The Random Matrix Theory Of The Classical Compact Groups Cambridge Tracts In Mathematics 1st Edition Elizabeth S Meckes

jerielgeedan 6 views 89 slides May 23, 2025
Slide 1
Slide 1 of 89
Slide 1
1
Slide 2
2
Slide 3
3
Slide 4
4
Slide 5
5
Slide 6
6
Slide 7
7
Slide 8
8
Slide 9
9
Slide 10
10
Slide 11
11
Slide 12
12
Slide 13
13
Slide 14
14
Slide 15
15
Slide 16
16
Slide 17
17
Slide 18
18
Slide 19
19
Slide 20
20
Slide 21
21
Slide 22
22
Slide 23
23
Slide 24
24
Slide 25
25
Slide 26
26
Slide 27
27
Slide 28
28
Slide 29
29
Slide 30
30
Slide 31
31
Slide 32
32
Slide 33
33
Slide 34
34
Slide 35
35
Slide 36
36
Slide 37
37
Slide 38
38
Slide 39
39
Slide 40
40
Slide 41
41
Slide 42
42
Slide 43
43
Slide 44
44
Slide 45
45
Slide 46
46
Slide 47
47
Slide 48
48
Slide 49
49
Slide 50
50
Slide 51
51
Slide 52
52
Slide 53
53
Slide 54
54
Slide 55
55
Slide 56
56
Slide 57
57
Slide 58
58
Slide 59
59
Slide 60
60
Slide 61
61
Slide 62
62
Slide 63
63
Slide 64
64
Slide 65
65
Slide 66
66
Slide 67
67
Slide 68
68
Slide 69
69
Slide 70
70
Slide 71
71
Slide 72
72
Slide 73
73
Slide 74
74
Slide 75
75
Slide 76
76
Slide 77
77
Slide 78
78
Slide 79
79
Slide 80
80
Slide 81
81
Slide 82
82
Slide 83
83
Slide 84
84
Slide 85
85
Slide 86
86
Slide 87
87
Slide 88
88
Slide 89
89

About This Presentation

The Random Matrix Theory Of The Classical Compact Groups Cambridge Tracts In Mathematics 1st Edition Elizabeth S Meckes
The Random Matrix Theory Of The Classical Compact Groups Cambridge Tracts In Mathematics 1st Edition Elizabeth S Meckes
The Random Matrix Theory Of The Classical Compact Groups Cam...


Slide Content

The Random Matrix Theory Of The Classical
Compact Groups Cambridge Tracts In Mathematics
1st Edition Elizabeth S Meckes download
https://ebookbell.com/product/the-random-matrix-theory-of-the-
classical-compact-groups-cambridge-tracts-in-mathematics-1st-
edition-elizabeth-s-meckes-11300550
Explore and download more ebooks at ebookbell.com

Here are some recommended products that we believe you will be
interested in. You can click the link to download.
The Random Matrix Theory Of The Classical Compact Groups Meckes
https://ebookbell.com/product/the-random-matrix-theory-of-the-
classical-compact-groups-meckes-10513816
The Oxford Handbook Of Random Matrix Theory Online Version Gernot
Akemann Jinho Baik Philippe Di Francesco
https://ebookbell.com/product/the-oxford-handbook-of-random-matrix-
theory-online-version-gernot-akemann-jinho-baik-philippe-di-
francesco-10441710
The Random Factor How Chance And Luck Profoundly Shape Our Lives And
The World Around Us Mark Robert Rank
https://ebookbell.com/product/the-random-factor-how-chance-and-luck-
profoundly-shape-our-lives-and-the-world-around-us-mark-robert-
rank-56900470
The Randomcluster Model 1st Edition Geoffrey R Grimmett Auth
https://ebookbell.com/product/the-randomcluster-model-1st-edition-
geoffrey-r-grimmett-auth-4207814

The Randomcluster Model 1st Ed 2006 Corr 2nd Printing Geoffrey R
Grimmett
https://ebookbell.com/product/the-randomcluster-model-1st-
ed-2006-corr-2nd-printing-geoffrey-r-grimmett-1290014
The Random Reader New Zealand Short Stories Volume One Various Authors
https://ebookbell.com/product/the-random-reader-new-zealand-short-
stories-volume-one-various-authors-48856744
Beyond The Random Walk A Guide To Stock Market Anomalies And Lowrisk
Investing Vijay Singal
https://ebookbell.com/product/beyond-the-random-walk-a-guide-to-stock-
market-anomalies-and-lowrisk-investing-vijay-singal-2476816
The Political Potential Of Sortition A Study Of The Random Selection
Of Citizens For Public Office Oliver Dowlen
https://ebookbell.com/product/the-political-potential-of-sortition-a-
study-of-the-random-selection-of-citizens-for-public-office-oliver-
dowlen-5552150
My Meteorite Or Without The Random There Can Be No New Thing Harry
Dodge
https://ebookbell.com/product/my-meteorite-or-without-the-random-
there-can-be-no-new-thing-harry-dodge-11976184

THE RANDOM MATRIX THEORY OF THE CLASSICAL
COMPACT GROUPS
This is the first book to provide a comprehensive overview of foundational
results and recent progress in the study of random matrices from the classi-
cal compact groups, drawing on the subject’s deep connections to geometry,
analysis, algebra, physics, and statistics. The book sets a foundation with an
introduction to the groups themselves and six different constructions of Haar
measure. Classical and recent results are then presented in a digested, accessible
form, including the following: results on the joint distributions of the entries;
an extensive treatment of eigenvalue distributions, including the Weyl integra-
tion formula, moment formulae, and limit theorems and large deviations for
the spectral measures; concentration of measure with applications both within
random matrix theory and in high dimensional geometry, such as the Johnson–
Lindenstrauss lemma and Dvoretzky’s theorem; and results on characteristic
polynomials with connections to the Riemann zeta function. This book will be
a useful reference for researchers in the area, and an accessible introduction for
students and researchers in related fields.
elizabeth s. meckes is Professor of Mathematics at Case Western
Reserve University. She is a mathematical probabilist specializing in random
matrix theory and its applications to other areas of mathematics, physics, and
statistics. She received her PhD from Stanford University in 2006 and received
the American Institute of Mathematics five-year fellowship. She has also been
supported by the Clay Institute of Mathematics, the Simons Foundation, and
the US National Science Foundation. She is the author of 24 research papers
in mathematics, as well as the textbookLinear Algebra, coauthored with Mark
Meckes and published by Cambridge University Press in 2018.

CAMBRIDGE TRACTS IN MATHEMATICS
GENERAL EDITORS
B. BOLLOBÁS, W. FULTON, F. KIRWAN, P. SARNAK,
B. SIMON, B. TOTARO
A complete list of books in the series can be found atwww.cambridge.org/mathematics.
Recent titles include the following:
183. Period Domains over Finite andp-adic Fields. By J.-F. Dat, S. Orlik, and M. Rapoport
184. Algebraic Theories. By J. Adámek, J. Rosický, and E. M. Vitale
185. Rigidity in Higher Rank Abelian Group Actions I: Introduction and Cocycle Problem.
By A. Katok and V. Ni ¸ticˇa
186. Dimensions, Embeddings, and Attractors. By J. C. Robinson
187. Convexity: An Analytic Viewpoint. By B. Simon
188. Modern Approaches to the Invariant Subspace Problem. By I. Chalendar and
J. R. Partington
189. Nonlinear Perron–Frobenius Theory. By B. Lemmens and R. Nussbaum
190. Jordan Structures in Geometry and Analysis. By C.-H. Chu
191. Malliavin Calculus for Lévy Processes and Infinite-Dimensional Brownian Motion. By
H. Osswald
192. Normal Approximations with Malliavin Calculus. By I. Nourdin and G. Peccati
193. Distribution Modulo One and Diophantine Approximation. By Y. Bugeaud
194. Mathematics of Two-Dimensional Turbulence. By S. Kuksin and A. Shirikyan
195. A Universal Construction for Groups Acting Freely on Real Trees. By I. Chiswell and
T. Müller
196. The Theory of Hardy’s Z-Function. By A. Ivi´c
197. Induced Representations of Locally Compact Groups. By E. Kaniuth and K. F. Taylor
198. Topics in Critical Point Theory. By K. Perera and M. Schechter
199. Combinatorics of Minuscule Representations. By R. M. Green
200. Singularities of the Minimal Model Program. By J. Kollár
201. Coherence in Three-Dimensional Category Theory. By N. Gurski
202. Canonical Ramsey Theory on Polish Spaces. By V. Kanovei, M. Sabok, and J. Zapletal
203. A Primer on the Dirichlet Space. By O. El-Fallah, K. Kellay, J. Mashreghi, and
T. Ra n s f o r d
204. Group Cohomology and Algebraic Cycles. By B. Totaro
205. Ridge Functions. By A. Pinkus
206. Probability on Real Lie Algebras. By U. Franz and N. Privault
207. Auxiliary Polynomials in Number Theory. By D. Masser
208. Representations of Elementary Abelianp-Groups and Vector Bundles. By D. J. Benson
209. Non-homogeneous Random Walks. By M. Menshikov, S. Popov and A. Wade
210. Fourier Integrals in Classical Analysis (Second Edition). By C. D. Sogge
211. Eigenvalues, Multiplicities and Graphs. By C. R. Johnson and C. M. Saiago
212. Applications of Diophantine Approximation to Integral Points and Transcendence. By
P. Corvaja and U. Zannier
213. Variations on a Theme of Borel. By S. Weinberger
214. The Mathieu Groups. By A. A. Ivanov
215. Slenderness I: Foundations. By R. Dimitric
216. Justification Logic. By S. Artemov and M. Fitting
217. Defocusing Nonlinear Schrödinger Equations. By B. Dodson
218. The Random Matrix Theory of the Classical Compact Groups. By E. S. Meckes

The Random Matrix Theory of the
Classical Compact Groups
Cambridge Tracts in Mathematics 218
ELIZABETH S. MECKES
Case Western Reserve University

University Printing House, Cambridge CB2 8BS, United Kingdom
One Liberty Plaza, 20th Floor, New York, NY 10006, USA
477 Williamstown Road, Port Melbourne, VIC 3207, Australia
314–321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre,
New Delhi – 110025, India
79 Anson Road, #06–04/06, Singapore 079906
Cambridge University Press is part of the University of Cambridge.
It furthers the University’s mission by disseminating knowledge in the pursuit of
education, learning, and research at the highest international levels of excellence.
www.cambridge.org
Information on this title:www.cambridge.org/9781108419529
DOI:10.1017/9781108303453
©Elizabeth S. Meckes 2019
This publication is in copyright. Subject to statutory exception
and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2019
Printed in the United Kingdom by TJ International Ltd., Padstow, Cornwall
A catalogue record for this publication is available from the British Library.
Library of Congress Cataloging-in-Publication Data
Names: Meckes, Elizabeth S., author.
Title: The random matrix theory of the classical compact groups /
Elizabeth Meckes (Case Western Reserve University).
Description: Cambridge ; New York, NY : Cambridge University Press, 2019.|
Series: Cambridge tracts in mathematics ; 218|
Includes bibliographical references and index.
Identifiers: LCCN 2019006045|ISBN 9781108419529 (hardback : alk. paper)
Subjects: LCSH: Random matrices.|Matrices.
Classification: LCC QA196.5 .M43 2019|DDC 512.9/434–dc23
LC record available athttps://lccn.loc.gov/2019006045
ISBN 978-1-108-41952-9 Hardback
Cambridge University Press has no responsibility for the persistence or accuracy of
URLs for external or third-party internet websites referred to in this publication
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.

Image by Swallowtail Garden Seeds

To Mark

Contents
Preface pageix
1 Haar Measure on the Classical Compact Matrix Groups 1
1.1 The Classical Compact Matrix Groups 1
1.2 Haar Measure 7
1.3 Lie Group Structure and Character Theory 17
2 Distribution of the Entries 31
2.1 Introduction 31
2.2 The Density of a Principal Submatrix 38
2.3 How Much Is a Haar Matrix Like a Gaussian Matrix?42
2.4 Arbitrary Projections 53
3 Eigenvalue Distributions: Exact Formulas 60
3.1 The Weyl Integration Formula 60
3.2 Determinantal Point Processes 71
3.3 Matrix Moments 80
3.4 Patterns in Eigenvalues: Powers of Random Matrices85
4 Eigenvalue Distributions: Asymptotics 90
4.1 The Eigenvalue Counting Function 90
4.2 The Empirical Spectral Measure and Linear
Eigenvalue Statistics 106
4.3 Patterns in Eigenvalues: Self-Similarity 113
4.4 Large Deviations for the Empirical Spectral Measure117
5 Concentration of Measure 131
5.1 The Concentration of Measure Phenomenon 131
5.2 Logarithmic Sobolev Inequalities and Concentration133
vii

viii Contents
5.3 The Bakry–Émery Criterion and Concentration for the
Classical Compact Groups 141
5.4 Concentration of the Spectral Measure 153
6 Geometric Applications of Measure Concentration 161
6.1 The Johnson–Lindenstrauss Lemma 161
6.2 Dvoretzky’s Theorem 165
6.3 A Measure-Theoretic Dvoretzky Theorem 170
7 Characteristic Polynomials and Connections to the Riemann
ζ-function 181
7.1 Two-Point Correlations and Montgomery’s Conjecture181
7.2 The Zeta Function and Characteristic Polynomials
of Random Unitary Matrices 186
7.3 Numerical and Statistical Work 197
References 205
Index 212

Preface
This book grew out of lecture notes from a mini-course I gave at the 2014
Women and Mathematics program at the Institute for Advanced Study. When
asked to provide background reading for the participants, I found myself at a
bit of a loss; while there are many excellent books that give some treatment
of Haar distributed random matrices, there was no single source that gave a
broad, and broadly accessible, introduction to the subject in its own right. My
goal has been to fill this gap: to give an introduction to the theory of random
orthogonal, unitary, and symplectic matrices that approaches the subject from
many angles, includes the most important results that anyone looking to learn
about the subject should know, and tells a coherent story that allows the beauty
of this many-faceted subject to shine through.
The book begins with a very brief introduction to the orthogonal, unitary,
and symplectic groups – just enough to get started talking about Haar measure.
Section 1.2includes six different constructions of Haar measure on the clas-
sical groups;Chapter 1also contains some further information on the groups,
including some basic aspects of their structure as Lie groups, identification of
the Lie algebras, an introduction to representation theory, and discussion of the
characters.
Chapter 2is about the joint distribution of the entries of a Haar-distributed
random matrix. The fact that individual entries are approximately Gaussian is
classical and goes back to the late nineteenth century. This chapter includes
modern results on the joint distribution of the entries in various senses: total
variation approximation of principal submatrices by Gaussian matrices, in-
probability approximation of (much larger) submatrices by Gaussian matrices,
and a treatment of arbitrary projections of Haar measure via Stein’s method.
Chapters 3and4deal with the eigenvalues.Chapter 3is all about exact
formulas: the Weyl integration formulas, the structure of the eigenvalue pro-
cesses as determinantal point processes with explicit kernels, exact formulas
ix

x Preface
due to Diaconis and Shahshahani for the matrix moments, and an interesting
decomposition (due to Eric Rains) of the distribution of eigenvalues of powers
of random matrices.
Chapter 4deals with asymptotics for the eigenvalues of large matrices: the
sine kernel microscopic scaling limit, limit theorems for the empirical spectral
measures and linear eigenvalue statistics, large deviations for the empirical
spectral measures, and an interesting self-similarity property of the eigenvalue
distribution.
Chapters 5and6are where this project began: concentration of measure
on the classical compact groups, with applications in geometry.Chapter 5
introduces the concept of concentration of measure, the connection with log-
Sobolev inequalities, and derivations of optimal (at least up to constants)
log-Sobolev constants. The final section contains concentration inequalities
for the empirical spectral measures of random unitary matrices.
Chapter 6has some particularly impressive applications of measure concen-
tration on the classical groups to high-dimensional geometry. First, a proof of
the celebrated Johnson–Lindenstrauss lemma via concentration of measure on
the orthogonal group, with a (very brief) discussion of the role of the lemma in
randomized algorithms. The second section is devoted to a proof of Dvoretzky’s
theorem, via concentration of measure on the unitary group. The final section
gives the proof of a “measure-theoretic” Dvoretzky theorem, showing that, sub-
ject to some mild constraints, most marginals of high-dimensional probability
measures are close to Gaussian.
Finally,chapter 7gives a taste of the intriguing connection between eigenval-
ues of random unitary matrices and zeros of the Riemann zeta function. There
is a section on Montgomery’s theorem and conjecture on pair correlations and
one on the results of Keating and Snaith on the characteristic polynomial of a
random unitary matrix, which led them to exciting new conjectures on the zeta
side. Some numerical evidence (and striking pictures) is presented.
Haar-distributed random matrices appear and play important roles in a wide
spectrum of subfields of mathematics, physics, and statistics, and it would never
have been possible to mention them all. I have used the end-of-chapter notes in
part to give pointers to some interesting topics and connections that I have not
included, and doubtless there are many more that I did not mention at all. I have
tried to make the book accessible to a reader with an undergraduate background
in mathematics generally, with a bit more in probability (e.g., comfort with
measure theory would be good). But because the random matrix theory of the
classical compact groups touches on so many diverse areas of mathematics,
it has been my assumption in writing this book that most readers will not be
familiar with all of the background that comes up. I have done my best to give
accessible, bottom-line introductions to the areas I thought were most likely

Preface xi
to be unfamiliar, but there are no doubt places where an unfamiliar (or, more
likely, vaguely familiar, but without enough associations for comfort) phrase
will suddenly appear. In these cases, it seems best to take the advice of John
von Neumann, who said to a student, “in mathematics you don’t understand
things. You just get used to them.”
One of the greatest pleasures in completing a book is the opportunity to
thank the many sources of knowledge, advice, wisdom, and support that made it
possible. My thanks first to the Institute for Advanced Study and the organizers
of the Women and Mathematics program for inviting me to give the lectures that
inspired this book. Thanks also to the National Science Foundation for generous
support while I wrote it.
Persi Diaconis introduced me to random matrix theory (and many other
things) and taught me to tell a good story.
Amir Dembo encouraged me to embark on this project and gave me valuable
advice about how to do it well.
I am grateful to Pierre Albin and Tyler Lawson for their constant willingness
to patiently answer all of my questions about geometry and algebra, and, if they
didn’t already know the answers, to help me wade through unfamiliar literature.
Experienced guides make all the difference.
Many thanks to Jon Keating, Arun Ram, and Michel Ledoux for answering
my questions about their work and pointing me to better approaches than the
ones I knew about. Particular thanks to Nathaël Gozlan for explaining tricky
details that eluded me.
My sincere thanks to Andrew Odlyzko for providing the figures based on his
computations of zeta zeros.
Thanks to my students, especially Tianyue Liu and Kathryn Stewart, whose
questions and comments on earlier drafts certainly enriched the end result.
The excellent and topical photograph for the frontispiece was found (I still
don’t know how) by Tim Gowers.
As ever, thanks to Sarah Jarosz, this time forUndercurrent, which got me
most of the way there, and to Yo-Yo Ma forSix Evolutions, which carried me
to the finish line.
And how to thank my husband and collaborator, Mark Meckes? We have
discussed the material in this book for so long and in so many contexts that his
viewpoint is inextricably linked with my own. He has lived with the writing of
this book, always willing to drop a (probably more important) conversation or
task in order to let me hash out a point that suddenly felt terribly urgent. If my
writing helps to illuminate the ideas I have tried to describe, it is because I got
to talk it out first at the breakfast table.

1
Haar Measure on the Classical Compact
Matrix Groups
1.1 The Classical Compact Matrix Groups
The central objects of study in this book are randomly chosen elements of the
classical compact matrix groups: the orthogonal groupO(n), the unitary group
U(n), and the symplectic groupS
p(2n). The groups are defined as follows.
Definition
1. Ann×nmatrixUoverRisorthogonalif
UU
T
=U
T
U=I n, (1.1)
whereI
ndenotes then×nidentity matrix, andU
T
is the transpose ofU.
The set ofn×northogonal matrices overRis denotedO(n).
2. Ann×nmatrixUoverCisunitaryif
UU

=U

U=I n, (1.2)
whereU

denotes the conjugate transpose ofU.Thesetofn×nunitary
matrices overCis denotedU(n).
3. A 2n×2nmatrixUoverCissymplecticifU∈U(2n)and
UJU
T
=U
T
JU=J, (1.3)
where
J:=
e
0I
n
−In0
l
. (1.4)
The set of 2n×2nsymplectic matrices overCis denotedS
p(2n).
Alternatively, the symplectic group can be defined as the set ofn×n
matricesUwith quaternionic entries, such thatUU

=In,whereU

is the
(quaternionic
H={a+bi+cj+dk:a,b,c,d∈R}
1

2 Haar Measure on the Classical Compact Matrix Groups
the skew-field of quaternions, satisfying the relations
i
2
=j
2
=k
2
=ijk=−1,
quaternionic conjugation is defined by
a+bi+cj+dk=a−bi−cj−dk.
Quaternions can be represented as 2×2 matrices overC:themap
a+bi+cj+dki−→
e
a+bi c+di
−c+di a−bi
l
is an isomorphism ofHonto
ie
zw
−wz
l
:z,w∈C
z
.
More generally, ifA,B,C,D∈M
n(R), then the matrix
M=A+Bi+Cj+Dk∈M
n(H)
is associated to the matrix
M
C=I2⊗A+iQ 2⊗B+Q 3⊗C+iQ 4⊗D,
where
Q
2:=
e
10 0−1
l
Q 3:=
e
01
−10
l
Q 4:=
e
01 10
l
and⊗denotes the Kronecker product. Any matrixM∈M
2n(C)of this form
has the property that
MJ=J
M
forJ=Q
3⊗Inas above, and the conditionUU

=InforU∈M n(H)is
equivalent toU
CU

C
=InoverC.
We will generally consider the symplectic group in its complex version, as a
subgroup of the (complex) unitary group, although certain geometric properties
of the group can be more cleanly characterized in the quaternionic form.
Note that it is immediate from the definitions thatUis orthogonal if and only
ifU
T
is orthogonal, andUis unitary or symplectic if and only ifU

is.
The algebraic definitions given above are nicely compact but may not make
the importance of these groups jump right out; the following lemma gives some
indication as to why they play such a central role in many areas of mathematics.

1.1 The Classical Compact Matrix Groups 3
Lemma 1.11. Let M be an n×n matrix overRorC. Then M is orthogonal
or unitary if and only if the columns of M form an orthonormal basis ofR
n
,
resp.C
n
.
2. For U an n×n matrix overR,U∈O(n)if and only if U acts as an isometry
onR
n
; that is,
bUv,Uwt=bv,wt
for all v,w∈R
n
.
3. For U an n×n matrix overC,U∈U(n)if and only if U acts as an isometry
onC
n
:
bUv,Uwt=bv,wt
for all v,w∈C
n
.
4. ConsiderC
2n
equipped with the skew-symmetric form
ω(v,w)=v
1wn+1+···+v nw2n−vn+1w1−···−v 2nwn=
a
k,i
Jklvkwi,
where
J=
e
0I
n
−In0
l
as above. For a2n×2n matrix U overC,U∈S
p(2n)if and only if U is
an isometry ofC
2n
which preservesω:
bUv,Uwt=bv,wtandω(Uv,Uw)=ω(v,w)
for all v,w∈C
2n
.
5. If U∈O(n)or U∈U(n),then|det(U)|=1.IfU∈S
p(2n),then
det(U)=1.
ProofNote that the(i,j)
th
entry ofU
T
U(ifUhas real entries) orU

U(ifU
has complex or quaternionic entries) is exactly the inner product of theith and
jth columns ofU.SoU
T
U=I norU

U=I nis exactly the same thing as
saying the columns ofUform an orthonormal basis ofR
n
orC
n
.
ForU∈M
n(R),bUv,Uwt=
b
U
T
Uv,w
t
,andsobUv,Uwt=bv,wtfor allv
andwif and only ifU
T
U=I. The proofs of parts 3 and 4 are similar. For part
5, on any of the groups,
|det(U)|
2
=det(U)
det(U)=det(U)det(U

)=det(UU

)=det(I n)=1.
The easiest way to see that ifU∈S
p(2n), then in fact det(U)=1istouse
the Pfaffian: for a skew-symmetric matrixA, the Pfaffian pf(A)is defined by a

4 Haar Measure on the Classical Compact Matrix Groups
sum-over-permutations formula along the lines of the determinant, and has the
property that for 2n×2nmatricesAandB,
pf(BAB
T
)=det(B)pf(A).
Applying this to the defining relation ofS
p(2n),
pf(J)=pf(UJU
T
)=det(U)pf(J),
and so (using the easily verified fact that pf(J) =0), det(U)=1. e
We sometimes restrict attention to the “special” counterparts of the orthog-
onal and unitary groups, defined as follows.
DefinitionThe setSO(n)⊆O(n)ofspecial orthogonal matricesis
defined by
SO(n):={U∈O(n):det(U)=1}.
The setSO

(n)⊆O(n)(thenegative coset) is defined by
SO

(n):={U∈O(n):det(U)=−1}.
The setSU(n)⊆U(n)ofspecial unitary matricesis defined by
SU(n):={U∈U(n):det(U)=1}.
Since the matrices of the classical compact groups all act as isometries of
C
n
, all of their eigenvalues lie on the unit circleS
1
⊆C. In the orthogonal and
symplectic cases, there are some built-in symmetries:
Exercise 1.2Show that each matrix inSO(2n+1)has 1 as an eigenvalue,
each matrix inSO

(2n+1)has−1 as an eigenvalue, and each matrix in
SO

(2n+2)has both−1and1aseigenvalues.
The setsO(n),U(n),S
p(2n),SO(n),andSU(n)of matrices defined above
arecompact Lie groups; that is, they are groups (with matrix multiplication as
the operation), and they are compact manifolds, such that the multiplication and
inverse maps are smooth. Moreover, these groups can naturally be viewed as
closed submanifolds of Euclidean space:O(n)andSO(n)are submanifolds of
R
n
2
;U(n)andSU(n)are submanifolds ofC
n
2
;andS p(2n)is a submanifold
ofC
(2n)
2
. Rather than viewing these matrices asn
2
-dimensional vectors, it is
more natural to view them as elements of the Euclidean spacesM
n(R)(resp.
M
n(C))ofn×nmatrices overR(resp.C), where the Euclidean inner products
are written as
bA,Bt
HS:=Tr(AB
T
)

1.1 The Classical Compact Matrix Groups 5
forA,B∈M
n(R),and
bA,Bt
HS:=Tr(AB

)
forA,B∈M
n(C). These inner products are called theHilbert–Schmidt inner
productson matrix space.
The Hilbert–Schmidt inner product induces a norm on matrices; it is some-
times called the Frobenius norm or the Schatten 2-norm, or just the Euclidean
norm. This norm isunitarily invariant:
UBV
HS=B HS
whenUandVare unitary (as is easily seen from the definition). This implies in
particular that ifU∈O(n)(resp.U(n)), then the mapR
U:Mn(R)→M n(R)
(resp.R
U:Mn(C)→M n(C)) defined by
R
U(M)=UM
is an isometry onM
n(R)(resp.M n(C)) with respect to the Hilbert–Schmidt
inner product.
The Hilbert–Schmidt norm is alsosubmultiplicative:
AB
HS≤A HSBHS.
In fact, this is true of all unitarily invariant norms (subject to the normalization
E
11=1), but it is particularly easy to see for the Hilbert–Schmidt norm: letB
have columnsb
1,...,b n;thenB
2
HS
=

n
j=1
|bj|
2
,where|·|is the Euclidean
norm onC
n
.Now,ABhas columnsAb 1,...,Ab n,andso
AB
2
HS
=
n
a
j=1
|Abj|
2
≤A
2
op
B
2
HS
,
whereA
op=sup
|x|=1
|Ax|is the operator norm ofA; i.e., the largest singular
value ofA. Writing the singular value decompositionA=UzVand using the
unitary invariance of the Hilbert–Schmidt norm,
A
2
op

2
1

n
a
j=1
σ
2
j
=z
2
HS
=A
2
HS
,
from which the submultiplicativity follows. Indeed, the sharper estimate
AB
HS≤A opBHS
is often useful.

6 Haar Measure on the Classical Compact Matrix Groups
The discussion above gives two notions of distance on the classical compact
matrix groups: first, the Hilbert–Schmidt inner product can be used to define
the distance between two matricesAandBby
d
HS(A,B):=A−B HS:=

bA−B,A−Bt
HS=

Tr

(A−B)(A−B)


.
(1.5
Alternatively, since, for example,A,B∈U(n)can be thought of as living in a
submanifold of Euclidean spaceM
n(C), one can consider thegeodesic distance
d
g(A,B)betweenAandB; that is, the length, as measured by the Hilbert–
Schmidt metric, of the shortest path lying entirely inU(n)betweenAandB.
In the case ofU(1), this is arc-length distance, whereas the Hilbert–Schmidt
distance defined inEquation (1.5)is the straight-line distance between two
points on the circle. Ultimately, the choice of metric is not terribly important:
Lemma 1.3Let A,B∈U(n).Then
d
HS(A,B)≤d g(A,B)≤
π
2
d
HS(A,B).
That is, the two notions of distance are equivalentin a dimension-free way.
ProofThe inequalityd
HS(A,B)≤d g(A,B)follows trivially from the fact that
the Hilbert–Schmidt distance is the geodesic distance in Euclidean space.
For the other inequality, first note thatd
g(A,B)≤
π
2
dHS(A,B)forA,B∈
U(1); that is, that arc-length on the circle is bounded above by
π
2
times
Euclidean distance.
Next, observe that bothd
HS(·,·)andd g(·,·)are translation-invariant; that is,
ifU∈U(n),then
d
HS(UA,UB)=d HS(A,B)andd g(UA,UB)=d g(A,B).
In the case of the Hilbert–Schmidt distance, this is immediate from the fact
that the Hilbert–Schmidt norm is unitarily invariant. For the geodesic distance,
translation invariance follows from the fact that, since any matrixU∈U(n)
acts as an isometry of Euclidean space, every path betweenAandBlying in
U(n)corresponds to a path betweenUAandUBof the same length, also lying
inU(n).
Now fixA,B∈U(n)and letA
−1
B=UtU

be the spectral decomposition
ofA
−1
B. Then for either distance,
d(A,B)=d(I
n,A
−1
B)=d(I n,UtU

)=d(U

U,t)=d(I n,t),
and so it suffices to assume thatA=I
nandBis diagonal.

1.2 Haar Measure 7
WriteB=diag(e
iθ1,...,e
iθn). Then the length of the path inU(n)fromA
toBgiven byU(t):=diag(e
itθ1,...,e
itθn),for0≤t≤1is
r
1
0p
p
U

(t)
p
p
HS
dt=
r
1
0p
p
pdiag(iθ
1e
itθ1
,...,iθ ne
itθn
)
p
p
p
HS
dt
=
r
1
0
θ
2
1
+···+θ
2
n
dt

π2
r
1
0
|1−e
iθ1|
2
+···+|1−e
iθn|
2
dt
=
π
2
p
p
pI
n−diag(e
iθ1
,...,e
iθn
)
p
p
p
HS
,
using the fact that
θ
2
=dg(1,e

)
2

π
2
4
d
HS(1,e

),
as noted above. e
1.2 Haar Measure
The main goal of this book is to answer the broad general question: What is
a random orthogonal, unitary, or symplectic matrix like? To do this, a natural
probability measure on each of these groups is needed.
Just as the most natural probability measure (i.e., uniform measure) on the
circle is defined by rotation invariance, ifGis one of the matrix groups defined
in the last section, a “uniform random element” ofGshould be a randomU∈G
whose distribution istranslation-invariant;thatis,ifM∈Gis any fixed matrix,
then the equality in distribution
MU
d
=UM
d
=U
should be satisfied. Phrased slightly differently, the distribution of a uniform
random element ofGshould be a translation-invariant probability measureμ
onG: for any measurable subsetA⊆Gand any fixedM∈G,
μ(MA)=μ(AM)=μ(A),
whereMA:={MU:U∈A}andAM:={UM:U∈A}.
It is a theorem due to A. Haar that there is one, and only one, way to do this.
Theorem 1.4Let G be any ofO(n),SO(n),U(n),SU(n),orS
p(2n).
Then there is a unique translation-invariant probability measure (calledHaar
measure)onG.

8 Haar Measure on the Classical Compact Matrix Groups
The theorem is true in much more generality (in particular, any compact Lie
group has a Haar probability measure). In the most general case the property
of left-invariance is not equivalent to that of right-invariance, but in the case
of compact Lie groups, left-invariance implies right-invariance and vice versa,
so the phrase “translation invariance” will be used in what follows, and will be
assumed to include both left- and right-invariance.
Exercise 1.5
1. Prove that a translation-invariant probability measure onO(n)is invariant
under transposition: ifUis Haar-distributed, so isU
T
.
2. Prove that a translation-invariant probability measure onU(n)is invariant
under transposition and under conjugation: ifUis Haar-distributed, so are
bothU
T
andU

.
Theorem 1.4is an existence theorem that does not itself provide a description
of Haar measure in specific cases. In the case of the circle, i.e.,U(1),itis
clear that Haar measure is just (normalized
section gives six different constructions of Haar measure onO(n), with some
comments about adapting the constructions to the other groups. For most of the
constructions, the resulting measure is only shown to be invariant one one side;
the invariance on the other side then follows from the general fact mentioned
above that on compact Lie groups, one-sided invariance implies invariance on
both sides.
The Riemannian Perspective
It has already been noted thatO(n)⊆M
n(R)and that it is a compact sub-
manifold. It has two connected components:SO(n)andSO

(n),thesetof
orthogonal matricesUwith det(U)=−1. At each pointUofO(n),there
is a tangent spaceT
U(O(n)), consisting of all the tangent vectors toO(n)
based atU.
A map between manifolds induces a map between tangent spaces as follows.
LetM
1,M2be manifolds andϕ:M 1→M 2.Ifx∈T pM1, then there is a curve
γ: [0, 1]→M
1such thatγ(0)=pandγ

(0)=x.Thenϕ◦γis a curve in
M
2withϕ◦γ(0)=ϕ(p),and(ϕ◦γ)

(0)is a tangent vector toM 2atϕ(p).
We take this to be the definition ofϕ
∗(x)(it must of course be checked that this
gives a well-defined linear map onT
pM1for eachp).
A Riemannian metricgon a manifoldMis a family of inner products, one
on the tangent spaceT
pMtoMat each pointp∈M. The submanifoldO(n)
inherits such a metric from M
n(R), since at each pointUinO(n),T U(O(n))
is a subspace ofT
U(Mn(R))

=M n(R). Because multiplication by a fixed

1.2 Haar Measure 9
orthogonal matrixVis an isometry of M
n(R), the induced map on tangent
spaces is also an isometry: ifU∈O(n)withX
1,X2∈TU(O(n))tangent
vectors toO(n)atU,andR
V:O(n)→O(n)denotes multiplication by a
fixedV∈O(n),then
g
VU((RV)∗X1,(RV)∗X2)=g U(X1,X2).
On any Riemannian manifold, the Riemannian metric uniquely defines a
notion of volume. Since the metric is translation-invariant, the normalized vol-
ume form onO(n)is a translation-invariant probability measure; that is, it is
Haar measure.
Since each of the classical compact matrix groups is canonically embedded
in Euclidean space, this construction works the same way in all cases.
An Explicit Geometric Construction
Recall thatU∈O(n)if and only if its columns are orthonormal. One way to
construct Haar measure onO(n)is to add entries to an empty matrix column by
column (or row by row), as follows. First choose a random vectoru
1uniformly
from the sphereS
n−1
⊆R
n
(that is, according to the probability measure
defined by normalized surface area). Takeu
1as the first column of the matrix;
by construction,u
1=1. Now chooseu 2randomly according to surface area
measure on
o
u

1
d
∩S
n−1
=
á
x∈R
n
:x=1,bx,u 1t=0
ý
and let this be the second column of the matrix. Continue in this way; each
column is chosen uniformly from the unit sphere of vectors that are orthogonal
to each of the preceding columns. The resulting matrix


||
u
1...u n
||


is obviously orthogonal; the proof that its distribution is translation-invariant is
as follows.
Observe that ifMis a fixed orthogonal matrix, then since
M


||
u
1...u n
||

⎦=


||
Mu
1...Mu n
||

⎦,

10 Haar Measure on the Classical Compact Matrix Groups
the first column ofM


||
u
1...u n
||

⎦is constructed by choosingu
1uniformly
fromS
n−1
and then multiplying byM.ButM∈O(n)means thatMactsasa
linear isometry ofR
n
, so it preserves surface area measure onS
n−1
.Thatis,the
distribution ofMu
1is exactly uniform onS
n−1
.
Now, sinceMis an isometry,bMu
2,Mu1t=0, and becauseMis an isometry
ofR
n
, it follows thatMu 2isuniformly distributedon
(Mu
1)

∩S
n−1
:=
á
x∈R
n
:|x|=1,bMu 1,xt=0
ý
.
So the second column ofM

u
1...un

is distributed uniformly in the unit sphere
of the orthogonal complement of the first column.
Continuing the argument, the distribution ofM

u
1...un

is exactly the same
as the distribution of

u
1...un

; i.e., the construction is left-invariant. It follows
by uniqueness that it produces Haar measure onO(n).
To construct Haar measure onU(n), one need only draw the columns uni-
formly from complex spheres inC
n
. To get a random matrix inSO(n),the
construction is identical except that there is no choice about the last column;
the same is true forSU(n).
The analogous construction on the representation of elements ofS
p(2n)by
2n×2nunitary matrices works as follows. ForUto be inS
p(2n), its first
columnu
1must lie in the set
{x∈C
2n
:x=1,bx,Jxt=0},
whereJis the matrix defined in (1.4). This conditionbx,Jxt=0 defines a
hyperboloid inC
n
(Jis unitarily diagonalizable and has eigenvaluesiand−i,
each with multiplicityn). The set above is thus the intersection of the sphere
with this hyperboloid; it is an(n−2)-dimensional submanifold ofC
n
from
which we can choose a point uniformly: this is how we chooseu
1.Ifn>1,
one then chooses the second column uniformly from the set
{x∈C
2n
:x=1,bx,u 1t=0,bx,Jxt=0,bx,Ju 1t=0};
forn=1, one chooses the second column uniformly from
{x∈C
2
:x=1,bx,u 1t=0,bx,Jxt=0bx,Ju 1t=−1}.
The construction continues: thekth columnu
kis chosen uniformly from the
intersection of the unit sphere, the hyperboloid{x:bx,Jxt=0}, and the (affine)
subspaces given by the conditionsbx,Ju
it=0for1≤i≤min{k−1,n}and
bx,Ju
it=−1forn+1≤i<k(ifk≥n+2). The argument that this

1.2 Haar Measure 11
construction is invariant under right-translation by an element ofS
p(2n)is
similar to the argument above, making use of the fact that forM∈S
p(2n)
given,Mis an isometry ofC
n
andMJ=JM.
A Different Inductive Construction
In this construction, a random element ofO(n)is built up successively from
smaller groups. Since it is clear how to choose a random element ofO(1)(flip
a coin!), one need only describe how to get a random element ofO(n)from a
random element ofO(n−1).
LetU
n−1be distributed according to Haar measure onO(n−1),andletM∈
O(n)be independent ofU
n−1and have its first column distributed uniformly
onS
n−1
⊆R
n
. (The distribution of the remaining columns is irrelevant.) Then
defineU
nby
U
n:=M
e
10
0U
n−1
l
. (1.6)
It is not hard to see that the columns ofU
nare distributed as described in the
previous approach: It is clear that in the matrix
e
10
0U
n−1
l
,
the second column is uniformly distributed in the orthogonal complement of
the first, the third is uniform in the orthogonal complement of the first two, etc.
So for a deterministicM∈O(n), the first column of
M
e
10
0U
n−1
l
is justm
1(the first column ofM), the second column is uniformly distributed in
the orthogonal complement ofm
1, etc. TakingMto be random but independent
ofU
n−1, it follows that the distribution of the columns ofU nis exactly as in
the previous construction.
The construction works exactly the same way for the unitary and special
orthogonal and unitary groups, replacingU
n−1by a Haar-distributed element
of the next smaller-rank group, and choosingMin the desired group with its
first column uniformly in the sphere inR
n
orC
n
, as appropriate.
For the symplectic group, the construction is similar, albeit slightly more
complicated: letU
n−1be Haar-distributed inS p(2(n−1)),andletM∈
S
p(2n)be independent ofU n−1and with its first columnm 1uniform in

12 Haar Measure on the Classical Compact Matrix Groups
{x∈C
2n
:x=1,bx,Jxt=0}
and its(n+1)
st
columnm n+1uniform in
{x∈C
2n
:x=1,bx,m 1t=0,bx,Jxt=0bx,Jm 1t=−1}.
WriteU
n−1as a 2×2 matrix of(n−1)×(n−1)blocks:
U
n−1=
e
(U
n−1)1,1(Un−1)1,2
(Un−1)2,1(Un−1)2,1
l
,
and defineU
n∈Sp(2n)by
U
n:=M




1000
0(U
n−1)1,10(U n−1)1,2
0010
0(U
n−1)2,10(U n−1)2,2




.
One can then check thatU
n∈Sp(2n)and has its columns distributed as in the
previous construction of Haar measure.
The Gauss–Gram–Schmidt Approach
This construction is probably the most commonly used description of Haar
measure, and also one that is easy to implement on a computer.
Generate a random matrixXby filling ann×nmatrix, with independent,
identically distributed (i.i.d.) standard Gaussian entries{x
i,j}. That is, the joint
density (with respect to
ü
n
i,j=1
dxij)ofthen
2
entries ofXis given by
1
(2π)
n
2
2
n
v
i,j=1
e

x
2
ij
2=
1
(2π)
n
2
2
exp
´

X
2
HS
2
f
(the collection of these random matrices is known as the real Ginibre ensemble).
The distribution ofXis invariant under multiplication by an orthogonal matrix:
by a change of variables, the density of the entries ofY=MXwith respect to
ü
dy
ijis
|det(M
−1
)|
(2π)
n
2
2
exp
i

M
−1
Y
2
2
z
=
1
(2π)
n
2
2
exp
i

Y
2
2
z
,
sinceM
−1
is an isometry.
That is, the distribution above is translation-invariant, but it is not Haar mea-
sure because it does not produce an orthogonal matrix. To make it orthogonal,

1.2 Haar Measure 13
we use the Gram–Schmidt process. Performing the Gram–Schmidt process
commutes with multiplication by a fixed orthogonal matrixM:letx
idenote
the columns ofX. Then, for example, when thex
1component is removed from
x
2,x2is replaced withx 2−πx1,x2tx2. If the result is then multiplied byM,the
resulting second column is
Mx
2−πx1,x2tMx2.
If, on the other hand, the multiplication byMis done before applying the Gram–
Schmidt algorithm, the result is a matrix with columnsMx
1,...,Mx n.Now
removing the component in the direction of column 1 from column 2, the new
column2is
Mx
2−πMx 1,Mx2tMx1=Mx2−πx1,x2tMx2,
sinceMis an isometry.
In other words, ifT:GL
n(R)→O(n)is the map given by performing the
Gram–Schmidt process (GL
n(R)denotes the group ofn×ninvertible matrices
overR), then for a fixed orthogonal matrixM,
MT(X)=T(MX)
d
=T(X).
That is, the distribution ofT(X)is supported onO(n)and is translation-
invariant, meaning that we have once again constructed Haar measure.
Remarks
1. In different terminology, the argument above says that ifXis a matrix of i.i.d.
standard Gaussian random variables andX=QRis theQR-decomposition
obtained via the Gram–Schmidt process, thenQis Haar-distributed on the
orthogonal group. But,WARNING:TheQR-decomposition of a matrix is
not uniquely defined, and most computer algebra packagesdo notuse the
Gram–Schmidt algorithm to produce it. The result of which is that having
a computer generate a matrix of i.i.d. standard Gaussian random variables
and then returningQfrom its internalQR-algorithm will not produce a Haar-
distributed matrix; see [81]and[42] for further discussion.
2. The same algorithm as above works overCorHto produce a random
unitary or symplectic matrix. To produce a random element ofSO(n)or
SU(n), one simply needs a final step: after carrying out the Gram–Schmidt
process, multiply the final column by the necessary scalar in order to force
the determinant of the matrix to be 1; it is not hard to see that this produces
Haar measure on the reduced group.

14 Haar Measure on the Classical Compact Matrix Groups
A Second Gaussian Construction
Again, letXbe ann×nrandom matrix with i.i.d. standard Gaussian entries. It
is easy to see thatXhas ranknwith probability 1, which implies in particular
thatX
T
Xis a symmetric ranknmatrix, and so
X
T
X=V
T
diag(d 1,...,d n)V
for someV∈O(n)andd
1,...,d n>0;(X
T
X)
−1/2
is then defined by
(X
T
X)
−1/2
=V
T
diag(d
−1/2
1
,...,d
−1/2
n
)V.
Now define the random matrixUby
U:=X(X
T
X)
−1/2
. (1.7)
ThenUis orthogonal:
U
T
U=(X
T
X)
−1/2
X
T
X(X
T
X)
−1/2
=In,
and since forV∈O(n)fixed,VX
d
=X,
U
d
=VX((VX)
T
VX)
−1/2
=VX(X
T
X)
−1/2
=VU.
That is,Uis distributed according to Haar measure onO(n).
Haar Measure onSO(n)andSO

(n)
The constructions above describe how to choose a uniform random matrix
fromO(n), but as we noted above,O(n)decomposes neatly into two pieces,
those matrices with determinant 1 (SO(n)) and those with determinant−1
(SO

(n)).Theorem 1.4says thatSO(n)has a unique translation-invariant
probability measure; it is clear that this is simply the restriction of Haar measure
onO(n)toSO(n).
There is also a measure that is sometimes called Haar measure onSO

(n),
which is the restriction of Haar measure fromO(n)toSO

(n).Theset
SO

(n)is of course not a group, but rather a coset of the subgroupSO(n)
in the groupO(n); we continue to use the name “Haar measure onSO

(n)”
because this measure is a probability measure invariant under translation
withinSO

(n)by any matrix fromSO(n). Haar measure onSO(n)and Haar
measure onSO

(n)are related as follows: ifUis Haar-distributed inSO(n)
andjUis any fixed matrix inSO

(n),thenjUUis Haar-distributed inSO

(n).

1.2 Haar Measure 15
Euler Angles
Recall the spherical coordinate system inR
n
: eachx∈R
n
has spherical coor-
dinatesr,θ
1,...,θ n−1, such that
x
1=rsin(θ n−1)···sin(θ 2)sin(θ1)
x
2=rsin(θ n−1)···sin(θ 2)cos(θ 1)
.
.
.
x
n−1=rsin(θ n−1)cos(θ n−2)
x
n=rcos(θ n−1).
Here,
0≤r<∞,
0≤θ
1<2π,
0≤θ
k≤π,2≤k≤n.
The spherical coordinates of a point are uniquely determined, except in the
cases that someθ
k(k≥2) is 0 orπ,orr=0.
Spherical coordinates are the basis for the parametrization ofSO(n)by the
so-called Euler angles. Forθ∈[0, 2π)and 1≤k≤n−1, let
U
k(θ):=




I
k−1
cos(θ)sin(θ)
−sin(θ)cos(θ)
I
n−k−1




.
All matrices inSO(n)can be decomposed as a product of matrices of this type,
as follows.
Proposition 1.6For any U∈SO(n), there are angles (called Euler angles)

k
j
}1≤k≤n−1
1≤j≤k with0≤θ
k
1
<2πand0≤θ
k
j
<πfor j =1, so that
U=U
(n−1)
···U
(1)
,
where
U
(k)
=U1(θ
k
1
)···U k(θ
k
k
).
The Euler angles are unique except if someθ
k
j
is 0 orπ(j≥2).
ProofObserve that the result is vacuous forn=1 and holds trivially for
n=2. Suppose then that it holds onSO(n−1),andletU∈SO(n).Let

16 Haar Measure on the Classical Compact Matrix Groups
1,θ
n−1
1
,...,θ
n−1
n−1
be the spherical coordinates ofUe n,wheree jdenotes thejth
standard basis vector ofR
n
.Then
U
1(θ
n−1
1
)···U n−1(θ
n−1
n−1
)en=U1(θ
n−1
1
)···U n−2(θ
n−1
n−2
)








0
.
.
.
0
sin(θ
n−1
n−1
)
cos(θ
n−1
n−1
)








=···=







sin(θ
n−1
n−1
)···sin(θ
n−1
2
)sin(θ
n−1
1
)
sin(θ
n−1
n−1
)···sin(θ
n−1
2
)cos(θ
n−1
1
)
···
sin(θ
n−1
n−1
)cos(θ
n−1
n−2
)
cos(θ
n−1
n−1
)







;
that is,
U
1(θ
n−1
1
)···U n−1(θ
n−1
n−1
)en=Uen,
and so
˜
U
1(θ
n−1
1
)···U n−1(θ
n−1
n−1
)

−1
U=
e
jU0
01
l
(1.8
for somejU∈SO(n−1). By the induction hypthesis,jUcan be written as
jU=U
(n−2)
···U
(1)
.
By mild abuse of notation we now consider each of the implicit factorsU
i(θ
k
i
),a
priori elements ofSO(n−1),tobeelementsofSO(n)fixinge
n. The claimed
factorization ofUfollows by multiplying both sides of (1.8)byU
(n−1)
:=
U
1(θ
n−1
1
)···U n−1(θ
n−1
n−1
). e
Haar measure onSO(n)can be characterized as a distribution on the Euler
angles. Observe first that by a right-to-left version of the column-by-column
construction, ifU∈SO(n)is distributed according to Haar measure, thenUe
n
is a uniform random point onS
n−1
. Recall that the uniform probability measure
on the sphereS
n−1
is given in spherical coordinates by

!
n
2
"

n/2
sin
n−2
(θn−1)···sin(θ 2)dθ1···dθ n−1.
That is, the{θ
n−1
k
}1≤k≤n−1 subset of the Euler angles ofUhave density

!
n
2
"

n/2
sin
n−2

n−1
n−1
)···sin(θ
n−1
2
)

1.3 Lie Group Structure and Character Theory 17
with respect todθ
n−1
1
···dθ
n−1
n−1
.Now,givenUe n, the vectorUe n−1is uniformly
distributed on the unit sphere in orthogonal complement ofUe
n; equivalently,
givenθ
n−1
1
,...,θ
n−1
n−1
,
˜
U
1(θ
n−1
1
)···U n−1(θ
n−1
n−1
)

−1
Uen−1
is distributed uniformly on the(n−1)-dimensional unit sphere ofe

n
.Since

n−2
k
}1≤k≤n−2 are exactly the (angular) spherical coordinates of this vector, it
follows that{θ
n−2
k
}1≤k≤n−2 are independent of{θ
n−1
k
}1≤k≤n−1 , with density

o
n−1
2
d

(n−1)/2
sin
n−3

n−2
n−2
)···sin(θ
n−2
2
).
Continuing in this way, the Euler angles ofUare independent, with joint density
n−1
v
k=1
#

!
k
2
"

k/2
$
k
v
j=1
sin
j−1

k
j
). (1.9)
From this construction, one can describe a natural analog forSO

(n);any
U∈SO

(n)can be written as
U=U
(n−1)
···U
(1)
U
(0)
, (1.10)
where
U
(0)
=
e
−1
I
n−1
l
,
andU
(n−1)
···U
(1)
is the Euler angle decomposition ofUU
(0)
; i.e., the matrix
with the same columns asU, except for a sign change in the first column. That
is, Haar measure onSO

(n)can be described by choosing angles{θ
k
j
}1≤k≤n−1
1≤j≤k
according to the density in (1.9) and then lettingUbe given by the formula in
(1.10).
To chooseUaccording to Haar measure onO(n), one simply chooses the
Euler angles according to (1.9), and then includes theU
(0)
factor with proba-
bility
1
2
, independent of the choice of angles.
1.3 Lie Group Structure and Character Theory
The Classical Groups as Lie Groups
As we have already noted, the classical compact matrix groups are Lie
groups; i.e., they are groups and they are differentiable manifolds such that the

18 Haar Measure on the Classical Compact Matrix Groups
multiplication operation(A,B)i→ABand the inversion operationAi→A
−1
are smooth maps. Each of the groups has an associatedLie algebra,which
is the tangent space to the group at the identity matrix. Lie algebras play an
important role in understanding Lie groups, because they are relatively simple
objects geometrically (vector spaces!), but they come equipped with an extra
algebraic structure called theLie bracket, which encodes much of the geometry
of the group itself.
The tool that connects the Lie algebra to the Lie group is the exponential map.
While one can define the exponential map on Riemannian manifolds in general,
the definition in the setting of the classical compact matrix groups can be made
very concrete; this in particular makes the source of the terminology clear.
DefinitionLetX∈M
n(C). The matrix exponentiale
X
is defined by
e
X
:=In+X+
1
2
X
2
+
1
3!
X
3
+···.
We will make frequent use of the following basic facts.
Lemma 1.71. The sum defining the matrix exponential is convergent for any
X∈M
n(C).
2. If X∈M
n(C),e
X
is invertible with inverse e
−X
.
3. If X and Y commute, then e
X+Y
=e
X
e
Y
; this need not be true if X and Y do
not commute.
Proof1. Recall that the Hilbert–Schmidt norm is submultiplicative; from this
it follows that
N
a
j=0
p
p
X
j
p
p
HS
j!

N
a
j=0
X
j
HS
j!


a
j=0
X
j HS
j!
<∞
for allN,andso


j=1
1
j!
X
j
is convergent.
2. This point follows easily from the next, sinceXand−Xcommute.
3. ForN∈N,sinceXandYcommute, we have
N
a
j=0
1
j!
(X+Y)
j
=
N
a
j=0
1
j!
j
a
k=0
%
j
k
&
X
k
Y
j−k
=
N
a
k=0
1k!
X
k
N
a
j=k
1
(j−k)!
Y
j−k
=
N
a
k=0
1
k!
X
k
N−k
a
j=0
1
(j)!
Y
j
.
Now letN→∞.
It is easy to cook up examples of noncommutingXandYfor which
e
X+Y
=e
X
e
Y
. e

1.3 Lie Group Structure and Character Theory 19
Suppose now thatX=UAU

withU∈U(n).Then
e
X
=

a
j=0
1
j!
(UAU

)
j
=

a
j=0
1
j!
UA
j
U

=Ue
A
U

.
In particular, ifXis unitarily diagonalizable (i.e., normal), withX=Udiag
(d
1,...,d n)U

,thene
X
=Udiag(e
d1,...,e
dn)U

. More generally, given a
functionfonC, for unitarily diagonalizableXas above, wedefine f(X)by this
route: ifX=Udiag(d
1,...,d n)U

,then
f(X):=Udiag(f(d
1),...,f(d n))U

.
This prodecure is referred to as thefunctional calculus.
The functionγ(t)=e
tX
defines a one-parameter subgroup ofGL n(C)(the
group of invertiblen×nmatrices overC), sincee
tX
e
sX
=e
(t+s)X
.More
geometrically,γ(t)is a curve inGL
n(C)withγ(0)=I nandγ

(0)=X.
In general, the role of the exponential map in Riemannian geometry is that
it gives a local diffeomorphism of the tangent space to a manifold at a point to
a neighborhood of that point in the manifold. In the present context, it is what
maps the Lie algebra of a closed subgroup ofGL
n(C)down to a neighborhood
ofI
nin the subgroup. The following lemma, whose proof we omit, gives the
precise statement needed.
Lemma 1.8If G is any closed subgroup of GL
n(C), then X is an element of
the Lie algebra of G if and only if e
tX
∈G for all t≥0.ThemapXi→e
X
gives a diffeomorphism of a neighborhood of0in the Lie algebra of G to a
neighborhood of I
nin G.
The lemma allows us to identify the Lie algebras of the classical groups
concretely, as follows.
Lemma 1.9
1. The Lie algebra ofO(n)is
o(n)=
'
X∈M
n(R):X+X
T
=0
(
.
2. The Lie algebra ofSO(n)is
so(n)=o(n)=
'
X∈M
n(R):X+X
T
=0
(
.
3. The Lie algebra ofU(n)is
u(n)=
á
X∈M
n(C):X+X

=0
ý
.

20 Haar Measure on the Classical Compact Matrix Groups
4. The Lie algebra ofSU(n)is
su(n)=
á
X∈M
n(C):X+X

=0, Tr(X)=0
ý
.
5. The Lie algebra ofS
p(2n)⊆U(n)is
sp(2n)=
á
X∈M
2n(C):X+X

=0,XJ+JX

=0
ý
.
The quaternionic form of the Lie algebra ofS
p(2n)⊆M n(H)is
su
H(n)=
á
X∈M n(H):X+X

=0
ý
,
where X

denotes the quaternionic conjugate transpose.
Proof
1. First, ifγ: [0, 1]→O(n)is a curve withγ(0)=I
n,thenγ(t)γ (t)
T
=In
for eacht. Differentiating gives that
γ

(t)γ (t)
T
+γ(t)γ

(t)
T
=0
for allt, and so in particular, ifX=γ

(0)is the tangent vector toγatI n,
thenX+X
T
=0.
On the other hand, givenXwithX+X
T
=0,
e
tX
(e
tX
)
T
=e
tX
e
tX
T
=e
tX
e
−tX
=In,
and soγ(t)=e
tX
is a curve inO(n)withγ(0)=I nandγ

(0)=X.That
is,Xis in the tangent space toO(n)atI
n.
2. SinceSO(n)is a subgroup ofO(n), the tangent space toSO(n)atI
nis
a subspace ofo(n). In fact, it is clear geometrically that the Lie algebras
ofO(n)andSO(n)must be the same, sinceO(n)is just the union of two
disconnected copies ofSO(n)(the second copy being the negative coset
SO

(n)).
3. Exactly analogous to 1.
4. As in the orthogonal case, sinceSU(n)is a subgroup ofU(n), the tan-
gent space toSU(n)atI
nis a subspace ofu(n); in this case, however, it
is not the whole space. The additional condition for a curveγ(t)to lie
inSU(n)is that detγ(t)=1forallt. Using the functional calculus, if
X=Udiag(d
1,...,d n)U

,
det(e
X
)=det(Udiag(e
d1
,...,e
dn
)U

)=e
d1+···+d n
=e
TrX
.
In particular, ifγ(t)=e
tX
,then
d
dt
(det(γ (t)))=
d
dt
o
e
tTrX
d
=(TrX)e
tTrX
,

1.3 Lie Group Structure and Character Theory 21
and soXis tangent toSU(n)atI
nif and only ifX∈u(n)(i.e.,X+X

=0)
and TrX=0.
5. For the complex form, sinceS
p(2n)is a subgroup ofU(2n), the Lie algebra
ofS
p(2n)is a subspace ofu(2n), hence the first condition insp(2n).For
the second, observe that ifγ: [0, 1]→S
p(2n)is a curve withγ(0)=I 2n,
then differentiating the requirement thatγ(t)Jγ(t)

=Jfor alltgives that
γ

(t)Jγ(t)

+γ(t)Jγ

(t)

=0; ifX=γ

(0), then evaluating att=0gives
thatXJ+JX

=0.
On the other hand, ifXsatisfiesX+X

=0andXJ+JX

=0, then
e
tX
J=

a
n=0
t
n
n!
X
n
J=

a
n=0
t
n
n!
(−X
n−1
JX

)=

a
n=0
t
n
n!
(X
n−2
J(X

)
2
)
=···=J

a
n=0
(−tX

)
n
n!
=J(e
−tX
)

,
and soe
tX
∈Sp(2n)for allt.
Verifying the quaternionic form is exactly the same as the unitarycase.
e
Observe that even though the matrices inu(n),su(n),andsp(2n)have com-
plex entries, they arerealvector spaces only; they are not closed under multipli-
cation by complex scalars. They do inherit a real inner product structure from
the Euclidean structure of the spaces in which they reside: the inner product on
u(n),su(n),andsp(2n)is
bX,Yt:=Re(Tr(XY

)).
This is the unique real inner product that defines the same norm as the usual
Hilbert–Schmidt inner product onu(n),su(n),andsp(2n).
Representation Theory
We have already encountered two natural actions of the classical compact
groups on vector spaces; namely, onR
n
orC
n
by matrix-vector multiplication,
and onM
n(R)orM n(C)by multiplication. The topic of representation theory
is to understandallthe ways a group can act on a vector space. This is a vast
and well-developed field, but in the context of the random matrix theory on
these groups, our main interest will be not in representation theory in its full
glory, but specifically in character theory. Essentially the reason for this is that
the irreducible characters form a basis for the space of class functions on a

22 Haar Measure on the Classical Compact Matrix Groups
group; i.e., those functions that are constant on conjugacy classes. Conjugacy
classes of matrices within the classical groups are exactly the set of matrices
with a given set of eigenvalues, and so ultimately, a good basis of the space of
class functions gives us a route to studying eigenvalues.
Arepresentationof a finite groupGis a group homomorphismρ:G→
GL(V),whereVis a finite-dimensional vector space andGL(V)is the group of
invertible linear maps onV.Arepresentationof a Lie groupGis again a map
ρ:G→GL(V)for a finite-dimensional vector spaceV, which is both a homo-
morphism of groups and is required to be smooth. That is, a representiation of
Gis a way of seeingGas acting onV, in a way that respects the structure ofG
as a (Lie) group. Usually the notation for the map itself is suppressed, and one
refers to a representationVof a groupG, and writesg·vor justgvrather than
ρ(g)(v).
A representationVofGisirreducibleifVhas no proper nontrivial sub-
spaces which are invariant under the action ofG.
ExampleLetS
nbe the symmetric group onnletters;S nacts onR
n
by per-
muting the coordinates of a vector. This representation ofS
nis not irreducible,
since the subspaceV
1spanned by(1, 1,...,1)is invariant under the action of
S
n, as is the complementary subspaceV 2={(x 1,...,x n):x1+···+x n=0}.
Of course,V
1is an irreducible representation ofS n(it is called the trivial one-
dimensional representation). Less obviously, though not too hard to see,V
2is
also an irreducible representation ofS
n;itiscalledthestandard representation.
Known representations of a group give rise to new ones in various ways.
The simplest are by taking direct sums or tensor products: ifVandWare
representations ofG,thenV⊕WandV⊗Ware also representations ofG,
via the actions
g((v,w))=(gv,gw)g(v⊗w)=(gv)⊗(gw).
A representationValso induces a representation on the dual spaceV

of scalar-
valued linear functions onV:ifv

∈V

,then
(gv

)(v):=v

(g
−1
v).
A linear mapϕ:V→Wbetween representationsVandWof a groupGis
calledG-linear if for allv∈Vand allg∈G,
g·ϕ(v)=ϕ(g·v).
Two representationsVandWareisomorphicif there is aG-linear isomorphism
between the vector spacesVandW.

1.3 Lie Group Structure and Character Theory 23
A fundamental fact is that finite-dimensional representations of finite groups
and compact Lie groups can be decomposed into direct sums of irreducible
representations. The proof in either case is essentially the same, and rests on
the fact that for any representationVof such a group, one can define an inner
product onVsuch that each group element acts as an isometry. Indeed, take
any inner productb·,·tonV, and define
bv,wt
G:=
´
1
|G|

g∈G
bgv,gwt,Gfinite;
)
bgv,gwtdg, Ga compact Lie group,
where the integration in the second case is with respect to normalized Haar
measure onG.ThenifWeVis a nonzero subspace that is invariant under the
action ofG, the orthogonal complementW

ofWwith respect tob·,·t
Gis also
G-invariant, andV=W⊕W

. Continuing in this way defines a decomposition.
Suppose now thatV=⊕
k
i=1
V
⊕ai
i
=⊕
i
j=1
W
⊕bj
j
, with theV iand theW j
irreducible representations. If a given summandV imeets a summandW jnon-
trivially, then they must be equal because otherwise their intersection would be
a non-trivialG-invariant subspace of at least one of them. The two decompo-
sitions can thus differ at most by permuting the summands. It therefore makes
sense to talk about the number of times an irreducible representationV
ioccurs
in a representationV.
A basic tool in the representation theory of finite groups and compact Lie
groups is the following.
Lemma 1.10 (Schur’s lemma)Let G be a finite group or a compact Lie group,
and let V and W be finite-dimensional irreducible complex representations of
G. Letϕ:V→W be a G-linear map.
1. Eitherϕis an isomorphism orϕ=0.
2. If V=W, then there is aλ∈Csuch thatϕ=λ·I, with I the identity map
on V.
Proof1. SinceϕisG-linear, kerϕis an invariant subspace ofV, and sinceV
is irreducible, this means that either kerϕ=V(and henceϕ=0) or else
kerϕ={0},sothatϕis injective. In that case, imϕis a nonzero invariant
subspace ofW, and hence imϕ=W:ϕis an isomorphism.
2. SinceVis a complex vector space,ϕmust have at least one eigenvalue; i.e.,
there is aλ∈Csuch that ker(ϕ−λI) ={0}. But then sinceVis irreducible,
ker(ϕ−λI)=Vand thusϕ=λI. e
Given a representationρ:G→GL(V)of a groupG,thecharacterof the
representation is the function

24 Haar Measure on the Classical Compact Matrix Groups
χ
V(g)=Tr(ρ(g)).
Note that ifh∈G,then
χ
V(hgh
−1
)=Tr(ρ(hgh
−1
))=Tr(ρ(h)ρ(g)ρ(h)
−1
)=Tr(ρ(g)),
and soχ
Vis a class function onG. The following properties are easy to check.
Proposition 1.11Let V and W be representations of G. Then
•χ
V(e)=dim(V)
•χ
V⊕W=χV+χW
•χV⊗W=χVχW
•χV
∗=
χV.
ProofExercise. e
Since all finite-dimensional representations can be decomposed into irre-
ducible representations, the second property above says that all characters can
be written as sums of characters corresponding to irreducible representations;
these are referred to as the irreducible characters of the group. The irreducible
characters satisfy two important orthogonality relations with respect to the
inner product(·,·)
Ggiven by
(α,β)
G=
´
1
|G|

g∈G
α(g)β(g),Gfinite;
)
G
α(g)
β(g)dg, Ga compact Lie group,
whereα,β:G→Care class functions. The first is the following.
Proposition 1.12 (First orthogonality relation)The irreducible characters of
a finite group or compact Lie group G are orthonormal with respect to(·,·)
G.
In fact, one can prove that the irreducible characters form an orthonormal
basis of the space of class functions onG(in the compact Lie group case, this
should be interpreted as saying that the irreducible characters form a complete
orthonormal system inL
2
(G)). In the finite group case, this has the following
consequence.
Proposition 1.13 (Second orthogonality relation)Letχ
1,...,χ Nbe the irre-
ducible characters of the finite group G. Then
N
a
j=1
χj(g)
χj(g

)=
´
|G|
c(g)
,g∼g

;
0,otherwise,
where c(g)is the size of the conjugacy class of g and g∼g

means that g and
g

are conjugate.

1.3 Lie Group Structure and Character Theory 25
ProofLetg∈G,andlet
1[g](h)=
´
1,h∼g;
0, otherwise.
Since
1[g]is a class function and the irreducible characters are an orthonormal
basis,
1[g]can be expanded as
1[g]=
N
a
j=1
(1[g],χj)Gχj=
N
a
j=1
#
1
|G|
a
h∈G
1[g](h)
χj(h)
$
χ j=
c(g)
|G|
N
a
j=1
χj(g)χj.
Multiplying both sides by
|G|
c(g)
and evaluating atg

∈Ggives the claimed
orthogonality. e
Exercise 1.14Give a second proof by observing that the matrix
˜
c(g)
|G|
χj(g)

,
with rows indexed byj∈{1,...,N}and columns indexed by conjugacy classes
ofG, is unitary.
There are many important consequences of the orthogonality relations, too
many to go into here. It is worth mentioning, however, that the first orthogo-
nality relation implies that a representation is uniquely determined by its char-
acter. Indeed, given a representationV, we have seen that it is possible to write
V=⊕
k
i=1
V
⊕ai
i
for irreducible representationsV iand integersa i.Thenχ V=

k
i=1
aiχVi
. But since theχ Vi
are orthonormal, we have thata i=(χV,χVi
)G.
That is, the decompositionV=⊕
k
i=1
V
⊕ai
i
can be recovered fromχ V.
We now turn to the irreducible characters of the classical compact groups.
First, we will need to introduce some basic notions about integer partitions.
Given a nonnegative integerN,apartitionλofNis an ordered tupleλ=

1,...,λ k)withλ i∈Nfor eachi,λ 1≥λ2≥···≥λ k,andλ 1+···+λ k=N.
Theλ
iare called thepartsofλ. It is sometimes convenient to choosekto be
larger than what might seem to be the obvious choice by tacking 0’s onto the end
ofλ; if two partitions differ only in this final string of 0’s, they are considered
to be the same. The number of nonzero parts of a partitionλis called thelength
ofλand is denotedi(r). The integerN, i.e., the sum of the parts ofλ, is called
theweightofλand is denoted|λ|.
The following altenative notation for integer partitions is often useful. Given
k∈N, we writeλ=(1
a1,2
a2,...,k
ak)for the partition ofN=a 1+2a 2
+···+ka kwhich hasa 1parts of size 1,a 2parts of size 2, and so on. In this
notation, the partition(3, 3, 2, 1)of 9 would thus be written(1
1
,2
1
,3
2
).
Integer partitions are often represented byYoung diagrams: for a partition
λ=(λ
1,...,λ k), its Young diagram is a collection of boxes drawn from the

26 Haar Measure on the Classical Compact Matrix Groups
top-left corner
1
, withλ 1boxes in the top row,λ 2boxes in the next row, and
so on. Theconjugate partitionλ

ofλis then the one corresponding to the
reflection of the Young diagram ofλacross the diagonal. Here are the Young
diagrams of the partitionsλ=(5, 4, 1)andλ

=(3, 2, 2, 2, 1):
Now, for a multi-indexα=(α 1,...,α n), define the anti-symmetric polyno-
mial
a
α(x1,...,x n)=
σ
π∈Sn
sgn(π)x
α1
π(1)
···x
αn
π(n)
.
Note that ifσ∈S
n,then
a
α(xσ(1),...,x σ(n))=sgn(σ )a α(x1,...,x n).
In particular, this shows thata
α(x1,...,x n)=0ifx i=xjfor anyi =jand that
if anyα
i=αjwithi =j,thena α≡0.
Assume, then, thatα
1>α2>···>α n≥0; in particular,α 1≥n−1,
α
2≥n−2, and so on. Writeα=λ+δ,whereδ=(n−1,n−2,...,0)and
λis a partition of length at mostn.Then
a
α(x1,...,x n)=
σ
π∈Sn
sgn(π)
n
v
j=1
x
λj−n+j
π(j)
=det
α

x
λj+n−j
i
1≤i,j≤n
β
.
Since we have already observed thata
α(x1,...,x n)vanishes ifx i=xj,
a
α(x1,...,x n)is divisible byx i−xjfor eachi<j, and therefore by their
product, which is the Vandermonde determinant
v
1≤i<j≤n
(xi−xj)=det
α

x
n−j
i

1≤i,j≤n
β
=a
δ(x1,...,x n).
The quotient
s
λ(x1,...,x n)=
a
λ+δ(x1,...,x n)
aδ(x1,...,x n)
1
Some authors start from the bottom-right instead.

1.3 Lie Group Structure and Character Theory 27
is thus a symmetric polynomial inx
1,...,x n. The polynomials λis called the
Schur functioncorresponding toλand is defined as long asi(r≤n; by con-
vention,s
λ=0ifi(rn. The Schur functions{s λ(x1,...,x n):i(r≤n}
form one of many possible bases of the symmetric polynomials inx
1,...,x n
overZ. Moreover, we have the following.
Theorem 1.15For U∈U(n), denote the eigenvalues of U by e
iθ1,...,e
iθn.
The functions
χ
λ(U)=s λ(e
iθ1
,...,e
iθn
)
fori(r≤n are (distinct U(n).
These characters do not comprise a complete list of the irreducible characters
ofU(n), but in fact there are not too many more. For eachλ=(λ
1,...,λ n),
withλ
1≥···≥λ n,defineλ

by
λ

=(λ1−λn,λ2−λn,...,λ n−1−λn,0).
Since theλ
iare not required to be nonnegative,λmay not be a partition, butλ

always is. The functions
χ
λ(U)=det(U)
λn

(e
iθ1
,...,e
iθn
)
give the complete list of the irreducible characters ofU(n). Note from the
definition ofs
λabove that if theλin the statement ofTheorem 1.15is in fact a
partition, then the two definitions given forχ
λagree.
To give similar descriptions of (at least some of) the irreducible characters
for the other groups, one needs the following analogs of the Schur functions.
Given a partitionλwithi(r≤n,
s
(b)
λ
(x1,...,x n):=
det


,
x
λj+n−j+
1
2
i
−x

o
λj+n−j+
1
2
d
i-
1≤i,j≤n
⎞ ⎠
det
⎛⎝
,
x
n−j+
1
2
i
−x

o
n−j+
1
2
d
i-
1≤i,j≤n
⎞ ⎠
,
s
(c)
λ
(x1,...,x n):=
det
%
˜
x
λj+n−j
i
+x
−(λj+n−j)
i

1≤i,j≤n
&
det
%
˜
x
n−j
i
+x
−(n−j)
i

1≤i,j≤n
&,

28 Haar Measure on the Classical Compact Matrix Groups
and
s
(d1)
λ
(x1,...,x n)
:=
det
%
˜⊥
x
λj+n−j
i
+x
−(λj+n−j)
i
d
1λj+n−j =0 +1λj+n−j=0

1≤i,j≤n
&
det
%
˜⊥
x
n−j
i
+x
−(n−j)
i
d
1n−j =0 +1n−j=0

1≤i,j≤n
& .
Ifi(r≤n−1, let
s
(d2)
λ
(x1,...,x n−1)=
det
%
˜
x
λj+n−j+1
i
−x
−(λj+n−j+1 )
i

1≤i,j≤n
&
det
%
˜
x
n−j+1
i
−x
−(n−j+1)
i

1≤i,j≤n−1
&.
Exercise 1.16The functionss
(b)
λ
,s
(c)
λ
,ands
(d1)
λ
are polynomials inx 1,x
−1
1
,...,
x
n,x
−1
n
,ands
(d2)
λ
is a polynomial inx 1,x
−1
1
,...,x n−1,x
−1
n−1
.
In the case of the remaining classical compact groups, we will restrict atten-
tion to polynomial representations. A representationρ:G→GL(V)of a
matrix group is called apolynomial representationif there is a basis ofV
such that the entries of the matrix ofρ(g)are polynomials in the entries ofg.
(If there is one such basis, this is in fact true for all bases ofV.) The following
then gives the analog ofTheorem 1.15for the other groups.
Theorem 1.17
1. Let n∈N.ForU∈SO(2n+1), denote the eigenvalues of U by
e
±iθ1,...,e
±iθn,1. The irreducible polynomial representations ofO(2n+1)
are indexed by partitionsλsuch thatλ

1


2
≤2n+1. For such a partition
withi(rn, let˜λbe the partition defined by
˜λ

i
=
´
λ

i
, i>1;
2n+1−λ

1
,i=1.
Letχ
λdenote the character of the irreducible representation ofO(2n+1)
corresponding toλ.Then
χ
λ(U)=
´
s
(b)
λ
(e
iθ1,...,e
iθn),i(r≤n;
s
(b)
˜λ
(e
iθ1,...,e
iθn),i(rn.
For U∈SO

(2n+1),−U∈SO(n)and
χ
λ(U)=(−1)
|λ|
χλ(−U).

1.3 Lie Group Structure and Character Theory 29
2. Let n∈N.ForU∈SO(2n), denote the eigenvalues of U by e
±iθ1,...,e
±iθn;
for U∈SO

(2n), denote the eigenvalues of U by e
±iφ1,...,e
±iφn−1,1,−1.
The irreducible polynomial representations ofO(2n)are indexed by parti-
tionsλsuch thatλ

1


2
≤2n. For such a partition withi(rn, let˜λbe
the partition defined by
˜λ

i
=
´
λ

i
, i>1;
2n−λ

1
,i=1.
Letχ
λdenote the character of the irreducible representation ofO(2n)
corresponding toλ. Then for U∈SO(2n),
χ
λ(U)=
´
s
(d1)
λ
(e
iθ1,...,e
iθn),i(r≤n;
s
(d1)
˜λ
(e
iθ1,...,e
iθn),i(rn.
For U∈SO

(2n),ifi(r=n, thenχ λ(U)=0.Otherwise,
χ
λ(U)=
´
s
(d2)
λ
(e
iφ1,...,e
iφn−1),i(rn;
−s
(d2)
˜λ
(e
iφ1,...,e
iφn−1),i(rn.
3. Let n∈N.ForU∈S
p(2n), denote the eigenvalues of U by e
±iθ1,...,e
±iθn.
The irreducible polynomial representations ofS
p(2n)are indexed by par-
titionsλsuch thati(r≤n, and the value of the character corresponding
toλat U is
χ
λ(U)=s
(c)
λ
(e
iθ1
,...,e
iθn
).
Notes and References
We have given only the barest introduction to matrix analysis on the classical
compact groups; for more, the books by Horn and Johnson [53,54] and Bhatia
[11] are invaluable references.
Of the many constructions of Haar measure presented, most are part of the
folklore of the field, and it seems impossible to sort out who first wrote them
down. The parametrization by Euler angles is presumably the oldest, having
been used by Hurwitz in 1897; see [33] for a modern perspective. A survey of
methods of generating random matrices from a more computational viewpoint
isgivenin[81].
For the Lie group theory of the classical compact groups, Bump [16]givesa
beautifully written introduction; [16] also contains some of the representation
theory that appears here, and the application to the empirical spectral measure

30 Haar Measure on the Classical Compact Matrix Groups
of a random unitary matrix that appears inSection 4.2of this book. The book
by Fulton and Harris [47] gives an accessible and modern treatment of repre-
sentation theory in general, focusing on the case of Lie groups and Lie algebras.
The book [31] by Diaconis gives an introduction to the representation theory
of the symmetric group and its applications in probability that a probabilist can
understand.
The character theory on the classical groups in the form presented here is
hard to ferret out of the literature; Littlewood’s [72] and Weyl’s [105] books
(from 1940 and 1939, respectively) are the standard references, but while they
are both gems, they are becoming increasingly inaccessible to the modern
reader (especially if her area of expertise is somewhat distant). In the case
ofO(2n)andO(2n+1), Littlewood and Weyl deal only with the characters
indexed by partitions of length at mostn; the formulae presented here for those
cases can be found (written slightly differently) in section 11.9 of Littlewood.
The characters corresponding to the longer partitions are related by so-called
modification rules due to King [66]. Ram [91] has a bottom-line summary for
evalutation of characters onSO(2n),SO(2n+1),andS
p(2n).

2
Distribution of the Entries
2.1 Introduction
We begin this section with a few useful and important properties of Haar mea-
sure on the classical compact groups that follow easily from translation invari-
ance and the definitions of the groups themselves. In fact, one such property
has already appeared (inExercise 1.5): the distribution of a Haar random matrix
is invariant under transposition (and conjugate transposition). The next is an
obvious but important symmetry of Haar measure.
Lemma 2.1Let U be distributed according to Haar measure in G, where G
is one ofO(n),U(n),SO(n),SU(n), andS
p(2n). Then the entries of U are
identically distributed.
ProofTo each permutationσ∈S
n, associate the matrixM σwith entries in
{0, 1}, such thatm
ij=1 if and only ifσ(i)=j;thenM σ∈O(n). Multiplication
on the left byM
σpermutes the rows byσ
−1
and multiplication on the right by
M
σpermutes the columns byσ. One can thus move any entry of a matrixUinto,
say, the top-left corner by multiplication on the right and/or left by matrices in
O(n). By the translation invariance of Haar measure, this means that all entries
have the same distribution ifGisO(n)orU(n).
To complete the proof for the remaining groups, the permutation matrices
may need to be slightly modified. Since forσ∈S
n,det(M σ)=sgn(σ ),the
permutation matrixM
σ∈SO(n)if and only ifσis even. ForG=SO(n)
orSU(n), one therefore replaces one of the nonzero entries ofM
σwith−1
to obtain a matrix fromSO(n); choosing which entry intelligently (or rather,
failing to make the one possible non-intelligent choice) shows that any entry of
U∈SO(n)orSU(n)can be moved to the top left corner by left- and right-
multiplication by matrices within the group.
For the symplectic group, it is a straightforward exercise to show that the
permutation matrices that are inS
p(2n)are exactly those which permute the
31

32 Distribution of the Entries
indices{1,...,n}among themselves and haveσ(i+n)=σ(i)+nfor the rest.
This allows one to move any of the firstnrows to the top of a random symplectic
matrix and any of the firstncolumns all the way to the left, without changing
the distribution. For the remaining rows and columns, note that a permutation
withσ(i)∈{n+1,...,2n}for 1≤i≤nandσ(i+n)=σ(i)−ncorresponds
to a kind of anti-symplectic matrix: for suchσ,
M
σJM
T
σ
=−J.
However, changing the signs of the entries in either all of the firstncolumns (or
rows) or all of the secondncolumns (or rows) produces a symplectic matrix,
which then can be used to move rows, resp. columns, of a random symplectic
matrix to the top, resp. left. e
Exercise 2.2IfUis Haar-distributed inU(n), the distributions of Re(U
11)
and Im(U
11)are identical.
The symmetries above can be exploited to easily calculate moments of the
matrix entries, as follows.
ExampleLetUbe Haar distributed inG,forGas above.
1.E[u
11]=0: note that Haar measure onO(n)andU(n)is invariant under
multiplication on the left by





−10 0
01
.
.
.
01





; (2.1)
doing so multiplies the top row (so in particularu
11)ofUby−1 but does not
change the distribution of the entries. Sou
11
d=−u 11, and thusE[u 11]=0.
ForG=SO(n)orSU(n), just change the sign of any of the remaining ones
in the matrix in (2.1); forG=S
p(2n), change the sign of the(n+1,n+1)
entry in the matrix in (2.1).
2.E|u
11|
2
=
1
n
: becauseU∈G, we know that

n
j=1
|u1j|
2
=1, and because
all the entries have the same distribution, we can write
E|u
11|
2
=
1
n
n
a
j=1
E|u1j|
2
=
1
n
E


n
a
j=1
|u1j|
2

⎠=
1
n
.
(ForG=S
p(2n),thenshould be replaced with 2n.)

2.1 Introduction 33
Exercise 2.3ForU=

u
ij

n
j=1
, compute Cov
!
u ij,uki
"
and Cov
!
|u
ij|
2
,|uki|
2
"
for alli,j,k,i.
Understanding the asymptotic distribution of the individual entries of Haar-
distributed matrices is of course more involved than just calculating the first
couple of moments, but follows from classical results. Recall that one construc-
tion of Haar measure onO(n)involves filling the first column with a random
point on the sphere. That is, the distribution ofu
11is exactly that ofx 1,where
x=(x
1,...,x n)is a uniform random point ofS
n−1
⊆R
n
. The asymptotic
distribution of a single coordinate of a point on the sphere has been known for
more than a hundred years; the first rigorous proof is due to Borel in 1906, but
it was recognized by Maxwell and others decades earlier.
Theorem 2.4 (Borel’s lemma)Let X=(X
1,...,X n)be a uniform random
vector inS
n−1
⊆R
n
.Then
P

nX1≤t

n→∞
−−−→
1


r
t
−∞
e

x
2
2dx;
that is,

nX1converges weakly to a Gaussian random variable, as n→∞.
The lemma is also often referred to as thePoincaré limit. There are many
proofs; the one given below is by the method of moments. The following propo-
sition gives a general formula for integrating polynomials over spheres to be
used below.
Proposition 2.5Let P(x)=|x
1|
α1|x2|
α2···|x n|
αn. Then if X is uniformly
distributed on

nS
n−1
,
E

P(X)

=

1)···(β n)(
n
2
)n
o
1
2

α
i
d
(β1+···+β n)π
n/2
,
whereβ
i=
1
2
(αi+1)for1≤i≤n and
(t)=
r

0
s
t−1
e
−s
ds=2
r

0
r
2t−1
e
−r
2
dr.
The proof is essentially a reversal of the usual trick for computing the nor-
malizing constant of the Gaussian distribution.
Proof of Borel’s lemma by momentsFixm∈N; to prove the lemma, we need
to show that if for eachn,Y
nis distributed as the first coordinate of a uniform
random point onS
n−1
,then
lim
n→∞
E
!√
nYn
"
m
=E

Z
m

, (2.2)

34 Distribution of the Entries
whereZis a standard Gaussian random variable. Recall that the moments of
the standard Gaussian distribution are
E

Z
m

=
´
(m−1)(m−3)(m−5)...(1),m=2k;
0, m=2k+1.
(2.3
To prove (2.2), first note that it follows by symmetry thatE[X
2k+1
1
]=0for
allk≥0. Next, specializingProposition 2.5toP(X)=X
2k
1
gives that the even
moments ofX
1are
E

X
2k
1

=

o
k+
1
2
d

o
1
2
d
n−1

!
n
2
"
n
k

!
k+
n2
"
πn2
.
Using the functional equation(t+1)=t(t)and the fact that
o
1
2
d
=

π, this simplifies to
E

X
2k
1

=
(2k−1)(2k−3)...(1)n
k
(n+2k−2)(n+2k−4)...(n)
. (2.4)
Equation (2.2)follows immediately. e
Corollary 2.6For each n, let U
nbe a random orthogonal matrix. Then the
sequence{

n[Un]1,1}converges weakly to the standard Gaussian distribution,
as n→∞.
More recent work has made it possible to give a much more precise statement
that quantifies Borel’s lemma; doing so has important implications for the joint
distribution of the entries of a random orthogonal matrix. We first give a brief
review of various notions of distance between measures.
Metrics on Probability Measures
LetXbe a metric space. The following are some of the more widely used
metrics on the set of Borel probability measures onX.
1. Letμandνbe Borel probability measures onX.Thetotal variation
distancebetweenμandνis defined by
d
TV(μ,ν):=sup
A⊆X
|μ(A)−ν(A)|,

2.1 Introduction 35
where the supremum is over Borel measurable sets. Equivalently, one can
define
d
TV(μ,ν):=
1
2
sup
f:X→R
0
0
0
0
r
fdμ−
r
fdν
0
0
0
0
,
where the supremum is over functionsfthat are continuous, such that
f
∞≤1. Whenμandνhave densitiesfandg(respectively) with respect
to aσ-finite measureλonX,then
d
TV(X,Y)=
1
2
r
|f−g|dλ.
The total variation distance is a very strong metric on probability mea-
sures; in particular, a discrete distribution cannot be approximated by a con-
tinuous distribution in total variation.
Exercise 2.7
1. Prove that the first two definitions are equivalent, and are equivalent to
the third when the measures have density.
Hint: The Hahn-Jordan decomposition of the signed measureμ−νis
useful here.
2. Prove that the total variation distance between a discrete distribution and
a continuous distribution is always 1.
2. Thebounded Lipschitz distanceis defined by
d
BL(μ,ν):=sup
g
BL≤1
0
0
0
0
r
gdμ−
r
gdν
0
0
0
0
,
where the bounded-Lipschitz normg
BLofg:X→Ris defined by
g
BL:=max
á
g
∞,|g|L
ý
,
where|g|
L=sup
x =y
|g(x)−g(y)|
d(x,y)
is the Lipschitz constant ofg.IfXis a sepa-
rable metric space, the bounded-Lipschitz distance is a metric for the weak
topology on probability measures onX(see, e.g., [39, theorem 11.3.3]).
3. TheKolmogorov distancefor probability measures onRis defined by
d
K(μ,ν):=sup
x∈R
0
0

o
(−∞,x]
"
−ν
!
(−∞,x]
"
d0
0
0.
Convergence in Kolmogorov distance is in general stronger than weak con-
vergence, because for weak convergence, the distribution functions need
only converge to the limiting distribution function at its continuity points,
and the convergence is not required to be uniform.

36 Distribution of the Entries
4. TheL
pKantorovich distanceforp≥1 is defined by
W
p(μ,ν):=inf
π
er
d(x,y)
p
dπ(x,y)
l
1
p
,
where the infimum is over couplingsπofμandν; that is, probability
measuresπonX×Xsuch thatπ(A×R
n
)=μ(A)andπ(R
n
×B)=ν(B).
TheL
pKantorovich distance is a metric for the topology of weak conver-
gence plus convergence of moments of orderpor less. It is often called the
L
pWasserstein distance, and in the case ofp=1, the earth-mover distance.
Whenp=1, there is the following alternative formulation:
W
1(μ,ν):=sup
|f|L≤1
0
0
0
0
r
fdμ−
r
fdν
0
0
0
0
.
The fact that this is an equivalent definition ofW
1to the one given above
is the Kantorovich–Rubenstein theorem. There are dual representations for
W
pforp>1 as well, but they are more complicated and will not come up
in this book.
As a slight extension of the notation defined above, ifYandZare random
variables taking values in a metric space,d
TV(Y,Z)is defined to be the total
variation distance between the distributions ofYandZ,etc.
Quantitative Asymptotics for the Entries of Haar-Distributed Matrices
As noted above, it is a consequence of Borel’s lemma that the individual entries
of a random orthogonal matrix are approximately Gaussian for large matrices.
Borel’s lemma has been strengthened considerably, as follows.
Theorem 2.8 (Diaconis–Freedman [35])Let X be a uniform random point on

nS
n−1
,forn≥5, and let1≤k≤n−4. Then if Z is a standard Gaussian
random vector inR
k
,
d
TV
!
(X
1,...,X k),Z
"

2(k+3)
n−k−3
.
That is, not only is an individual coordinate of a random point on the sphere
close to Gaussian, but in fact the joint distribution of anykcoordinates is close
in total variation toki.i.d. Gaussian random variables, ifk=o(n).Inthe
random matrix context, this implies that fork=o(n), one can approximate any
kentries from the same row or column ofUby independent Gaussian random
variables. This led Persi Diaconis to raised the question: How many entries ofU
can be simultaneously approximated by independent normal random variables?

2.1 Introduction 37
The answer to this question of course depends on the sense of approximation.
In the strongest sense, namely in total variation, the sharp answer was found
independently by T. Jiang and Y. Ma [58] and by K. Stewart [97], following
earlier work of Diaconis–Eaton–Lauritzen [37]andJiang[59].
Theorem 2.9Let{U
n}be a sequence of random orthogonal matrices with
U
n∈O(n)for each n, and suppose that pnqn=o(n).LetU n(pn,qn)denote
the top-left p
n×qnblock of Un, and let Z(p n,qn)denote a pn×qnrandom
matrix of i.i.d. standard normal random variables. Then
lim
n→∞
dTV(

nUn(pn,qn),Z(p n,qn))=0.
That is, ap
n×qnprincipal submatrix can be approximated in total varia-
tion by a Gaussian random matrix, as long asp
nqnun; in particular, this
recovers the theorem of Diaconis and Freedman (without the explicit rate of
convergence) whenq
n=1. The theorem is sharp in the sense that ifp n∼x

n
andq
n∼y

nforx,y>0, thend TV(

nUn(pn,qn),Z(p n,qn))does not tend
to zero.
If one relaxes the sense in which entries should be simulatenously approx-
imable by i.i.d. Gaussian variables, one can approximate a larger collection of entries, as in the following theorem. Recall that a sequence of random variables
{X
n}tends to zero in probability (denotedX n
P−−−→
n→∞
0) if for alln>0,
lim
n→∞
P[|Xn|>n]=0.
Theorem 2.10 (Jiang [59])For each n, let Y
n=

yij

n
i,j=1
be an n×nmatrix
of independent standard Gaussian random variables and let
n=

γij

n
i,j=1
be
the matrix obtained from Y
nby performing the Gram–Schmidt process; i.e., n
is a random orthogonal matrix. Let
n
n(m)=max
1≤i≤n,1≤j≤m
0
0

nγij−yij
0 0
.
Then
n
n(mn)
P
−−−→
n→∞
0
if and only if m
n=o
o
n
log(n)
d
.
That is, in an “in probability” sense, as many aso
o
n
2
log(n)
d
entries ofUcan
be simultaneously approximated by independent Gaussians.
Theorems 2.9and2.10are the subject ofSection 2.3.

Other documents randomly have
different content

hiiliaineksia ja muodostui teräkseksi. Tämä karkaistiin vielä
hehkuvaksi kuumentamalla ja upottamalla kylmään veteen.
Jonkin ajan kuluttua heillä oli puukkoja, kirveitä, piiluja, sahanteriä,
talttoja, vasaroita ja nauloja.
Oli toukokuun 6. päivä, jota pohjoisella pallonpuoliskolla vastaa
marraskuun 6. Kylmä vuodenaika oli tulossa. Koska Lincolnin saari oli
vastaavassa vyöhykkeessä kuin Espanja tai Kreikka Euroopassa,
siellä oli odotettavissa kylmiä ilmoja, jopa lunta ja pakkastakin.
Piakkoin oli ainakin tuleva pitkällinen sadeaika ankarine myrskyineen.
Siirtolaisten täytyi sen takia ajatella paremman ja turvallisemman
asunnon hankkimista, ja sellaista päätettiin lähteä etsimään Grantin
järven ja meren väliseltä itärannalta. Niillä tienoin tahtoi Cyrus Smith
niin ikään saada selville, missä kohti Grantin järvi laski mereen.
Samana päivänä he lähtivät liikkeelle ensin Punapuron lähteille,
kulkeakseen sitten puron vasenta rantaa Grantin järvelle, siitä järven
koillispäähän ja sieltä pitkin sen itäistä rantaa. Matka kävi
aarniometsäin halki ilman muita seikkailuja kuin että Top kerran
metsässä seisahtui suuren, neljä, jopa viisi metriä pitkän käärmeen
eteen. Nab tappoi käärmeen yhdellä ainoalla nuijan iskulla, ja
tutkittaessa huomattiin, ettei se ollutkaan myrkyllinen, vaan kuului
niin sanottuihin timanttikäärmeisiin, joita alkuasukkaat Uudessa
Etelä-Walesissa käyttävät ruoakseen. Luultavaa kumminkin oli, ettei
saaresta puuttunut myrkyllisiäkään käärmeitä.
Järven pohjoispäässä ei huomattu mitään laskupaikkaa eikä sellaista
Cyrus Smithin suureksi kummastukseksi löydetty järven
itärannaltakaan. Siirtolaiset olivat jo saapuneet ennestään
tuntemilleen seuduille, kun Top äkkiä kävi levottomaksi. Se juoksi
rantaa edestakaisin, seisahtui kuin saalista vainutessaan, haukkui
väliin kiivaasti, väliin vaikeni äkisti.
— Vainuaisikohan Top veden alla jonkin pedon, alligaattorin
esimerkiksi?
— Tuskin, insinööri vastasi, — sillä näillä leveyksillä harvoin näkee
alligaattoreja.

Top hyppeli edelleen rannan korkeassa ruohikossa nähtävästi
seuraten jotakin vedenalaista olentoa. Mutta järven pinta oli aivan
tyyni, ei näkynyt pienintäkään virettä.
Se oli varsin kummallista.
Vähitellen saavuttiin järven kaakkoiseen päähän eikä vieläkään ollut
löydetty laskupaikkaa.
— Mutta miksi tuo järven laskupaikka teidän mielestänne on niin
tärkeä? Gideon Spilett kysyi.
— Varsinkin siksi, että jos järven aallot ovat puhkaisseet tien rannan
vuorenseinämän lävitse, niin on meidän mahdollista löytää sen
rotkoista sopiva talviasunto.
— Mutta eikö ole mahdollista, kysäisi Harbert, — että järvellä on
pohjassaan jokin laskupaikka?
— On kyllä; siinä tapauksessa täytyykin meidän itse ruveta
rakentamaan talviasuntoa.
Kello oli nyt viisi iltapäivällä, ja aiottiin juuri kääntyä kotiin päin, kun
Top taas äkkiä alkoi haukkua vimmatusti, ja ennenkuin osattiin
aavistaakaan, se oli syöksynyt veteen.
Siirtolaiset juoksivat lähemmä. Top oli viiden, kuuden metrin päässä
rannasta. Cyrus Smith rupesi huutelemaan koiraansa takaisin, mutta
samassa kohosi suunnattoman suuri pää veden pintaan.
Turpeasta turvasta ja viiksistä sekä suurista silmistä Harbert luuli
tuntevansa siinä merilehmän. Ja sen sukulaiseläin se olikin, niin
sanottu dugonki (Halicore dugong). Kita ammollaan se hyökkäsi
koiraa kohti, ja ennenkuin siirtolaiset olivat ennättäneet ryhtyä
mihinkään toimenpiteisiin, se oli siepannut koirasta kiinni ja vetänyt
sen veden alle.
Nab, rautapäinen sauva kädessä, yritti hyökätä järveen hänkin,
mutta insinööri sai hänet pidätetyksi.
Sillä välin kävi veden alla ankara taistelu, sen näki veden
kuohahtelusta. Ennakolta oli jo selvää, miten tällainen taistelu

päättyisi. Mutta äkkiä vesiryöpyn keskeltä ilmestyi Top, joka
ikäänkuin jonkin salaperäisen voiman sinkoamana lennähti kolme,
neljä metriä korkealle ilmaan ja putosi samentuneeseen veteen
takaisin. Hetken kuluttua se oli jälleen rannassa. Sanottavia haavoja
sen ruumiissa ei näkynyt.
Kerrassaan käsittämätön seikka. Ja yhä näkyi jatkuvan ottelua veden
alla. Jokin vielä suurempi eläin oli kaiketi käynyt dugongin kimppuun,
niin että koira oli päässyt irti.
Mutta taistelua ei kestänyt kauan. Vesi alkoi punertua verestä, eikä
aikaakaan, niin dugongin ruumis nousi pintaan, leveä purppurajuova
perässään, ja tarttui viimein matalikolle lähelle rantaa.
Siirtolaiset kiirehtivät sen luokse. Dugonki oli kuollut, kaulassaan
suuri haava, ilmeisesti jollain teräaseella viilletty.
Millainen mahtoikaan olla se peto, joka oli iskenyt siihen niin
ammottavan haavan? Sitä ei osannut kukaan sanoa.
Siirtolaiset palasivat vihdoin hormeille keskustellen matkalla
vilkkaasti tästä kerrassaan käsittämättömästä tapauksesta.

Kahdestoista luku
Dugongin haava. Järven laskukohta.
Räjähdysaine. Uusi vesiputous. Tunnelissa.
Syvä kuilu. Luolasaleja.
Seuraavana päivänä, toukokuun 7:ntenä, Cyrus Smith ja Gideon
Spilett lähtivät jälleen Grantin järvelle. Toiset menivät lauttaamaan
polttopuita hormeille.
Sillä matalikolla, missä dugongin raato virui, oli varsin vähän vettä,
mutta siitä näytti järven pohja äkkijyrkkänä laskevan keskijärveä
kohti. Lintuja oli suurissa parvin keräytynyt haaskan ympärille. Ne
piti karkottaa kivillä pois, sillä Cyrus Smith aikoi käyttää hyväkseen
dugongin rasvaa.
Siinä he seisoivat kumpainenkin kummastellen eilisillan tapausta.
Mitä enemmän he katselivat dugongin leuan alla olevaa haavaa, sitä
oudommalta heistä asia näytti. Mikä väkevä käsi lieneekään tehnyt
tämän ilmeisen puukonhaavan ja viskannut Topin järvestä ilmaan?...
Ja kuinka oli Cyrus Smith pallosta mereen pudottuaan ja veteen
vajottuaan voinut pelastua ilman vähintäkään vammaa
rantakallioiden toiselle puolelle?
Tässä piili merkillinen salaisuus, johon he toivottavasti
tulevaisuudessa saisivat selityksen. Toistaiseksi he päättivät olla
puhumatta siitä mitään muille tovereilleen.
Heidän siinä näitä asioita miettiessään ja järven rantaa kävellessään
Cyrus Smith huomasi äkkiä järven pinnassa nopean virtapaikan. Hän
heitti siihen muutamia puunpalasia, ja ne lähtivät heti kiitämään
virran mukana. Ystävykset seurasivat niitä, kunnes saapuivat järven
eteläisimpään päähän. Siinä vesi katosi ikäänkuin rakoon maan

sisään. Insinööri painoi korvansa maahan ja kuuli selvästi veden
solinaa.
— Tässä varmaankin järven vedet laskevat johonkin rantakallioiden
halkeamaan.
Hän katkaisi ja karsi pitkän oksan ja pisti sen rantaviivan
muodostamaan kulmaukseen noin puolen metrin verran
vedenpinnan alle. Keppi kirposi heti hänen kädestään ja katosi
vedenalaiseen koloon.
— Nyt ei enää epäilystäkään! Cyrus Smith huudahti. — Tässä on
järven vesillä laskupaikkansa, ja minä saan sen näkyviin.
— Millä keinoin?
— Lasken järven pinnan puolitoista metriä alemmaksi.
— Mutta millä tavalla?
— Aukaisen isomman väylän.
— Tuohon graniittiseinäänkö?
— Siihen juuri. Minä räjäytän sen, vesi alenee, ja tunneli tulee
näkyviin. Samalla syntyy vesiputous, josta meille vielä saattaa olla
paljonkin hyötyä.
He palasivat hormeihin, ja siellä Cyrus Smith ilmoitti ystävilleen
aikeensa.
Hänen mielestään täytyi siinä osassa Näkötornin ylänköä, joka
muodosti graniittiseinän järven ja meren välille, olla suurempia tai
pienempiä rotkoja ja luolia, ja ne hän oli päättänyt saada esiin. Sitä
varten piti ensinnäkin raivata suurempi aukko järven vesille ja laskea
siten järven pinta alemmaksi. Siihen taas tarvittiin räjähdysaineita. Ja
nämä olisi valmistettava tarjona olevista runsaista
luonnonrikkauksista.
Rikkikiisua oli saarella runsaasti. Siitä tehtäisiin rikkihappoa.
Salpietaria oli niinikään Franklinin vuoren juurella. Levistä ja
merikasveista poltettaisiin niin sanottua luonnollista soodaa. Soodan

avulla saataisiin taas hylkeenrasvasta saippuaa ja glyseriiniä.
Salpietarista ja rikkihaposta valmistuu typpihappoa, siitä ja
glyseriinistä nitroglyseriiniä.
— Selvä kuin päivä! huudahti Pencroff.
Niin aivan selvää se ei silti ollut. Varsinkin oli rikkihapon
valmistaminen vaikeaa ja monimutkaista, se kun tehtaissa vaatii
erittäin suuria ja kalliita laitteita. Mutta Cyrus Smith osasi käyttää
hyväkseen sitä valmistustapaa, jonka antamaa tuotetta nimitetään
Nordhausenin hapoksi. Rikkikiisu sai hienoksi jauhettuna rapautua
tasaisella kalliolla, ja näin syntynyt rautasulfaatti pasutettiin
savisessa tislausastiassa.
Työskenneltyään lopulta jonkin aikaa syrjässä ihan itsekseen — työ
kun oli varsin vaarallista — sai Cyrus Smith vihdoinkin aikaan
astiallisen öljymäistä, kellertävää nestettä. Hyvillään hän näytti sen
ystävilleen.
— Tämä on nitroglyseriiniä! hän lausui.
Siinä oli tosiaan tuota hirveää ainetta, jonka räjähdysvoima on
laskettava kymmenen kertaa niin suureksi kuin ruudin. Sen
vaarallisuus on onneksi paljon vähentynyt sen jälkeen, kun on opittu
valmistamaan sitä kiinteänä, jolloin sen nimi on dynamiitti.
— Ja tuoko mokoma litku se nyt kalliot räjähdyttää? merimies
kysäisi.
— Tämä, ja sitä suuremmalla voimalla, mitä lujempaa graniitti on.
— Milloin se tapahtuu?
— Huomenna, kun ensin olemme saaneet poratuksi reiän
graniittiseinään.
Seuraavana päivänä siirtolaiset lähtivät siihen kohtaan Grantin
järveä, jossa sen itäinen lahdeke oli ainoastaan viidensadan askelen
päässä meren rannasta. Itse ylänkö oli tässä kohden alempana
järven pintaa, mutta ylängön tasainen laki esti vettä laskemasta
mereen. Tämä harja oli raivattava pois.

Ensin porattiin reikä harjaan, yläreunasta lähtien vinoon suuntaan,
niin että sen toinen pää tuli olemaan alempana järven pintaa. Aukko
oli tehtävä melko suureksi, sillä insinööri oli päättänyt käyttää
räjähdyttämiseen vähintään kymmenen litraa nitroglyseriiniä.
Mitä panoksen sytyttämiseen tulee, päätti Cyrus Smith käyttää
hyväkseen sitä nitroglyseriinin ominaisuutta, että se räjähtää
tärähdyksestä. Sitä varten hän asetti porausreiän kohdalle kolmijalan
ja ripusti siihen kasvinkuiduista punottuun köyteen usean kilon
painoisen rautalohkareen. Ennakolta rikillä voideltu lanka johti
ensinmainitun keskikohdalta muutaman metrin verran ulkopuolelle
aukkoa maata pitkin. Kun langan irtonainen pää sytytettäisiin, palaisi
lanka verkalleen, kunnes tuli ehtisi painoa kannattavaan köyteen.
Tämä syttyisi ja katkeaisi, ja silloin putoaisi rautalohkare
nitroglyseriinin päälle ja saisi sen räjähtämään.
Lähetettyään ystävänsä jo ennakolta hormeille Cyrus Smith nosti
irtonaisen langanpään, sytytti sen ja lähti juoksujalkaa muitten luo.
Köyden oli määrä palaa viisikolmatta minuuttia, ja aivan oikein:
viidenkolmatta minuutin kuluttua kuului räjähdys, valtavampi kuin he
olivat osanneet aavistaakaan. Tuntui siltä kuin koko saari olisi
vavahtanut perustuksiaan myöten. Suunnaton joukko kiviä sinkoutui
ilmaan kuin tulivuoresta. Hormien kalliotkin vavahtivat. Ilmanpaine
oli niin ankara, että siirtolaiset, vaikka olivat vähää vaille neljän
kilometrin päässä, eivät kestäneet jaloillaan, vaan kaatuivat kumoon
joka mies.
Heti he kuitenkin nousivat ja kiiruhtivat paikalle.
Heiltä pääsi kolminkertainen eläköönhuuto. Ylängön harja oli
murtunut. Hurjasti syöksyi vesi järvestä ylängön poikki ja putosi
yhdeksänkymmenen metrin korkeudesta meren rantaan!
Parin tunnin kuluessa järvi laski vähintään puoli metriä. Veden
entinen laskupaikka vuoren sisällä oli nyt tarkemmin tutkittava, ja
sitä varten haettiin hormeista kuokkia, rautakankia, kasvin-kuiduista
punottuja köysiä ja tulukset.

Pencroff oli ylen ihastunut insinöörin suurenmoisesta, onnistuneesta
yrityksestä.
— Mr Smith, hän virkkoi, — tällä teidän rasvaisella liemellännehän
voisi tärskäyttää ilmaan vaikka koko saaren, vai?
— Vaikka koko maapallon, kun vain käyttää tarpeeksi nitroglyseriiniä.
— Sepä se. Mutta eikö sopisi käyttää sitä ruudin asemasta
pyssyissä?
— Sellaisenaan se olisi liian voimakasta: ei lujinkaan pyssynpiippu
sitä kestäisi. Sen sijaan sopisi meidän valmistaa siitä niin sanottua
pumpuliruutia, mutta sekin olisi turhaa, koska meillä ei ole pyssyjä.
— Niin, niin, mutta jos te vain tahtoisitte, niin pianhan ne pyssyt...
Mahdotonta — se oli sana, jonka Pencroff oli kokonaan pyyhkäissyt
pois Lincolnin saaren sanakirjasta.
Vuoren graniittiseinässä, ylempänä vedenpintaa, näkyi nyt selvästi
tuo kauan etsitty aukko, noin kuusi metriä leveä, mutta ainoastaan
puoli metriä korkea. Kuokilla hakattiin tämä portti avarammaksi, niin
että oli helppoa kulkea ainakin kumarassa tunnelia myöten. Lattian
kaltevuuskaan tuossa käytävässä ei ollut sen suurempi kuin että
siinä saattoi helposti liikkua.
Käsissään pihkaisia oksia tulisoihtuina siirtolaiset lähtivät käytävään,
Top muitten edellä. Vastoin odotuksia aukko alkoi käydä yhä
korkeammaksi, niin että vaeltajat pian mahtuivat kulkemaan aivan
suorina. Pohja oli veden jäljeltä vielä märkä, niin että täytyi koko
ajan varoa liukastumista. Kaikki osoitti, että tämä käytävä ei ollut
virran luoma, vaan oli aikain alussa muodostunut tähän eli toisin
sanoen oli yhtä vanha kuin saari itsekin.
Satasen askelta laskeuduttuaan kulkijat huomasivat käytävän äkkiä
laajenevan jonkinlaiseksi väljäksi luolaksi. Siihen pysähtymättä he
kulkivat vielä noin viisikymmentä askelta alemmaksi. Silloin alkoi
edestä päin äkkiä kuulua ääniä, kovia ja kaikuvia kuin kaiuttimesta.
— Se on Top! huusi Harbert.

— Ja haukkuukin niin vimmatusti, lisäsi Pencroff.
— Haukkukoon mitä tahansa, Gideon Spilett virkkoi, — onhan meillä
aseet kädessä. Eteenpäin vain!
Siirtolaiset kiirehtivät yhä alemmaksi, ja yhä rajummaksi kävi koiran
haukunta. Olisiko se ottelussa jonkin eläimen kanssa? Miehet
melkein laskivat liukua luolan kaltevaa pohjaa myöten ja saavuttivat
tuokin kuluttua koiransa pariakymmentä metriä alempana.
Siellä oli tunneli laajentunut suureksi, komeaksi luolaksi. Top
juoksenteli edestakaisin vimmatusti haukkuen. Pencroff ja Nab
heiluttivat tulisoihtujaan paremmin valaistakseen luolaa. Siirtolaiset
samosivat sitä ristiin rastiin, mutta kaikkialla oli vastassa vain tumma
graniittiseinä.
— Onhan tässä täytynyt olla aukko, josta vesi on virrannut mereen,
lausui insinööri.
— Ollaanpa varovaisia, Pencroff arveli, — muutoin tässä vielä
syöstään suinpäin kuoppaan.
— Etsi, Top, etsi, etsi!
Ja entistä kiivaammin hyökkäsi Top eteenpäin rajusti haukkuen.
Vihdoin saavuttiin syvän kaivon partaalle. Koetettiin tulisoihduilla
valaista sen seinämiä, mutta turhaan. Tämä ei enää ollut luisua
käytävää, vaan kohtisuoraan alas johtava kuilu. Cyrus Smith tempasi
palavan oksan ja heitti sen syvyyteen. Pudotessaan se paloi entistä
kirkkaammin ja sammui sitten sähisten: se oli pudonnut meren
pintaan. Putoamisen ajasta insinööri laski kuilun syvyyden noin
kolmeksikymmeneksi metriksi.
Luolan lattia oli niin ollen kolmekymmentä metriä ylempänä
merenpintaa.
— Kas tässä meillä on maja tästä lähtien, insinööri virkkoi.
— Mutta tässä on varmaan ennen meitä asunut joku olento.

— Kaiketi, ja nyt se on paennut tuon kuilun kautta jättäen asuntonsa
meille.
— Pahus etten minä ollut Top! Olisin saanut nähdä tuon
kummituksen! huudahti Pencroff.
Oli miten oli, siirtolaiset olivat nyt toiveittensa perillä: tästähän he
saisivat tilavan ja mukavan asunnon. Kaksi hankaluutta vielä oli
jäljellä. Ensinnäkin luolaan oli saatava valoa; toiseksi siihen oli
hankittava sopivat portaat. Käytävän kaltevuudesta ja pituudesta
insinööri päätteli, ettei luolan perimmäinen, merenpuoleinen seinä
ollut kovinkaan paksu. Jos siihen saisi puhkaistuksi ikkunareiän,
syntyisi samalla ovikin, johon ei olisi vaikeata kiinnittää tikapuita.
Tuskin oli Cyrus Smith ilmaissut ajatuksensa ystävilleen, kun jo
Pencroff ryhtyi kuokallaan hakkaamaan seinää. Hänen väsyttyään
jatkoi työtä Gideon Spilett. Kului pari tuntia. Äkkiä kuokka kirposi
reportterin kädestä ja putosi puhjenneesta aukosta ulos.
— Eläköön! huusi Pencroff. — Onni suosii miehiä!
Seinä ei ollut täyttä metriäkään paksu.
Cyrus Smith katsahti aukosta ulos: hänen silmäinsä eteen aukeni
merenranta, Pelastuksenluoto ja sen toisella puolella valtameren
siintävä pinta. Sisään hulvahtanut valo kirkasti tämän suunnattoman
luolan sisustan. Vasen puoli oli lähes kolmekymmentä metriä pitkä,
kymmenen metriä leveä ja yhtä korkea, mutta oikeanpuoleisessa
sivustassa katto kohosi viisikolmatta metriä korkealle. Siellä täällä oli
graniittipylväitä, jotka kannattivat tuota jättiläismäistä holvia,
niinkuin suurissa tuomiokirkoissa pylväät kannattavat keskikupoolia.
Ja niitten välillä kierteli oikuttelevan monimutkaisina mitä erilaisimpia
kaaria. Luonto, suuri rakentaja, oli yhdistänyt tässä kaikki
bysanttilaisen, romaanisen ja goottilaisen rakennustyylin eri muodot.
Ihmetellen siirtolaiset katselivat tätä salaperäistä uutta, uljasta
Alhambraa. He olivat toivoneet löytävänsä ahtaan luolan, ja heidän
eteensä avautui komea linna. Nab oikein otti lakin päästään
hartauden vallassa.

— Mikä nimeksi tälle uudelle asunnolle? Harbert kysyi.
— Graniittilinna, insinööri vastasi.
Mutta tulisoihdut olivat sammumaisillaan. Ystävykset riensivät
takaisin ja saapuivat järven rantaan juuri, kun viimeinen niistä
sammui.

Kolmastoista luku
Useita huoneita. Kaniinitarha. Ovi ja ikkunoita.
Seuraavana päivänä siirtolaiset alkoivat varustella uutta kotiaan
asuttavaan kuntoon.
Ensinnäkin oli Cyrus Smithin ehdotuksesta hakattava
merenpuoleiseen päätyyn viisi ikkunaa ja ovi. Viimeksimainitun
tarpeellisuutta ei Pencroff voinut käsittää, koska järven puolelta oli
aivan hyvä käytävä Graniittilinnaan.
— Hyvä ystävä, insinööri sanoi, — jos meidän on helppoa tulla
sisään järven puolelta, niin yhtä helppoa on mereltäkin päin. Entisen
aukon aion päinvastoin kokonaan sulkea.
— Millä tavalla sitä sitten sisään päästään? Pencroff kyseli.
— Ulkopuolisia tikkaita, köysitikkaita myöten, jotka saa vetää sisään,
milloin tahtoo. Silloin ei pääse sisään ketään muita.
— Mutta miksi niin paljon varokeinoja? merimies intti. — Eiväthän
meitä tähän asti ole pedot ahdistaneet, eikä saarellamme ole
villejäkään.
— Oletteko siitä ihan varma?
— Se on pian selville saatu, kunhan joka kolkka saarta saadaan
tutkituksi.
— Aivan niin, sillä me tunnemme siitä vain vähäisen osan. Mutta ellei
meidän tarvitsekaan pelätä vihollisia omalla saarellamme, niin
saattaa niitä tulla ulkoapäin, sillä Tyynen meren rannikot ovat
vaarallisia seutuja.
Ja niinpä he ryhtyivät työhön, innokkaammin muita juuri Pencroff.
Rannalta kannettiin muutama tuhat valmista tiiltä Graniittilinnan

seinukselle. Kuokilla ja nitroglyseriinillä puhkaistiin mainitut viisi
ikkunaa ja ovi meren puolelle. Tiilistä muurattiin väliseinät, niin että
saatiin viisi erillistä asuinhuonetta, nimittäin ovesta oikealle eteinen,
sen viereen keittiö, edelleen ruokasali, kaksitoista metriä leveä,
sitten yhtä tilava makuuhuone, sen jälkeen seurusteluhuone ja
vihdoin suuri sali. Vastapäätä näitä perätysten olevia huoneita oli
toisella puolen käytävää suuri varastohuone.
Koska pääsy Graniittilinnaan entisen tunnelin kautta oli kovin
hankala, sinne kun oli kuljettava pitkiä ja vaivalloisia kiertoteitä ja
itse tunnelikin oli kovin vaikeakulkuinen, punottiin erään kasvin
lujista kuiduista tukevat köysitikkaat, joiden puolat veistettiin
punaisen seetripuun oksista. Koska ovi oli niin korkealla maasta,
täytyi tikkaisiin panna vähintään sata puolaa. Nousua helpotti se,
että vuoren seinämässä oli ulkoneva pengermä, niin että tikkaat
saattoi jakaa kahteen osaan. Tarpeen tullen saattoi kuitenkin tikkaat
kokonaisuudessaan vetää oven sisäpuolelle.
Toukokuun 28. päivänä olivat tikkaat lopullisesti valmiit. Lyhyessä
ajassa oppivat siirtolaiset varsin nopeasti kiipeämään niitä myöten
ylös. Hitain oli Top, luonnollisestikin, mutta oppimestarinsa Pencroffin
johdolla sekin tottui auttavasti kiipeämään, vaikka tosin opettaja
toisenkin kerran sai kantaa retuuttaa oppilaansa selässään ylös.
Talvi läheni nopein askelin, ja siksipä Harbertilla ja Gideon Spilettillä
oli täysi työ heidän hankkiessaan metsänriistaa varastoon.
Toistaiseksi he yhä metsästelivät Laupeudenjoen vasemmalla
rannalla, mutta siellä olikin vielä saalista riittämiin. Vasta seuraavana
keväänä oli aikomus siirtyä Kaukaisen Lännen suunnattomiin
aarniometsiin, jotka alkoivat joen toiselta rannalta. Näillä retkillä
sattui Harbert löytämään läheltä Grantin järven lounaista rantaa villin
puutarhan, jossa kasvoi monenlaisia hyvänhajuisia pensaita ja
kukkia, kuten minttuja, kuningasyrttejä, kynteliä, ja muita sellaisia,
joita kaniinit mielellään syövät.
Koska pöytä oli valmiiksi katettu, arveli reportteri, ei isäntäväkikään
voinut olla kovin kaukana.

Maassa näkyi lukemattomia kaniinien pesiä. Ja tuskin olivat
metsämiehet kulkeneet kymmentäkään askelta tarhassa, kun jo
satakunta jyrsijää, ns. amerikankaniineja, puikahti ulos pesistään
lähtien kiitämään pakoon. Pari kolme ehtivät metsästäjät kuitenkin
saada saaliikseen.
Toukokuun viimeisenä päivänä oli muuraustyö Graniittilinnassa saatu
loppuun suoritetuksi ja keittiöön oli kyhätty yksinkertainen liesi, josta
savutorvi johti suureen päätyseinässä olevaan aukkoon.
Ei unohtanut Cyrus Smith vedensaantiakaan. Entiseen järven
laskukohtaan hakattiin aukko hiukan vedenpinnan alapuolelle. Siitä
vesi virtasi entiseen väylään kaivettua kapeaa uomaa myöten kuiluun
asti tuoden Graniittilinnan uusille asukkaille runsaasti sata litraa vettä
päivässä.
Tunnelin entinen suu tukittiin suurilla kallionlohkareilla, jotka
muurattiin kiinni toisiinsa. Ne peitettiin mullalla ja turpeilla. Ikkunat
suljettiin tiiviillä luukuilla siihen saakka, kunnes Cyrus Smith ryhtyi
lasin tekoon.
Ihana näköala avautui Graniittilinnan ikkunoista valtamerelle, joka
kaukaisuudessa sulautui taivaanrantaan, rajoittui vasemmalla
Eteläleukaan ja oikealla Kynsiniemeen. Koko Liittolahti oli heidän
ihailtavanaan.
Pencroffin mielestä oli "niin kovin hauskaa elellä täällä viidennessä
kerroksessa".

Neljästoista luku
Sydäntalvi. Steariinikynttilöitä. Leivän puute.
Vehnänjyvä. Maan viljavuus.
Talvi alkoi kesäkuun ensimmäisinä päivinä; kesäkuuta vastaa
pohjoisella pallonpuoliskolla joulukuu. Sen enteinä olivat kovat
myrskyt ja rankkasateet. Siirtolaiset olivat muuttaneet
Graniittilinnaan päätettyään säilyttää entiset hormit työpajana
myöhemmin tehtäviä karkeita töitä varten.
Koko kesäkuu kului metsästellessä ja kalastellessa. Kylmän tultua —
niin päätettiin — lähdettäisiin pyytämään niitä mufloneja, joita kerran
oli nähty Franklinin vuorelle noustaessa ja joitten villasta siirtolaiset
insinöörinsä taitoon luottaen toivoivat pystyvänsä valmistamaan
lämpimiä vaatteita.
Tuli kesäkuun 4. päivä. Se oli helluntai. Sen päivän he päättivät pitää
pyhänä, niinkuin olivat pitäneet pitkänperjantain ja pääsiäisenkin.
Insinöörin ja reportterin kerran keskustellessa pian alkavista
pakkasista puuttui Pencroffkin puheeseen.
— Minun mielestäni, hän virkkoi, — on aika puhua pakkasista silloin,
kun pakkaset tulevat. Eiköhän sopisi keskustella siitä, mistä tässä
valoa saadaan, sillä päivät ovat käyneet merkillisen lyhyiksi?
— Se saadaan helposti aikaan, Cyrus Smith vastasi..
— Keskusteluko?
— Ei, vaan valo.
— Öljylamppujako?
— Steariinikynttilöitä.

Tämä ei ollutkaan mahdotonta, sillä insinöörillähän oli
käytettävänään sekä kalkkia että rikkihappoa, ja hylkeistä sai kyllä
rasvaa.
Rasvasta saa kalkin avulla eräänlaista kalkkisaippuaa, joka helposti
liukenee rikkihappoon. Tästä liuoksesta kalkki saostuu rikkihapon
suolana ja samalla erottuu rasvahappoja, joista palmitiini- ja
steariinihappo ovat juuri kynttiläin aineksia. Sydämiksi oli käytettävä
yhteenpunottuja kasvikuituja: tällainen sydän kastetaan sulaan
rasvaan, joka pian hyytyy sen ympärille ohuena kerroksena; sen
jälkeen se kastetaan uudestaan, jolloin uusi rasvakerros tarttuu
entiseen; näin jatketaan, kunnes kynttilä on tullut sopivan
paksuiseksi.
Kynttilänvalmistus onnistuikin hyvin, kun Pelastuksenluodolta oli
saatu muutamia hylkeitä, sillä erää puoli tusinaa.
Rattoisaa oli siirtolaisten työskennellä Graniittilinnassaan, milloin
rauta-, milloin puuseppinä, valmistellen tarpeellisia huone- ja
talouskaluja. Kirvesmiehiksikin heidän oli pystyttävä, sillä
vesiputouksen yli oli rakennettava kahdet porraspuut, toiset
ylängölle, toiset vuorenseinämän ja merenrannan välille.
Laupeudenjoen varrelta kerättiin suunnattomat määrät ostereita,
joitten makuun siirtolaiset olivat jo tottuneet.
Lihaa heillä oli runsaasti; ei liioin kasvisruoista ollut puutetta. Niin
sanottujen lohikäärmepuitten keitetyistä ja käyneistä juurista he
osasivat panna hapahkoa, oluentapaista juomaa. Sokeriakaan ei
heiltä puuttunut: sitä he saivat sokerivaahteran (Acer saccharinum)
mahlasta. Maukasta, niin sanottua oswego-teetä, he osasivat keittää
niistä monista kasveista, joita runsaasti kasvoi kaniinitarhan
seuduilla. Suolaakin heillä oli yllinkyllin.
Vain leipää puuttui.
Sallimus oli kuitenkin pitänyt heistä huolta tässäkin suhteessa
tavalla, jota he eivät olleet osanneet aavistaakaan.

Eräänä päivänä satoi kuin saavista. Siirtolaiset olivat tapansa
mukaan koolla Graniittilinnan suuressa salissa. Harbert näkyi
hypistelevän jotakin takkinsa liepeessä, vuoren ja päällisen välissä.
Äkkiä hän löysi etsimänsä ja huudahti riemuissaan:
— Mr Smith! Katsokaas, mitä minä löysin! Vehnänjyvän!
Asian selitys oli, että Harbertilla Richmondissa oli ollut tapana pitää
taskussaan vehnänjyviä ja ruokkia niillä kyyhkysiä.
Kaikki kokoontuivat ihmettelemään pikkuista jyvää.
— Voi lapsi raukka! Pencroff puheli. — Jopa nyt jotakin löysit! Yhden
ainoan jyväpahasen!
— Siitä me vielä paistamme leipää, lausui Cyrus Smith.
— Ja piiraita ja kaakkuja ja kalakukkoja, ivaili merimies.
Mutta insinööri loi häneen vakavan katseen ja kysäisi:
— Pencroff, tiedättekö, kuinka monta kortta nousee yhdestä jyvästä?
— Jyvästä korsi, kahdesta kaksi, toinen hämillään vastasi.
— Kymmenen. Kuinka monta jyvää korressa?
— En minä niitä ole laskenut.
— Keskimäärin kahdeksankymmentä, julisti insinööri juhlallisesti. —
Jos siis istutamme tämän, niin saamme ensi sadoksi 800 jyvää;
toisella kertaa niistä tulee jo 640 tuhatta, kolmannella 512 miljoonaa
ja neljännellä yli neljäsataa miljardia.
Kummastellen kuuntelivat siirtolaiset näitä suunnattomia lukuja,
mutta insinööri jatkoi:
— Eikä tämä ole vielä mitään sen rinnalla, mitä unikko ja tupakka
tuottavat. Yhdestä unikon siemenestä tulee 32 tuhatta siementä,
tupakan siemenestä 360 tuhatta. Muutamassa vuodessa nämä kasvit
täyttäisivät koko maan pinnan, ellei niiden leviämistä olisi
ehkäisemässä tuhansia syitä.

— Ja tiedättekö, Pencroff, Cyrus Smith yhä jatkoi, — kuinka monta
bushelia [busheli = 36 litraa] tulee 400 miljardista jyvästä?
— Kaikkea sitä minulta kysytäänkin! merimies jupisi. — En minä ole
opissa ollut enkä kouluissa kolunnut.
— Neljänneksen toista miljoonaa bushelia.
— Soo! Neljässä vuodessa?
— Kenties kahdessakin, sillä täällä korjataan luullakseni kaksi satoa
vuodessa.
Pencroffin tuli äkkiä niin hyvä mieli, että häneltä pääsi railakas
eläköönhuuto.
— Ja sen minä sanon, hän lisäsi, — että jos minä joskus vielä
tupakan siemenen löydän, niin en minä sitä järveen heitä.
Juhlallisesti kylvettiin sitten tämä kallis jyvä Graniittilinnan ylängölle,
huolellisesti muokattuun, muheaan multaan, tuulensuojaiseen
päivänpuoleiseen paikkaan, ja tiheä aitaus pystytettiin sen ympärille.

Viidestoista luku
Pakkasia. Haukkuvia kettuja. Talvitöitä.
Salaperäinen kuilu. Pekari. Herkullinen ateria.
Hauli hampaisiin.
Kesäkuun lopulla sadeaika lakkasi, ilma kylmeni ja lämpömittari laski
alle nollan. Polttopuut hupenivat silmissä, ja uusia oli hankittava.
Helposti niitä lauttaamalla saatiinkin, ja helposti ne hinattiin
Graniittilinnaan, johon lujista köysistä oli kyhätty varsin kätevä hissi.
Tuotiinpa Franklinin vuoren liepeiltä kivihiiltäkin, ja kun myös
ruokasaliin oli rakennettu tulisija, ei siirtolaisten tarvinnut kärsiä
kylmyyttä.
Tehtiin lyhyitä metsästysretkiä. Kerrankin päästiin Laupeudenjoen yli,
se kun oli vetäytynyt jäähän, ja tavattiin siellä omituisia kettuja,
jotka haukkuivat kuin koirat. Pencroff ei paljonkaan välittänyt niistä,
ne kun eivät olleet syötävää riistaa. Siitä huolimatta hän kuitenkin
rakenteli niille ketunkuoppia, ja silloin tällöin joku vikkeläjalkainen
repolainen erehtyi ja jäi merimiehen saaliiksi. Joka kerta hän
kuitenkin nurisi "mokomastakin riistasta". Hiukan häntä lohdutti
Cyrus Smithin neuvo, että hän rupeaisi käyttämään ketunraatoja
muitten otuksien syötteinä.
Ja kyllä oli merimies mielissään saatuaan kerran kuopasta erään
siansukuisen eläimen nimeltä pekari: sehän kelpasi ruoaksi saparosta
kärsään saakka.
Elokuun puolivälissä alkoi sataa lunta, ja pian sitä oli puolen metrin
paksuudelta. Samalla alkoivat rajut lumimyrskyt. Graniittilinnan
luukut suljettiin, ja siellä siirtolaiset miellyttävässä lämpimässä
kuuntelivat kaikessa rauhassa myrskyn ulvontaa, kun sen rajuista
puuskista suuret puut ryskyen kaatuivat maahan.

Elokuun lopulla alkoi sää vaihdella: satoi vuoroin vettä, vuoroin
lunta, mutta ilma pysyi yhä kolkkona, joten Graniittilinnan ikkuna- ja
oviluukut pidettiin edelleen suljettuina. Miehet alkoivat kyllästyä
talven yksitoikkoisuuteen ja pitkälliseen sisälläoloon, mutta vielä
rauhattomampi oli Top. Lakkaamatta se kuljeskeli huoneesta toiseen
ikäänkuin hakien aukkoa, mistä pääsisi ulos.
Cyrus Smith huomasi koiransa kovin usein kiertelevän pystysuoran
kuilun aukkoa, joka nyt oli laudoilla peitetty. Kummallisesti se
haukkua louskutteli siinä, ikäänkuin kuilun pohjalla olisi ollut jotain,
joka sitä ärsytti ja jota se samalla pelkäsi. Kuilu oli yhteydessä meren
kanssa, siitä ei ollut epäilystäkään. Tulikohan mereltä jokin
vedenalainen hirviö väliin levähtämään kuilun kohdalle?
Jotain kummallista siinä oli.
Lämpimäin talvivaatteiden puute sai heidät usein suunnittelemaan
villavarastojen hankkimista seuraavaksi talveksi. Franklinin vuoren
rinteillä oli laumoittain muflonlampaita. Sen he tiesivät. Parasta olisi
saada muutamia niistä kiinni elävinä ja kesyttää ne. Samalla voisi
pyydystää metsälintuja jonkinlaiseksi kanatarhaksi. Mutta sitä varten
pitäisi perustaa karjakartano, oikea farmi. Ja edullisinta olisi sijoittaa
sellainen Kaukaiseen Länteen.
Ensi kevätkelillä käytäisiin tarkastamassa noita tuntemattomia
seutuja.
Mutta jo ennen sitä sattui tapaus, joka sai heidät entistä
kiihkeämmin tutkimaan Lincolnin saarta.
Oli lokakuun 24:s. Sinä päivänä oli Pencroff saanut ketunkuopasta
emäpekarin ja kaksi parin kuukauden ikäistä porsasta.
Silloin oli siirtolaisilla loistopäivällinen: kengurunlihalientä,
savustettua kinkkua, porsaanpaistia, manteleita, juuriolutta ja
oswego-teetä. Kaiken kruununa oli tietysti pekarinpaisti.
Liemen jälkeen kävi Pencroff itse leikkaamaan pekaria ja viilteli aimo
viipaleita kunkin lautaselle. Nämä juottoporsaat maistuivat todellakin

aivan erinomaisilta. Kummako, että Pencroffkin ahnaasti kävi oman
annoksensa kimppuun.
Äkkiä hän parkaisi ja päästi suustaan karkean sanan.
— Mikä nyt? Cyrus Smith kysyi.
— Se ... se vain ... että minulta lohkesi hammas! merimies vastasi.
— Vai oli teidän pekarissanne kiviäkin? naurahti Gideon Spilett.
— Lienee ollut, lienee ollut ... Pencroff puheli vetäen suustaan kiven,
joka oli maksanut hänelle poskihampaan.
Mutta se ei ollutkaan kivi. Se oli — hauli!
TOINEN OSA

HYLÄTTY MIES

Ensimmäinen luku
Haulin johdosta. Pirogin tekoon. Kilpikonnan
katoaminen. Kömpelyyttäkö?
Seitsemän kuukautta olivat siirtolaiset jo olleet Lincolnin saaressa, ja
yhä enemmän he olivat varmistuneet siinä uskossa, että he olivat
saaren ainoat asukkaat. Mutta nyt tämä pikkuinen lyijypalanen
ilmeisesti todisti, että saarella täytyi olla joku muukin ihmisolento,
kenties useampiakin. Eihän ollut muuta mahdollisuutta kuin että
hauli oli ammuttu; siis oli ollut ampujakin.
— Oletteko varma, Cyrus Smith kysyi Pencroffilta haulia
tarkastellessaan, — oletteko varma siitä, ettei tuo pekarinporsas ollut
kuin kolmen kuukauden ikäinen?
— Ei täyttä kolmeakaan, koskapa imi vielä, kun sen kuopasta löysin.
— Siis on tällä saarella laukaistu pyssy korkeintaan kolme kuukautta
sitten, insinööri lausui, — ja niin ollen päättelen: joko on saaressa
ollut asukkaita ennen meidän tänne tuloamme, tai ihmisiä on
noussut tänne maihin enintään kolme kuukautta sitten. Ovatko he
tulleet ehdoin tahdoin vai haaksirikkoisina ajautuneet rannalle, siitä
saamme myöhemmin selon samoin kuin siitäkin, ovatko he
eurooppalaisia vai malaijilaisia ja vieläkö he ovat saarella vai ovatko
jo lähteneet.
— Ei ikipäivinä täällä ole ristinsielua muita kuin me! Pencroff
huudahti. — Olisihan ne nähty jo aikoja sitten. Se on tietty se.
— Mutta tietympää on, että pekarilla oli hauli ruumiissaan, reportteri
virkkoi. — Mistä se siihen olisi tullut?
— Kunhan ei vain Pencroffilla olisi ollut entisiä jätteitä ... yritti Nab.

— Soo-o! merimies loukkaantuneena huudahti. — Vai olisi minulla
ollut puolen vuoden verran lyijyrae hampaassa, enkä olisi sitä
huomannut! Tule katsomaan, Nab, ja jos löydät suustani yhden
ainoankaan rikkinäisen hampaan, niin saat kiskoa puoli tusinaa
purimia leukapielistäni irti.
— Nab iski kirveensä kiveen, huomautti Cyrus Smith naurahtaen. —
Mutta oli miten oli, meidän täytyy ruveta tutkimaan asiaa.
Pencroff oli paikalla valmis ja tarjoutui tekemään kevyen veneen,
jolla voisi kiertää ympäri saaren ja nousta Laupeudenjokea myöten
sen latvoille saakka. Hän ei tarkoittanut varsinaista kaaritettua ja
lautalaitaista venettä, vaan sellaista puun kaarnasta tehtyä kevyttä,
tasapohjaista ruuhta, jota intiaanit käyttävät ja jota sanotaan
pirogiksi. Sen hän lupasi saada valmiiksi viidessä päivässä.
Viipymättä hän ryhtyikin työhön, jota huomattavasti helpotti se, että
äskeinen myrsky oli kaatanut useita sopivia puita. Ei tarvinnut muuta
kuin kiskoa kaarna irti. Siihen soveliaita työkaluja kyllä puuttui, mutta
jotenkin merimies keinotteli tästä pulmasta, ja niinpä hänellä ennen
pitkää oli kaarnaa koossa enemmän kuin tarvitsikaan.
Pencroffin ja insinöörin ahertaessa pirogin kimpussa toiset
metsästelivät varoen kuitenkaan joutumasta kovin kauas
Graniittilinnasta.
Kerran — oli lokakuun 28:s — sattui taas omituinen tapaus.
Kävellessään Nabin kanssa meren rannikolla Harbert sattui
huomaamaan kallioitten välissä suuren kilpikonnan, joka parhaillaan
pyrki merta kohti. He päättivät ottaa sen kiinni ja käänsivät sen
sauvoillansa selälleen, siis sellaiseen asentoon, jossa kilpikonnan on
mahdotonta liikkua. Varmemmaksi vakuudeksi Harbert asetti
kuitenkin sen ympärille suuria kiviä estääkseen sitä karkaamasta.
He kiiruhtivat sitten kotiin hakemaan rattaita, sillä Harbert tahtoi
yllättää Pencroffin tuomalla sellaisen eläimen, jolla oli aina "maja
mukanaan". Kenenkään huomaamatta he ottivat rattaat ja lähtivät
takaisin merenrantaan.

Mutta suuri oli heidän hämmästyksensä, kun siellä ei ollutkaan enää
eläintä, jolla oli "maja mukanaan".
— Kappas tuota! Nab virkkoi. — Osaavatpa nuo konnat sittenkin
kääntyä vatsalleen.
— Nähtävästi, Harbert myönsi katsellen hajalleen joutuneita kiviä.
— Mutta, Nab sanoi, — emmepä hiisku koko tästä jutusta muille
mitään.
— Päinvastoin me kerromme koko asian, Harbert vastasi.
Noloina, he menivät rannalle, missä muut olivat täyttä höyryä pirogia
rakentamassa, ja kertoivat epäonnistuneesta yrityksestään.
— Aika poropeukaloita! huudahti Pencroff. — Päästää käsistään
ainakin puolisataa vadillista maukasta lientä! Käänsitte pedon
huonosti, siinä kaikki.
Kuultuaan Harbertin tehneen kiviaitauksenkin sen ympärille Pencroff
arveli:
— Sitten siinä on tapahtunut ilmeinen ihme!
— Kuinka kaukana rannasta te käänsitte sen selälleen? Cyrus Smith
kysyi.
— Korkeintaan viidentoista askelen päässä.
— Ja pakoveden aikana?
— Niin.
— Sitten on mahdollista, että nousuvesi kohotti sen maanpinnasta,
ja vedessähän kilpikonna pääsee kääntymään.
— Mutta olimmepa tosiaankin poropeukaloita! Nab huudahti.
— Sehän minulla oli kunnia teille vastikään ilmoittaa, ivasi Pencroff.
Sellaisen selityksen Cyrus Smith antoi. Luultavinta kuitenkin on, ettei
hän itsekään sitä uskonut.

Toinen luku
Ensimmäinen pirogiretki. Rannalla oleva hylky.
Arkun sisällys. Mitä Pencroff kaipaa?
Raamatun jae.
Pencroff piti sanansa: viiden päivän kuluttua pirogi oli valmis, kolme
ja puoli metriä pitkä ja painoltaan tuskin sataakaan kiloa. Runko oli
tehty lujista puunoksista, päällys kaarnasta. Kokassa ja perässä oli
tuhto, keskellä samoin, pitämässä laitoja kiinni. Kaksi airoa ja mela
— siinä aluksen kaikki varusteet.
Viivyttelemättä ja tyytyväisinä siirtolaiset astuivat pirogiinsa.
— Kovinhan sinun laivallasi on jano, ivaili Nab nähdessään pirogin
vuotavan.
— Älä yhtään huoli, Pencroff lohdutteli. — Parin päivän kuluttua ei
siinä ole vettä enempää kuin juomarin vatsassa.
Ja sitten lähdettiin liikkeelle, Nab ja Harbert airoissa, Pencroff
perämiehenä. Kulku ohjattiin ensin salmen poikki Pelastuksenluodon
eteläpäähän ja siitä pari kilometriä ulommaksi merelle, jonka pinta
oli tällä kertaa rasvatyyni. Sieltä soudettiin Laupeuden joen suulle ja
käännyttiin kaakkoon. Noin viisi kilometriä joen suusta pisti mereen
teräväkärkinen niemeke. Sitä kohti Pencroff alkoi ohjata venettä,
pysytellen jonkin matkan päässä rannasta.
Tultiin niemenkärjen kohdalle, ja Pencroff oli juuri kiertämäisillään
sen, kun Harbert äkkiä huudahti:
— Mitä tuolla rannalla on?
Kaikki kääntyivät sinne päin: rannalla näkyi jotain mustaa, puoleksi
hiekan peitossa.

— Kas, mistä ne ovat sinne joutuneet? virkkoi Pencroff.
— Mitä ne ovat? Mitä ne ovat?
— Tynnyreitä ne ovat, ja tuskinpa tyhjiä.
Nuolena kiiti vene pienen lahdelman rantaan, ja kaikki hyppäsivät
maihin.
Pencroff oli oikeassa. Rannalla, puoleksi hiekkaan vajonneena, oli
kaksi tynnyriä, joihin taitavasti oli köytetty kiinni isohko arkku. Tämä
oli ilmeisesti kellunut tynnyrien varassa meren pinnalla, kunnes aallot
olivat sen heittäneet maihin.
— Mitähän ihmettä siinä mahtaa olla? huudahti Pencroff hyökäten
arkun ääreen. — Luja on siinä kansi, mutta eiköhän lohkeaisi, kun
oikein koettaisi? Ja samassa hän sieppasi maasta suuren kiven.
— Pencroff! Cyrus Smith virkkoi pidättäen häntä. — Ettekö malttaisi
odottaa tunnin verran?
— Mutta, herra insinööri, ajatelkaa: siinähän saattaa olla juuri
sellaista, mitä meiltä puuttuu!
— Siitä saamme kohta selon, eikä liene syytä särkeä arkkua, josta
meille vielä voi olla paljonkin hyötyä. Me viemme sen Graniittilinnaan
ja avaamme siellä varovasti.
— Kai se parasta on, merimies myönsi. — Minä olen sellainen
hätähousu!
Arkku päätettiin hinata kotiin, sillä se oli heidän pienelle ruuhelleen
liian raskas.
Mistä ja miten arkku oli joutunut sinne? Rannikko tutkittiin tarkoin,
mutta muita hylkyjä ei näkynyt missään. Harbert ja Nab kiipesivät
ylängölle ja tähyilivät merelle: ei sielläkään laivaa, ei
purrenpahastakaan.
Ilmeisesti oli tapahtunut haaksirikko. Oliko sillä ja äskeisellä haulilla
mitään yhteyttä keskenään? Kenties olivat haaksirikkoiset itse saaren

toisessa päässä? — Se oli ainakin varmaa, etteivät he olleet
malaijilaisia merirosvoja, sillä näillä ei ole sellaisia varusteita.
Arkku oli tammesta, puolitoista metriä pitkä ja yhdeksänkymmentä
senttiä leveä, päällystetty paksulla, messinkinauloin kiinnitetyllä
nahalla. Umpinaiset tynnyrit oli sidottu arkkuun oikein merimiehen
solmuin, niin vakuutti Pencroff. Erittäin hyvin se oli säilynyt, niinkuin
luonnollistakin, koska se oli joutunut hiekalle eikä kivisärkälle.
Meressä se nähtävästi ei ollut vielä kovinkaan kauan uiskennellut.
Luultavasti oli haaksirikkoon joutuneessa laivassa häthätää kerätty
tärkein mukaan otettava ja heitetty arkussa mereen siinä toivossa,
että haaksirikkoiset sen myöhemmin löytäisivät.
— Me hinaamme tämän Graniittilinnaan, puhui insinööri, — ja
otamme tavarat siitä talteen. Jos kohtaamme haaksirikkoiset,
annamme ne heille takaisin, ellemme, niin...
— Niin pidetään hyvänämme. Mitähän kummaa siinä mahtaa
ollakaan?
Toinen tynnyreitä kiinnittävä köysi kierrettiin auki ja siten saatiin
tarpeeksi pitkä hinausköysi, jolla arkku sidottiin pirogin perään.
Nousuvesi kohotti pian sen vedenpinnalle, ja niin lähdettiin
hinaamaan sitä Graniittilinnaa kohti. Niemenkärki nimitettiin
Löytöniemeksi. Puolentoista tunnin kuluttua — niin kauan kesti
soutaa tuota viiden kilometrin matkaa — saavuttiin vihdoin
kotirantaan ja arkku nostettiin maihin. Lukot väännettiin hohtimilla
irti, jonka jälkeen kansi ponnahti helposti auki.
Sisällä oli toinen, sinkistä tehty arkku. Sinkkipäällys leikattiin pitkin
pituuttaan, puoliskot taivutettiin sivuun ja sitten ruvettiin arkusta
nostelemaan esineitä toistensa perästä Pencroffin huutaessa eläköön
joka kerta, kun uusi kapine ilmestyi. Harbert taputti käsiään, ja Nab
tanssi oikeata neekeritanssia.
Kas tässä luettelo arkun aarteista:
Työkaluja: 3 moniteräistä veistä, 2 suurta kirvestä, 2 veistokirvestä,
3 höylää, 2 isompaa ja pienempi piilu, 6 talttaa, 2 viilaa, 3 vasaraa, 3

näveriä, 2 naskalia, 10 pussillista nauloja ja ruuveja, 3 erisuuruista
sahaa, 2 rasiallista silmäneuloja.
Aseita: 2 piilukkopyssyä, 2 nallipyssyä, 2 takaaladattavaa luodikkoa,
5 metsäpuukkoa, 4 keihästä, 2 ruutitynnyriä, kummassakin noin 10
kiloa ruutia, 12 nallirasiaa.
Tieteellisiä kojeita: sekstantti, kaksoiskiikari, kaukoputki, harpikko,
taskukompassi, Fahrenheitin lämpömittari, aneroidi-ilmapuntari,
valokuvauskone kaikkine tarpeineen.
Vaatteita: 2 tusinaa paitoja omituisesta kankaasta, joka näytti
villalta, mutta oli ilmeisesti kasvinkuiduista kudottua, 3 tusinaa
sukkia samanlaisesta kankaasta.
Astioita: rautakattila, 6 tinattua kuparikattilaa, 3 rautapannua, 10
alumiinilautasta, 2 pataa, kamina, 6 pöytäveistä.
Kirjoja: Raamattu, kartasto, kokoelma polynesialaisia sanoja
useimmilla murteilla, luonnontieteellinen kuusiniteinen sanakirja, 3
riisiä valkoista paperia, 2 puhdasta muistikirjaa.
Mitä enemmän siirtolaiset näitä aarteita katselivat, sitä suuremmaksi
kävi heidän kummastuksensa. Ei ainoassakaan esineessä, ei edes
kirjoissa näkynyt mitään meikkiä, josta olisi saattanut arvailla
omistajan nimeä tai edes kansallisuutta. Tavaroiden hyvä järjestys,
osoitti, ettei niitä ollut viskottu umpimähkään arkkuun vaan
suunnitelmallisesti järjestetty. Kummallisinta kaikista oli
valokuvauskone tarpeineen.
Mutta olipa arkku nyt tullut tavalla tai toisella, siinä oli joka
tapauksessa kaikki se, mitä he niin kipeästi tarvitsivat ja mitä he itse
eivät olisi pystyneet tekemään. He olivat mielestään nyt rikkaita ja
kiittivät sydämestänsä kaiken hyvän antajaa.
Pencroff yksin oli hiukan nyreissään.
— Jokainen teistä sai jotain, minä en mitään, hän murahti.
— Mitä sinulle olisi pitänyt olla? Nab kysyi.

— Puoli naulaa tupakkaa, Pencroff vastasi vakavasti, — sitten olisin
ollut onnellisin ihminen maailmassa.
Toisten täytyi väkisinkin naurahtaa merimiehen mieliteoille.
Päätettiin lähteä seuraavana päivänä aamun koitteessa
tiedusteluretkelle tämän kummallisen löydön selvittämiseksi. Kaikki
esineet vietiin Graniittilinnaan ja asetettiin järjestykseen.
Tänään, lokakuun 29:ntenä, oli sunnuntai. Ennen maatapanoa
Harbert pyysi insinööriä lukemaan jotain Raamatusta.
Cyrus Smith otti Raamatun ja oli juuri avaamaisillaan sen, kun
Pencroff pidätti häntä sanoen:
— Herra insinööri! Minä olen hieman taikauskoinen, minä. Avatkaapa
kirja umpimähkään ja lukekaa se jae, johon silmänne ensiksi
sattuvat.
Hymähtäen Cyrus Smith avasi Raamatun juuri siltä kohdalta, missä
kahden lehden välissä oli merkki. Hänen silmänsä osuivat kohtaan,
johon punaisella kynällä oli merkitty risti. Se oli Matteuksen
evankeliumin 7. luvun 8. jakeen kohdalla. Hän luki:
"Jokainen anova saa, ja etsivä löytää."

Kolmas luku
Pirogilla. Runsas kasvillisuus ja eläinmaailma.
Eukalyptuspuita, apinoita. Yöleiri.
Seuraavana päivänä, lokakuun 30:ntenä, kello kuusi aamulla
hankkiuduttiin matkalle. Pirogiin pantiin pari suurta kirvestä ja eväitä
kolmeksi päiväksi. Ampuma-aseita otettiin kaksi piilukkoa ja luodikko,
edelliset siitä syystä, että sytyttimet niissä saattaisi helposti uusia ja
että nalleja säästyisi, mikäli mahdollista. Metsäpuukon pisti kukin
vyölleen. Näin varustettuina he lähtivät, Top tietysti mukana,
soutamaan Laupeuden joen suuta kohti käyttäen hyväkseen
nousuvettä.
Neitseellisenä humisi metsä joen kummallakin rannalla; ihmiskäden
jälkeä ei missään. Tuon tuostakin laskettiin rantaan, ja useimmiten
Gideon Spilett ja Harbert pistäytyivät viidakkoon, jonka uhkea
kasvillisuus hämmästytti ja ihastutti nuorta luonnontutkijaa. Siellä
kasvoi reheviä salavalajeja, laivanrakentajille niin mieluisia, ne kun
kestävät vedessä kauan aikaa lahoamatta; siellä oli taajoina ryhminä
Celtis-sukuisia puita, joitten pähkinöistä tehdään erinomaista öljyä;
siellä oli köynnöskasveja, joitten oksista saadaan erittäin kestäviä
köysiä.
Insinööri istui koko ajan pirogissa seuraten kompassin mukaan joen
suuntaa.
Kerran maissa käydessä Gideon Spilettin onnistui saada elävänä
kiinni kaksi kanan sukuista lintua, joilla oli pitkä, litteä nokka, pitkä
kaula, lyhyet siivet eikä lainkaan pyrstöä. Nämä tinamu-nimiset
linnut päätettiin kesyttää tulevan kanatarhan vanhemmiksi.
Kajahtipa viimein ensimmäinen pyssynlaukauskin Kaukaisen Lännen
metsissä, ja erämiesten saaliiksi jäi komea lintu, jäälintujen

sukulainen, jonka kankeat sulat vivahtelivat metallinvärisinä. Se oli
nimeltään jakamari, ja sen kunniaksi annettiin senpuoleiselle
metsälle nimeksi Jakamarin metsä.
Kello kymmenen tienoissa päästiin joen toiseen mutkaan. Siihen
pysähdyttiin aterioimaan. Kun matkaa puolen tunnin kuluttua
jatkettiin, huomattiin, ettei nousuvesi enää tuntunutkaan. Täytyi niin
ollen turvautua taas airoihin. Tiheiköt alkoivat harveta, mutta sitä
paksummiksi ja upeammiksi kävivät puut. Komeina kohosi joen
rannoilla varsinkin eukalyptus-puita, paikoin kuusikymmentäkin
metriä korkeina, tyvestä noin kuusi metriä ympäri mitaten. Niiden
toistakymmentä senttiä paksu kuori oli hyvänhajuisen hartsiverkon
peittämä.
— Nuopa ovat puitten puita! huudahti Nab, — mutta lieneekö niistä
mitään hyötyä?
— Katin kontit! Pencroff arveli. — Mutta pitäähän maailmassa olla
jättiläispuita, niinkuin on ihmisissäkin jättiläisiä, joilla ei ole muuta
virkaa kuin näytellä itseään markkinoilla.
— Suuri erehdys, Pencroff! selitti Harbert. — Nämä puut puhdistavat
ilmaa. Australiassa ja Uudessa Seelannissa niitä sanotaankin
kuumepuiksi.
— Siksikö, että niistä leviää kuumetta ja muuta horkkaa?
— Ei, vaan siksi, että ne estävät kuumetauteja, Cyrus Smith selitti.
— Sekä Euroopassa että Afrikassa on näitä puita istutettu
epäterveellisiin, suoperäisiin paikkoihin, ja siitä lähtien on asukkaiden
terveydentila huomattavasti parantunut.
— Näkyypä sitten tässä maassa olevan jos jotakin hyvää, ja jos vain
olisi vielä...
— Älkää hätäilkö, Pencroff, kukaties löydämme sitäkin.
Lähdettiin taas jatkamaan matkaa. Joen uoma alkoi nyt ilmeisesti
kaveta ja käydä matalammaksi, niin että vain sauvomalla päästiin
eteenpäin. Aurinko kallistui länteen, ja senvuoksi päätettiin yöpyä
rantaäyräälle. Meren länsirantaan, johon heidän oli määrä päästä, oli

Cyrus Smithin laskujen mukaan vielä noin yhdeksän, kymmenen
kilometriä. Viidakko tiheni taas molemmin puolin jokea, ja
eläinmaailma kävi yhä runsaammaksi. Pencroffin tarkka silmä oli
kauempana näkevinään apinoitakin hyppimässä puitten oksilla, ja
oikeassa hän olikin. Muutamia juoksi rannalla, ja siinä nuo nelikätiset
oudostellen mutta pelkäämättä katselivat pirogia ja sen omituisia
olentoja. Apinoita olisi ollut helppoa ampuakin, mutta turhaan siinä
vain olisi kulutettu ampumavaroja. Ja Pencroff olisi ollut harmissaan
riistasta, josta ei ollut pataan pantavaksi.
Heidän vielä hetken aikaa kuljettuaan alkoi edestä päin kuulua
kohinaa, ja pian näkyi puitten lomitse vesiputous. Kohta pirogin
pohja raapaisi kiviin. Täytyi siis pysähtyä. Siirtolaiset nousivat
maihin, sitoivat veneensä lujasti puuhun ja rupesivat hankkiutumaan
yöpuulle.
Illallinen valmistettiin tuota pikaa, sillä retkeilijät olivat kovin nälkäisiä
ja väsyneitä. Yöleirin ympärille sytytettiin nuotioita suojaksi pedoilta,
joitten karjuntaa alkoi kuulua pimeän tultua.

Welcome to our website – the perfect destination for book lovers and
knowledge seekers. We believe that every book holds a new world,
offering opportunities for learning, discovery, and personal growth.
That’s why we are dedicated to bringing you a diverse collection of
books, ranging from classic literature and specialized publications to
self-development guides and children's books.
More than just a book-buying platform, we strive to be a bridge
connecting you with timeless cultural and intellectual values. With an
elegant, user-friendly interface and a smart search system, you can
quickly find the books that best suit your interests. Additionally,
our special promotions and home delivery services help you save time
and fully enjoy the joy of reading.
Join us on a journey of knowledge exploration, passion nurturing, and
personal growth every day!
ebookbell.com