, 103 Jordan decomposition, 378, 381
Poisson-Jacobi, 103 equivalences between, 381ff.
theta series, 47, 102ff. Julin, D., 139
improper, 355 Justesen, J., 27, 82, 155
incident, 317 justifica tion, 302
indicator vector, 88
inequality, arithmetic-geometric, 226
Elias, 263, 265 k -parity, 151
mean, 259, 260 Kn lattice, see lattice, Kn
Mordel1, 10, 12, 167 Kabatiansky, G. A., ix, 19, 23, 245,
information source, 63 247,264
inhomogeneous minimum, 44 Kac, V. G., 569
inradius, 33 Kac-Moody Lie algebra, 569
integer, algebraic, 225 kaleidoscope, 95
cyclotomic, 215 Kantor, W. M., 244, 266, 323
Eisenstein, 52, 197ff. Kaplinskaja, I. M., x, 527, 532, 541
Gaussian, 197, 202 Karpovsky, M. R., 76
integral, as a quadratic form, 355 Kasahara, M., 27, 83
as a symmetric bilinear form, 355 Kassem, W., 68
Dirichlet, 451 Kerdock, A. M., 87, 191
integral1yequivalent, 355 kernel, 557
improperly, 355 Kershner, R., 32
properly, 355 Kilby, K. E., 76
integration, numerical, vi, II, 12, 89, kissing number, 21, 23, 24, 27, 46, 94,
93 99, 138
intersymbol interference, 74 bounds on, 23ff., 337ff.
interval of forms, 381 definition, 21
invariant, 186 problem, v, ix, 21, 337
Hasse-Minkowski,373 summary of results, 21
theory, 93, 186 tables of, 12, 15, 16, 23
2-adic, 380 see also under individual lattices
Iorgov, V. I., 194 Kitaoka, Y., 47
isobaric, see polynomial, isobaric Kleidman, P. B., 298
isometry, 90 Kneser, M., 47, 49, 352, 353, 389, 399,
isotropic vector, 523 439
Koch, H. V., 194
Koornwinder, T., 120,266
J -lattice, 53 Korkine, A., 10, 12
J 12 , 159
Jacobi, K. G. J., 103, 186
Jacobi identity, 103, 186,440 L 2(G),252
Jacobi theta function, 46, 95, 102ff. L 2(M),252
product expansion of, 105 Ln (q), see group, projective special
0; (z), 102, 105 linear
02(Z), 46, 102ff. laminated lattice An' ix, vi, 13, 14, 95,
OJ (z), 45, 102ff. 142, 145, 157ff.
oJ(rlz), 46,102,104 complex, 160
Index 653

laminated lattice (cont.) dual, 10, 53, 99

definition, 163 E 6, see E6 lattice
density, 14, 162 E 7, see E 7 lattice
determinant, 162 E 8, see E8 lattice
integral, 160 Eisenstein, 53, 161, 198jJ., 409
kissing number, 174 equivalent, 5, 10, 163
An> n ~ 8, 13, 164, 168 l1{A}, 238
An' 9 ~ n ~ 12,145, 170jJ. even, see lattice, Type II
A l2 , 148 extremal, 49, 182, 189, 193, 204
An, 13 ~ n ~ 16, 170jf. extremal 24-dimensional, see Leech
An' 17 ~ n ~ 24, 176jf. lattice
An, 25 ~ n ~ 48, 177jf. extremal 32-dimensional, 156,
A4o, 147jf. 194
quaternionic, 160 extremal 40-dimensional, 156, 194,
thickness, 36 221jf.
see also Barnes-Wall lattice AI6 extremal 48-dimensional, see 48-
see also Leech lattice A24 dimensional lattices P 48p' P 48q
Lang, G. R., 74 extremal 56-dimensional, 156, 194
Laplace-Beltrami operator, 266 extremal 64-dimensional, 156, 194,
lattice, v, 3, 31, 42, 355, 479 213
A 2> see hexagonal lattice extremal Type II, 156, 194,213
A 3, see face-centered cubic lattice extreme, 44, 182,227
A;, see body-centered cubic lattice face-centered cubic, see face-
An, see An lattice centered cubic lattice
fcc, see face-centered cubic lattice
A:, see A: lattice
from BCH codes, 152jJ., 206, 235
automorphism group of, 90
from Reed-Muller codes, 151, 206,
Barnes-Wall, see Barnes-Wall
234
lattice{s}
Gaussian, 53, 161,202
bee, see body-centered cubic lattice
Gosset, see E 6 , E7 and E8 lattices
body-centered cubic, see body-
hcp, see hexagonal close-packing
centered cubic lattice
hexagonal, see hexagonal lattice
Bravais, 3
Hurwitz quaternionic, 53, 160
checkerboard, see Dn lattice
hyperbolic, see lattice, Lorentzian
classically integral, 47
integral, 47, 48, 53, 355, 522
complex, 52jf., 160
best, 179jJ.
congruent, 10
integral part of, 422
constructed from a group, 92
K l2 , see Coxeter-Todd lattice
Coxeter, A~ - An[s), rs - n + I, K n , 145, I 59jf., 165
116,223
laminated, see laminated lattice
Coxeter-Todd, see Coxeter-Todd
A 16, see Barnes-Wall lattice AI6
lattice
An, see laminated lattice
Craig, 17,206, 222jJ., 227, 241
Leech, see Leech lattice
cubic, see cubic lattice
Lorentzian, vi, x, 43, 48, 91, 132,
D 3 , see face-centered cubic lattice
412, 522jf., 527jf., 532jf.
D; , see body-centered cubic lattice group of, 527jJ., 532jf.
D4, see D4 lattice McKay, 194, 206, 221jf.
Dn, see Dn lattice neighboring, 415, 421, 427
D:, see D: lattice Niemeier, vi, x, 406jJ., 526, 570
definition, 3 glue code for, 406jJ.
densest, 12 group of, 407
difference, 222 non-principal, 53
directly congruent, 10 norm of, 55
654 Index

lattice (cont.) characterization of, 331fJ., 437fJ.

odd, see lattice, Type I complex, 200, 201, 511
of determinant I, see lattice, Type I definition, 200, 293
of determinant 2, 399ff. group of, 293
of determinant 3, 399ff. minimal vectors in, 200
one-dimensional, 8 covering radius, ix, 134, 476, 478.IJ.
over Zcn ~ - I, 215ff. cyclotomic constructions, 132, 227,
packing, 3 229ff.
polar, 10 decoding, 443, 447
quadratic form associated with, 42 deep holes, vi, ix, 134, 326, 413,
quaternionic, 52ff., 160, 201 478.IJ.
Quebbemann, 14, 24, 194, 206, definition, 131ff., 145, 193,211,286,
213ff. 422
reciprocal, 10 finding closest lattice point, 443
root, 6, 94, 97, 162, 163,408, 450ff. from Golay code, 131, 145, 193,286
Voronoi cell of, 456, 459 from ternary code, 149, 156
see also A., D., £6, £7, £8 generator matrix, 133
lattices group of, 91, 134, 287
S -,484 see also Conway group Coo
self-dual, see Type I groups associated with, 291
similar, 10 initial shells, 133
24- and 25-dimensional Type I, 421 A24 /2A 24 , 289, 332
Type I (self-dua\), 48, 182, I 85ff., MOG definition, 132
408,522 norm-doubling map, 177, 242
mass constants for, 408ff. norm-trebling map, 243
Type II (even self-dual), 48, 55, odd, 0 24 , 135, 179, 422, 425
181ff., 191ff., 331, 408, 522 optimality of kissing number, 12, 22,
mass constants for, 408ff. 337,338,345
unimodular, 48, 53, 331, 406ff., orbits of vectors, 289
421ff., 427ff., 439ff., 522 other constructions, 156, 506.IJ.
Voronoi's principal lattice, see A: quaternionic, 201, 243, 512
weight, 432 sections, 145.IJ., 166
~-coordinates for, 43 shallow holes, 134, 513ff.
see also packing shorter, 0 23 , 179, 420, 439
see also quadratic form summary of properties, 131
law of inertia, Sylvester, 371 theta series, 51, 134, 135, 192, 193,
layer, 136, 142 294,438
Leavitt, D. W., 89 Turyn-type construction, 132, 210,
Lee, T.-A., 74 213
Leech, J., v, ix, xi, xii, 10, 13, 19, 21, 23 constructions for, 506fJ., 571
24, 131, 136, 159,478 type of vector in, 289
Leech lattice A24 - A, v, 10, 13, 22, ubiquity, 11
39, 48, 90ff., 131ff., 145, 156, uniqueness, 331.IJ., 340.IJ.
I 92ff., 206, 221, 33 Iff., 338, Voronoi cell, 134,475
340ff., 406, 412, 420, 422, 427, Leech root, x, 524.fJ., 532ff., 569
439, 476, 478ff., 523, 527, Legendre, A.-M., 372
558ff. Lekkerkerker, C. G., 224
as Coxeter diagram, vi, x, 527ff. Lempken, W., 298
as icosian lattice, 132, 210 length, 75
as laminated lattice, 132, 157, 159, Lenstra, A. K., 41
177 Lenstra, H. W. Jr., 41
as Lorentzian lattice, 132 Leon, J. S., 186, 204, 297
Index 655

Lepowsky, J. I., 568 outer automorphism of, 323

Levenshtein, V. I., ix, 19, 23, 245, 247, presentation, 273
262, 264, 265 Rubik-icosahedron construction, 328
lexicography, 327 shuffle definition, 323
Lidl, R., 266 subgroups, 285
Lie algebra, xi, 94, 519, 568ff. Mathieu groups M 22 , M 23 , M 24, ix, 85,
ligancy, see kissing number 296, 299ff., 319, 323, 330, 335,
Lin, Shu, 76 434, 521, 556ff.
Lindsey, J. H. II, 3, 13, 298 definitions, 274
line, at infinity, 314 shuffle numbering, 324
projective, 267 subgroups, 279ff., 285, 307ff.
line-code, 317 matrix, Cartan, 514, 519
linear programming, 28, 245, 258, 338 generator, 4. 11,77,90, 101
Lint, J. H. van, 75, 76, 85 Gram,4, II, 42, 101
Litsyn, S. N., 18, 76, 224, 227, 447 Hadamard, 85, 87, 138, 139, 221
Livingstone, D., 295 monomial, 77, 433
Ljunggren, W., 524 Mattson, H. F. Jr., 76, 196
Lloyd, S. P., 266 Mazo, J. E., 11, 74
logi, 18, 244 McEliece, R. J., 246
long root, 97 McKay, J., viii, 221, 298
Longstaff, F. M., 74 MeL, 291, 292
loop, 327, 554ff., 563ff. McLaughlin, J., 292
Lorentzian space, vi, x, II, 12, 29, 43, mean squared error, 57, 449
91, 326, 522ff., 527ff. bounds on, 57ff.
Lovasz, L., 41, 265 measure, Haar, 252
Lyons, R., 298 invariant, 252
Lebesgue, 89, 252
Meurman, A. E., 568
m.s.e., see mean squared error Meyer, J., x, 527, 532, 541
MACSYMA, 420, 442, 475 Meyer, W., 298
MacWilliams, F. J., 76, 78, 183, 186, microweight, 432
245 Milnor, J., II, 522, 523
MacWilliams identity, see identity, minimal angle, 24
MacWilliams distance, 75
Madelung constant, 47 norm, 42, 53, 102, 403
Magliveras, S. S., 89 vector, 41, 94, 106, 428
Mallows, C. L., 189, 205, 351 finding, 41
map, gluing, 164 MINIMOG, ix, 299, 320, 322, 323
Martinet, J., 18, 227 Minkowski, H., 14, 16, 18, 42, 224,
mass constant, see mass formula 247, 352, 396, 397, 408
mass formula, 49, 78, 197,403,408 Minkowski reduction, 41, 352, 396
Massforme, 408 -1 as prime number, 366, 370
Mathieu, E., 267, 271, 274, 280, 330 Miracle Octad Generator (MOO), ix,
Mathieu groups Mil and M 12, ix, 29, 299,303,312,322
85, 267, 296, 297, 299ff., 320, standard labeling of, 309
323, 330, 435 standard sextets of, 312
definitions, 271, 280 Mitchell, H. H., 96, 127, 129
duality between hexads and duads, mnemonic, 328
326 modem, vi
duality of, 323 MOG, see Miracle Octad Generator
loop construction, 327 monad,279
numberings for, 328 Monster algebra, 561ff., 568ff.
656 Index

Monster group M - FI> vi, 11, 91, number, automorphous, 390

297, 299, 513, 531, 554jf., p -adically, 390jf.
568jf. Bernoulli, 409
centralizers in, 297 Cayley, see Cayley number
construction of, 554jf. contact, see kissing number
subgroups of, 555 coordination, see kissing number
Monstrous Moonshine, 11, 568 Coxeter, 99, 101, 407, 408, 425,
Moody, R. V., 569 428, 429, 510, 514, 519
Mordell, L. J., 10, 524 Euler, 409
Mordell inequality, 10, 12, 167 Eulerian, 467
Muder, D. J., 3 Hurwitz, see Hurwitz integral
Muller, D. E., 83 quaternion
multiplicative property, 108 kissing, see kissing number
Newton, see kissing number
p -adic, 366jf.
nab, 289 integral, 366, 367
n-ad,279 rational, 366, 367
n -simplex, 451 Ramanujan, 51, 105, 187, 193, 295,
n-sphere, 3, 9, 10,450,452 332
Namekawa, T., 27, 82 rational, Q, 366
nearest neighbor region, see Voronoi Nyquist, H., vi
cell
neighbor, see lattice, neighboring
Neumaier, A., 266, 522 O'Connor, R. E., 135, 179
Neutsch, W., 298 O'Meara, O. T., 378
Newton, I., 21, 224 O'Mura, J. K., 75, 76
Newton number, see kissing number O'Nan, M., 298
Niemeier, H.-V., ix, 48, 101, 336, 352, octad, 276, 279, 283, 305, 480
353,406,421,427,478 counts, 278
Niemeier lattice, see lattice, Niemeier standard, 312
Nikulin, V. V., 386 octahedron, 25, 36, 144, 449, 450, 452,
9-dimensional packings, 22 464
96-dimensionallattice, 243 truncated, 450, 454, 464, 471
noise, 63 octave, 122
non-principal, 53 octern, 318
nonconstructive, 14, 18, 68, 402jf. odd interpretation, 304
nonresidues N, 221, 268 odd-man-out, 321
Nordstrom, A. W., 87, 190, 195 oddity, 371, 372, 380, 384
norm, 41, 42, 102, 224, 355, 364 condition, 383
Euclidean, 208 formula, 370, 371
minimal, 42, 53, 102, 403 fusion, 381
of quaternion, 52 Odlyzko, A. M., 76, 228, 266, 337,
quaternionic, 208 351,439,446
trace, 224, 225 Ogg, A., 187
see also lattice, norm of w-rule, 301
norm-doubler, 239 0,268
normalized second IT. ment, 34 0.,24,248
Norton, S. P., ix, xi, xii, 41, 162, 179, one-dimensional lattice Z, 8,9, 106
298, 300, 476 theta series, 45
Norton's method, ix, 41,174,476 l20-cell (5,3,3), 451, 455
Novikova, N. V., 397 I-rule, 301
NP-hard,40 operator, Laplace, 428
Index 657

orbit, 93 physics, vii, II

orthogonal subspace, 163 il 9•1 lattice, 527fJ.
Ozeki, M., 156, 194 il 17,1 lattice, 527fJ.
il 2S ,1 lattice, 527fJ.
picture processing, 57
P 48p, P 48q, see 48-dimensional lattices Pierce, J. N., 78
p 48p and P 48q plane, affine, 314
p-adic metric, 367 Cayley, 251, 256
p-adic number, 366fJ. projective of order 4, 314, 549
p-excess, 371, 372 planetary perturbations, 29
p-part, 370 playing cards, 323, 328
p-signature, 370 Plesken, W., 161, 164, 179
p-term,377 Pless, V., 86, 189fJ.
p.d.k., see positive-definite degenerate Plotkin, M., 246, 262
kernel Pohst, M., 161, 179
packing, 1fJ., 31, 69 point-code, 317
best, 12, 15-17, 20, 235, 241, 242, Poisson, S. D., 103
250 Poisson summation formula, 103
Best, 140, 185 Poisson-Jacobi identity, 103
cannon ball, 2 pollen grains, 25
diamond, see diamond structure polygon, regular, 455
II-dimensional, 22, 139, 144 polynomial, Bernoulli, 409
from BCH codes, 152fJ., 206, 235 Gegenbauer,256
from Justesen codes, 155 generator, 79
from Reed-Muller codes, 151, 234 Hahn, 256
fruit stand, 2 isobaric, 188
hexagonal close-, see hexagonal Jacobi, 256, 264, 337, 341
close-packing Krawtchouk, 255fJ.
lattice, see lattice ultraspherical, 256
n-dimensional, 3, 8 polytope, 451
nonlattice, 7, 8, 139fJ., 195 Gosset, 463
dual to, 185 reciprocal, 474
periodic, 8 regular, 454
problem, 1fJ., 31, 69 second moment of, 453, 454
radius, 6, 10, 42, 46, 53, 75, 101 space-filling, 450
summary of results, 12fJ. volume of, 453
tetrahedral, see diamond structure see also individual polytopes
12-dimensional, 139, 142 positive-definite degenerate kernel,
see also lattice 256fJ.
palindromic properties, 165, 166 power, average, 66
Pall, G., 135, 179, 366, 378 constraint, 69
parallelohedron, 450 Preparata, F. P., 41, 87, 191
parentheses, 406 presentation, 96, 456
Parker, R. A., ix, xi, xii, 298, 300, 478, primal problem, 258
521, 554fJ. prime, 366
Parker loop, 554fJ., 563fJ. archimedean, 370
PCM, see pulse code modulation infinite, 370
pennies, 22 intractable, 404
penumbral, 284 irrelevant, 392
permutation, turnaround, 324 -I as, 366, 370
permutohedron, 115, 472 tractable, 392
Peterson, W. W., 76 unit, 370
658 Index

prism, hexagonal, 450 Minkowski, 41, 352, 353, 396

problem, kissing number, see kissing Reed, I. S., 81, 83
number problem reflecting hyperplane, 97, 528
problem of thirteen spheres, 21 reflection, 95, 376
product formula, 353, 370, 371 subgroup, 97
product, inner, 4 region, fundamental, 358, 359
scalar, 4 relations, 96
semidirect, 432 relevant vector, 35, 70, 474
wreath,432 replacement lemma, 377
projective plane of order 4, 314, 549 representation, class I, 253
Promhouse, G., 76, 152 integral, 91, 92
proper, 355 irreducible, 252
PSL 2 (Z), 358 left regular, 252
PSLn (q), 267jJ. modular, 92, 298
pulse code modulation, 56, 65 of Monster and subgroups etc., 554,
559, 566
of sporadic groups, 298
q-ary-symmetric channel, 75 orthogonal, 93
Qp,366 permutation, 92
quadratic form, see form, quadratic unitary, 252
quadratic number field, 364 rescaling, 105
quadratic residues Q, 221, 268 rhombic dodecahedron, 34, 450, 462
quadric, 292 Riemannian manifold, 247
quantization, 31, 56jJ., 63, 443, 449 ring, 206, 356
quantizer problem, v, 59, 69 graded, 187, 193
lattice, 59 icosian, 207
summary of results, 59 Rivest, R. L., 446
quasicrystal, vii, 21 Robinson, J. P., 87, 190, 195
quasilattice, 21 Rodemich, E. R., 246
quaternion, 52, 119, 207 Rogers, C. A., 3, II, 13, 40
conjugate, 52 Romans, 315
norm of, 52 roof,458
unit, 54 root, 97
quaternionic integer, 52, 119 fundamental, 97, 528, 531, 532, 547
Quebbemann, H.-G., 14,24,225,244 Leech, x, 524jJ., 532jJ., 569
Queen, L., xii, 513, 568 p-adic, 391
Qureshi, S. U., 74 system, 94, 97, 427, 457, 482, 569
vector, 97
Ross, A. E., 403
radius, covering, see covering radius rounding off, 13, 56jJ.
packing, see packing radius Rubicon, 328
subcovering, 163 Rudakov, A. N., 386
Ramanujan, S., 193, 295 Rudvalis, A., 298
Ramanujan number, see number, Rumsey, H. C. Jr., 246
Ramanujan Rush, J. A., 19
Ramanujan's function, see number, Ryba, A. J. E., 298
Ramanujan Ryskov, S. S., 36, 39, 397
Rankin, R. A., 27, 187,247,264
rate, 67, 76, 246
Ray-Chaudhuri, D. K., 81 S n group, see group, symmetric
real Cayley algebra, 122 scale, 378, 380
reduction, 41 Schatz, J. R., 76
Index 659

Schliifli, L., 28, 119 signs, attaching, 322

Schoenberg, I. J., 257 Sigrist, F., 11
Schoeneberg, B., 187 similarity, 236
school district, 33 simplex, 452
Schutte, K., 21 centroid of, 452
Schur double cover, 293 property of shallow hole in A 24 , 514
Schwarzenberger, R. L. E., 11 regular, 453
score, 304 second moment of, 453
search problem, 12 {3,3,3},451
secant, 495 volume of, 453
second moment, 59,449 Sims, C. C., 292, 297
about an axis, 451 Singleton, R. C., 81
dimensionless, G (p), 450ff. 6-design, 89
in terms of faces, 453 600-cell {3,].,5}, 449, 451, 456
normalized, I(P), 450ff. 16-cell {3,3,4}, 451
unnormalized, U (P), 450ff. skew field, 119
section, 136, 145, 163 Skubenko, B. F., 160
Seidel, J. J., 90, 266, 522 Sleeper, H. A., 74
self-glue, 100 Slepian, D., 66, 68
semi-direct product, 92 slope, 301, 320
sequence, Cauchy, 367 Smith, F. L., 298
series, Eisenstein, 51, 294 Smith, H. J. S., 120
Poincare, 187, 192 Smith, N. K., 74
theta, see theta series Smith, P. E., 297
Serre, J.-P., 187,436,522 Smith, Stephen D., 555, 562, 568
7-design, spherical, 342 snub cube, 25
sextet, 279, 309 Sobolev, S. L., 90, 93
standard, 309 soft decoding, 443
Seymour, P. D., 90 Soicher, L. H., 298
Shafarevich, I. R., 18,206,227,386 Solomon, G., 81
shallow hole, 33, 570, 571 solution, feasible, 342
Shannon, C. E., vi, 66 sorting, 446
shape rule, 302 source decoder, 64
Shapiro, H. M., 74 source encoder, 63
Shekhovtsov, O. I., 447 space, Euclidean, Rn , 3, 249ff.
shell, 26, 95, 102 Hamming, Fi, 249ff.
Shephard, G. C., 129 hyperbolic, see hyperbolic space
short, 559 isotropic, 251
root, 97 Johnson, FV, 249ff.
vector, 332, 480 Lorentzian, see Lorentzian space
shuffle numbering, 328 n-dimensional, 3
shuffle, Mongean, 323 of cosets, 250
shuffling, 323 orthogonal, 163
Sidel'nikov, V. M., 247 projective, 251, 256, 264
Siegel, C. L., 187, 189,408 quaternionic, 251, 256, 264
Siersma, D., 11 rank one, 251
sign, 379, 380 spherical, n n, 249ff.
rule, 302, 563 2-point-homogeneous, 245, 250ff.
walking, 381 sparsity, 31
signal-to-noise ratio, 70 special, 276, 284
normalized, 71 node, 98, 510
signature, 371 speech,57
660 Index

sphere packing problem, v, 1,76 Sylvester, J. J., 269, 495

summary of results, 12 Sylvester law of inertia, 376, 379, 523
sphere, n-dimensional, 3, 450, 452 symbol, Jacobi-Legendre, 368
surface area of, 10 (-I)-adic, 379
volume of, 9 p-adic, 379
spheres, overlapping, 31 2-adic, 380
touching two spheres, 249, 250, 264 symbols, list of, xiv
spherical cap, 25 synthematic total, 269, 495
codes, construction of, 26jJ syntheme, 269, 495
coefficient, 428 system, root, see root system
harmonic, 256
simplex, 28, 458
area of, 28 t -design, vi, 88, 89
spinor genus, 352, 353, 366, 378, spherical, 89, 340jJ.
388jJ., 393, 397 tight, 90
computing, 388jJ. table, binary indefinite forms, 362-363
invariants for, 388jJ. binary positive definite forms, 360
spinor kernel, 388, 389, 390, 404 codes, best, I 52, 247, 248
spinor norm, 388, 390 coding gain of lattices, 74
spinor operator, 389 coverings, 12, 38
split extension, 92 crystallographic reflection groups, 98
spread-spectrum communications, II deep holes in A24 , 479, 515
square anti prism, 25 dodecad counts, 278
square class, p -adic, 367, 368 even unimodular 24-dimensional
square lattice Z2, 31, 32, 106, 184, 240 lattices, 407
as Gaussian lattice, 55 exceptional simple groups, 296jJ.
theta series, 106 genera of forms with Idetl ~ II,
squared length, 41 387
standard partition, 323 groups associated with Leech lattice,
Stanley, R. P., 186 291
Stanton, D., 266 icosian ring, 209
statistical analysis on spheres, 11 initial shells of £ 8, 123, 209
Steiner system S (t ,k, v), vi, 88 initial shells of A 24 , 133
definition, 88 integral lattices, best, 180
SO,4, 10),88,89, 140 kissing numbers, 12, 15, 16,23,174,
SO,4, 14),88,89,145 250
sO, 6, 22), 320 laminated lattices, 158, 174
S(4, 5, 11), 89 lattices from £8,241
S(4, 7, 23), 89, 319 lattices from A 24 , 242
S (5, 6, 12), 89, 139, 299, 320jJ, lattices from BCH codes, 235
326 lattices K n , 161, 162
S (5, 8, 24), 89, 145, 276, 299, 305, lattices of determinant 2, 400
335 lattices of determinant 3, 402
Stinaff, R. D., 76 Leech roots, 533jJ, 542jJ., 549, 551
strong generating set, 177 mass constants, 410jJ.
subcovering radius, 36 maximal subgroups of Mathieu
Sugiyama, Y., 27, 83 groups, 285, 308
sum formula, 371 minimal vectors in complex Leech
sum of four squares, 51 lattice, 200
sums of squares, 44jJ., 51, 108 MOG coordinates for K n , 161
superalgebra, 568 MOG coordinates for An, 160
superstring theory, vii, 12 Niemeier lattices, 407
Index 661

table (cont.) Thackray, J. G., 300

numbers represented by binary theorem, Assmus-Mattson, 196
forms, 374 Bieberbach, 99
octad counts, 278 Biirmann-Lagrange, 442
odd unimodular 24-dimensional complementary slackness, 342
lattices, 424 deep holes, x, 478
P(O)/M 24 ,284 Deligne,51
packings, 12, 15-17, 20, 235, 241, Delsarte, 245
242,250 Dirichlet, 372
permutations of S 12 mod M 12, 328 duality, 259
quantizers, best, 12, 61 Eichler, 388
record packings, etc., 12 Gleason, vi, 186, 192, 202
reflection groups, 98 Gleason-Pierce, 78
root lattices, 98 Hasse principle, 372
root systems, 98 Hecke, vi, 192,331
second moments, 450, 451 Jacobi, 103, 186, 440
sections of P 48p , 167 MacWilliams, 78, 103, 186, 245,
shallow holes in A24, 515jJ. 258
sporadic simple groups, 296 Minkowski, see bound, Minkowski
subgroups of Monster, 555 Minkowski-H1awka, 15
ternary indefinite forms, 399 Niemeier, ix, 48, 406, 410jJ., 427jJ.
ternary positive definite forms, 398 Peter-Weyl,252
theta series of bec lattice, 116 Poisson-Jacobi, 47, 103, 186,
theta series of D4 , 119 440
theta series of E 6, E 1, E 8, 123 sampling, 63jJ.
theta series of fcc lattice, 113 shallow holes, x, 513
theta series of hcp, 114 Shannon, 66jJ., 74
theta series of hexagonal lattice, 111 Sobolev,93
theta series of K 12, 129 Sylvester law of inertia, 376, 379,
theta series of A16, 130 523
theta series of A24 , 135 Venkov, 196
theta series of P 48p' P 48q' 196 Witt cancellation, 375, 376
theta series of Z2, 106 Witt, 94, 101
theta series of Z3, 107 Zador,58
2-elementary forms, 388 theta function, Jacobi, see Jacobi theta
unimodular lattices, 49, 416jJ. function
Tammela, P., 397 see also theta series
Tammes' problem, 25 theta series, vi, 45jJ., 53, 73, 95, 102,
Tarjan, R. E., 446 181,220,428
Tavares, S. E., 76, 152 average, 45
Teirlinck, L., 89 congruences for, 294
telephone, 56 definition, 45
tern, 318 does not determine lattice, 47
tet, 321 extremal, 189, 193,331,440
tetracode C 4, ix, 81, 200, 202, 320, for nonlattice packing, 45
408,436 of dual lattice, 47, 54, 103
definition, 81, 320 see also Jacobi theta function
tetrad, 321 see also under individual lattices
tetrahedral packing, see diamond thickness, 31, 32, 101,449
structure normalized, 32
tetrahedron, 25, 450, 464 36-dimensional lattice B 36, 234
truncated, 464 32-dimensionallattice Q32, 220
662 Index

Thompson, J. G., 149, 224, 229, 244, indicator, 88

296,297,351,422 isotropic, 523
{3,4,31, 119,216,451,455 minimal, 41, 94, 106,428
3-problem, 302 relevant, 474
Thue, A., 8 short, 480
Tietiiviiinen, A., 85, 262, 264 Weyl, vi, x, 472,510,514,519,528,
tip, 98 569
Tits, J., 135,298,307,315, 512, 554 Veldkamp, F. D., 11
Todd, J. A., 13, 127, 129, 159,279 Venkov, B. B., xi, xii, 90, 93, 196, 205,
Tonchev, V. D., 194 408, 427, 474
total, 269, 495 Verhoeff, T., 76
touching two spheres, 249, 250, 264 vertex figure, 474
tower, class field, 18, 206, 227 Vetchinkin, N. M., 12, 162
trace, 224 Vinberg, E. B., x, 527, 528, 532, 541,
tractable, 393 547, 569
train, 381 Viterbi, A. J., 75, 76
transform, by zonal spherical function, Vladuts, S. G., 266
258 volume formula, 514
fast Hadarmard, 447 volume of unit sphere, Vn , 9
Krawtchouk, 245 Voronoi, G. F., 33, 474
transposing, 313 Voronoi cell, v, 6, 33, 56jJ., 69, 95, 449
transverse, 283, 284 algorithms for, 41
trellis coded modulation, vi, 74 second moment of, 59jJ.
Trenerry, D. W., 36,62 Voronoi-relevant, 35, 70, 474
triad,279
triality, 557
trio, 279, 316 Waerden, B. L. van der, 21, 25
triple map, 557 Wales, D. B., 93, 298
trivial invariants, 376 Wall, G. E., 13, 129, 159,232
truncated octahedron, 34 Wang, H.-C., 251
truncating, 464 Ward, H. N., 190,440
Tsfasman, M. A., 18,224,227,266 Watson, G. L., 12, 22, 39, 141, 355,
Turyn, R., 307, 317 388, 389, 390, 396, 462
Tutte, W. T., 549 Watson, G. N., 524
twelve spheres, 21jJ., 29jJ. Wei, L.·F., 74
24-cell {3,4,31, 119,216,451,455 Wei, V. K., 74
2-adic symbol, canonical, 382 weight, 75, 186, 187
2-problem, 321 binary, 83
type, 289, 380, 384, 480 distribution, 78
scaled,380 enumerator, vi, 78
complete, 78
extremal, 189, 193, 194
umbral, 279, 283, 284 Hamming, 75, 433
Ungerboeck, G., 74 Weiss, R., 298
union bound, 70 Welch, L. R., 246
unit, 54 Weldon, E. J. Jr., 27, 76, 83
Urakawa, H., 266 Wells, A. F., vii, II
Welti, G. R., 73
Weyl group, 98, 408, 432, 456, 458
vector, characteristic, 88 affine, 456
fundamental, 506jJ., 523 see also under An' Dn , £6. £7, £8
glue, 99, 101,406, 506, 524, 570 lattices
Index 663

Weyl vector, vi, x, 472, 510, 514, 519, Yoshiara, S., 201, 298
528, 569 Young tableau, 255
Weyl-MacDonald-Kac formula, 571
Whitrow, G. J., 527
Wigner-Seitz cell, see Voronoi cell Z, see one-dimensional lattice
Wilson, J., 256 Z2, see square lattice
Wilson, R. A., 201, 298, 327 Z', see cubic lattice
Wilson, S. G., 74 ZP' 366
Witt, E., 47, 307, 315, 335, 352, 375, Zador, P. L., 58
399 Zaslavsky, T., 90
Woldar, A., 298 zero-sum check, 79
Worley, R. T., 61, 463 r<n), 14
wreath product, 432 Zink, T., 266
Wyner, A. D., 27 Zinoviev, V. A., 76
Zolotareff, G., 10, 12, 120

X-ray tomography, 11
Grundlehren der mathematischen Wissenschaften
Continued from page ii

248. Suzuki: Group Theory II

249. Chung: Lectures from Markov Processes to Brownian Motion
250. Arnold: Geometrical Methods in the Theory of Ordinary Differential Equations
251. Chow/Hale: Methods of Bifurcation Theory
252. Aubin: Nonlinear Analysis on Manifolds, Monge-Ampere Equations
253. Dwork: Lectures on p-adic Differential Equations
254. Freitag: Siegelsche Modulfunktionen
255. Lang: Complex Multiplication
256. Hormander: The Analysis of Linear Partial Differential Operators I
257. Hormander: The Analysis of Linear Partial Differential Operators II
258. Smoller: Shock Waves and Reaction-Diffusion Equations
259. Duren: Univalent Functions
260. FreidlinlWentzall: Random Perturbations of Dynamical Systems
261. RemmertiBosch/Giintzer: Non Archimedian Analysis-A Systematic Approach to
Rigid Analytic Geometry
262. Doob: Classical Potential Theory & Its Probabilistic Counterpart
263. Krasnosel'skii/Zabreiko: Geometrical Methods of Nonlinear Analysis
264. AubiniCellina: Differential Inclusions
265. Grauert/Remmert: Coherent Analytic Sheaves
266. de Rham: Differentiable Manifolds
267. Arbarello/CornalbaiGriffiths/Harris: Geometry of Algebraic Curves, Vol. I
268. Arbarello/CornalbaiGriffiths/Harris: Geometry of Algebraic Curves, Vol. II
269. Schapira: Microdifferential Systems in the Complex Domain
270. Scharlau: Quadratic and Hermitian Forms
271. Ellis: Entropy, Large Deviations, and Statistical Mechanics
272. Elliott: Arithmetic Functions and Integer Products
273. Nikol'skii: Treatise on Shift Operator
274. Hormander: The Analysis of Linear Partial Differential Operators III
275. Hormander: The Analysis of Linear Partial Differential Operators IV
276. Liggett: Interacting Particle Systems
277. Fulton/Lang: Riemann-Roch Algebra
278. Barr/Wells: Toposes, Triples, and Theories
279. Bishop/Bridges: Constructive Analysis
280. Neukirch: Class Field Theory
281. Chandrasekharan: Elliptic Functions
282. Lelong/Gruman: Entire Functions of Several Complex Variables
283. Kodaira: Complex Manifolds and Deformation of Complex Structures
284. Finn: Equilibrium Capillary Surfaces
285. Burago/Zalgaller: Geometric Inequalities
286. Andrianov: Quadratic Forms and Hecke Operators
287. Maskit: Kleinian Groups
288. Jacod/Shiryaev: Limit Theorems for Stochastic Processes
289. Manin: Gauge Field Theory and Complex Geometry
290. Conway/Sloane: Sphere Packings, Lattices and Groups
291. HahnlO'Meara: The Classical Groups and K-Theory