Professional Documents
Culture Documents
Textbook Geometry of Quantum States An Introduction To Quantum Entanglement Ingemar Bengtsson Ebook All Chapter PDF
Textbook Geometry of Quantum States An Introduction To Quantum Entanglement Ingemar Bengtsson Ebook All Chapter PDF
https://textbookfull.com/product/geometry-of-quantum-states-an-
introduction-to-quantum-entanglement-ingemar-bengtsson-2/
https://textbookfull.com/product/an-introduction-to-quantum-
optics-and-quantum-fluctuations-first-edition-edition-milonni/
https://textbookfull.com/product/an-introduction-to-quantum-
physics-first-edition-anthony-p-french/
https://textbookfull.com/product/an-introduction-to-quantum-
field-theory-first-edition-michael-e-peskin/
An Introduction to Quantum and Vassiliev Knot
Invariants David M. Jackson
https://textbookfull.com/product/an-introduction-to-quantum-and-
vassiliev-knot-invariants-david-m-jackson/
https://textbookfull.com/product/an-introduction-to-quantum-
transport-in-semiconductors-1st-edition-david-k-ferry/
https://textbookfull.com/product/introduction-to-quantum-
mechanics-david-j-griffiths/
https://textbookfull.com/product/an-introduction-to-quantum-
optics-photon-and-biphoton-physics-2nd-edition-yanhua-shih/
https://textbookfull.com/product/introduction-to-quantum-field-
theory-with-applications-to-quantum-gravity-1st-edition-iosif-l-
buchbinder-ilya-shapiro/
G E O M E T RY O F Q UA N T U M S TAT E S
An Introduction to Quantum Entanglement
SECOND EDITION
INGEMAR BENGTSSON
Stockholm University, Sweden
K A RO L Ż Y C Z KOW S K I
Jagiellonian University, Poland
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
4843/24, 2nd Floor, Ansari Road, Daryaganj, Delhi – 110002, India
79 Anson Road, #06-04/06, Singapore 079906
www.cambridge.org
Information on this title: www.cambridge.org/9781107026254
DOI: 10.1017/9781139207010
© Ingemar Bengtsson and Karol Życzkowski 2017
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 2006
Second edition 2017
Printed in the United Kingdom by Clays, St Ives plc
A catalogue record for this publication is available from the British Library.
ISBN 978-1-107-02625-4 Hardback
ISBN 978-1-107-65614-7 Paperback
Cambridge University Press has no responsibility for the persistence or accuracy
of URLs for external or third-party Internet Web sites referred to in this publication
and does not guarantee that any content on such Web sites is, or will remain,
accurate or appropriate.
Contents
Preface page xi
Acknowledgements xv
1 Convexity, colours and statistics 1
1.1 Convex sets 1
1.2 High dimensional geometry 9
1.3 Colour theory 15
1.4 What is ‘distance’? 18
1.5 Probability and statistics 25
Problems 28
2 Geometry of probability distributions 29
2.1 Majorisation and partial order 29
2.2 Shannon entropy 35
2.3 Relative entropy 41
2.4 Continuous distributions and measures 46
2.5 Statistical geometry and the Fisher–Rao metric 48
2.6 Classical ensembles 55
2.7 Generalised entropies 56
Problems 62
3 Much ado about spheres 63
3.1 Spheres 63
3.2 Parallel transport and statistical geometry 68
3.3 Complex, Hermitian, and Kähler manifolds 74
3.4 Symplectic manifolds 80
3.5 The Hopf fibration of the 3-sphere 82
3.6 Fibre bundles and their connections 88
3.7 The 3-sphere as a group 95
v
vi Contents
Epilogue 544
Appendix A Basic notions of differential geometry 545
A.1 Differential forms 545
A.2 Riemannian curvature 546
A.3 A key fact about mappings 547
Appendix B Basic notions of group theory 549
B.1 Lie groups and Lie algebras 549
B.2 SU(2) 550
B.3 SU(N) 551
B.4 Homomorphisms between low-dimensional groups 552
Appendix C Geometry: do it yourself 553
Appendix D Hints and answers to the exercises 557
References 572
Index 614
Preface
xi
xii Preface
bits that their sample spaces conspire to form a space – namely the surface of the
ball – large enough to carry the canonical transformations that are characteristic of
mechanics. Hence the word quantum mechanics.
It is not particularly difficult to understand a three–dimensional ball, or to see
how this description emerges from the usual description of a qubit in terms of a
complex two-dimensional Hilbert space. In this case we can take the word geometry
literally – there will exist a one-to-one correspondence between pure states of
the qubit and the points of the surface of the Earth. Moreover, at least as far
as the surface is concerned, its geometry has a statistical meaning when transcribed
to the qubit (although we will see some strange things happening in the interior).
As the dimension of the Hilbert space goes up, the geometry of the state spaces
becomes very intricate, and qualitatively new features arise – such as the subtle way
in which composite quantum systems are represented. Our purpose is to describe
this geometry. We believe it is worth doing. Quantum state spaces are more won-
derful than classical state spaces, and in the end composite systems of qubits may
turn out to have more practical applications than the bits themselves already have.
A few words about the contents of our book. As a glance at the Contents will
show, there are 17 chapters, culminating in a long chapter on ‘entanglement’. Along
the way, we take our time to explore many curious byways of geometry. We expect
that you – the reader – are familiar with the principles of quantum mechanics at
the advanced undergraduate level. We do not really expect more than that, and
should you be unfamiliar with quantum mechanics we hope that you will find some
sections of the book profitable anyway. You may start reading any chapter – if
you find it incomprehensible we hope that the cross-references and the index will
enable you to see what parts of the earlier chapters may be helpful to you. In the
unlikely event that you are not even interested in quantum mechanics, you may
perhaps enjoy our explanations of some of the geometrical ideas that we come
across.
Of course there are limits to how independent the chapters can be of each other.
Convex set theory (Chapter 1) pervades all statistical theories, and hence all our
chapters. The ideas behind the classical Shannon entropy and the Fisher–Rao
geometry (Chapter 2) must be brought in to explain quantum mechanical entropies
(Chapter 12) and quantum statistical geometry (Chapters 9 and 13). Sometimes
we have to assume a little extra knowledge on the part of the reader, but since no
chapter in our book assumes that all the previous chapters have been understood,
this should not pose any great difficulties.
We have made a special effort to illustrate the geometry of quantum mechan-
ics. This is not always easy, since the spaces that we encounter more often than
not have a dimension higher than three. We have simply done the best we could.
To facilitate self-study each chapter concludes with problems for the reader, while
some additional geometrical exercises are presented in the final appendix.
Preface xiii
Once and for all, let us say that we limit ourselves to finite-dimensional state
spaces. We do this for two reasons. One of them is that it simplifies the story very
much, and the other is that finite-dimensional systems are of great independent
interest in real experiments.
The entire book may be considered as an introduction to quantum entanglement.
This very non-classical feature provides a key resource for several modern applica-
tions of quantum mechanics including quantum cryptography, quantum computing
and quantum communication. We hope that our book may be useful for graduate
and postgraduate students of physics. It is written first of all for readers who do not
read the mathematical literature everyday, but we hope that students of mathematics
and of the information sciences will find it useful as well, since they also may wish
to learn about quantum entanglement.
We have been working on the book for about five years. Throughout this time
we enjoyed the support of Stockholm University, the Jagiellonian University in Cra-
cow, and the Center for Theoretical Physics of the Polish Academy of Sciences in
Warsaw. The book was completed in Waterloo during our stay at the Perimeter Insti-
tute for Theoretical Physics. The motto at its main entrance – AOYATO
EPI EMETPIA MHEI EIIT – proved to be a lucky omen indeed,
and we are pleased to thank the Institute for creating optimal working conditions
for us, and to thank all the knowledgeable colleagues working there for their help,
advice, and support.
We are grateful to Erik Aurell for his commitment to Polish–Swedish colla-
boration; without him the book would never have been started. It is a pleasure
to thank our colleagues with whom we have worked on related projects: Johan
Brännlund, Åsa Ericsson, Sven Gnutzmann, Marek Kuś, Florian Mintert,
Magdalena Sinołȩcka, Hans-Jürgen Sommers and Wojciech Słomczyński. We
are grateful to them, and to many others who helped us to improve the manuscript.
If it never reached perfection, it was our fault, not theirs. Let us mention some
of the others: Robert Alicki, Anders Bengtsson, Iwo Białynicki-Birula, Rafał
Demkowicz-Dobrzański, Johan Grundberg, Sören Holst, Göran Lindblad, and
Marcin Musz. We have also enjoyed constructive interactions with Matthias
Christandl, Jens Eisert, Peter Harremoës, Michał, Paweł and Ryszard Horodeccy,
Vivien Kendon, Artur Łoziński, Christian Schaffner, Paul Slater, and William
Wootters.
Five other people provided indispensable support: Martha and Jonas in Stock-
holm, and Jolanta, Jaś, and Marysia in Cracow.
Waterloo, 12 March 2005 Ingemar Bengtsson
Karol Życzkowski
xiv Preface
Figures 4.10–13 and 16.1–3 have appeared already in our reference [112], which
was published in the International Journal of Physics A17 Issue 31 (2002), coau-
thored with Johan Brännlund. They appear here with the kind permission of the
copyright holder, World Scientific.
Figure 5.2 and the Figure accompanying the Hint to Problem 5.2 are reprinted from
Dr. Kate Blanchfield’s PhD thesis, with the kind permission of its copyright holder
and author.
xv
1
Convexity, colours and statistics
What picture does one see, looking at a physical theory from a distance, so that the details
disappear? Since quantum mechanics is a statistical theory, the most universal picture which
remains after the details are forgotten is that of a convex set.
—Bogdan Mielnik1
Our object is to understand the geometry of the set of all possible states of a
quantum system that can occur in nature. This is a very general question; especially
since we are not trying to define ‘state’ or ‘system’ very precisely. Indeed we will
not even discuss whether the state is a property of a thing, or of the preparation of a
thing, or of a belief about a thing. Nevertheless we can ask what kind of restrictions
are needed on a set if it is going to serve as a space of states in the first place. There
is a restriction that arises naturally both in quantum mechanics and in classical
statistics: the set must be a convex set. The idea is that a convex set is a set such that
one can form ‘mixtures’ of any pair of points in the set. This is, as we will see, how
probability enters (although we are not trying to define ‘probability’ either).
From a geometrical point of view a mixture of two states can be defined as a
point on the segment of the straight line between the two points that represent the
states that we want to mix. We insist that given two points belonging to the set of
states, the straight line segment between them must belong to the set too. This is
certainly not true of any set. But before we can see how this idea restricts the set
of states we must have a definition of ‘straight lines’ available. One way to proceed
is to regard a convex set as a special kind of subset of a flat Euclidean space En .
Actually we can get by with somewhat less. It is enough to regard a convex set as a
subset of an affine space. An affine space is just like a vector space, except that no
1
2 Convexity, colours and statistics
special choice of origin is assumed. The straight line through the two points x1 and
x2 is defined as the set of points
x = μ1 x1 + μ2 x2 , μ1 + μ2 = 1. (1.1)
If we choose a particular point x0 to serve as the origin, we see that this is a one-
parameter family of vectors x − x0 in the plane spanned by the vectors x1 − x0 and
x2 − x0 . Taking three different points instead of two in Eq. (1.1) we define a plane,
provided the three points do not belong to a single line. A k-dimensional k–plane
is obtained by taking k + 1 generic points, where k < n. For k = n we describe
the entire space En . In this way we may introduce barycentric coordinates into an
n-dimensional affine space. We select n + 1 points xi , so that an arbitrary point x
can be written as
x = μ0 x0 + μ1 x1 + . . . + μn xn , μ0 + μ1 + . . . + μn = 1. (1.2)
The requirement that the barycentric coordinates μi add up to one ensures that they
are uniquely defined by the point x. (It also means that the barycentric coordinates
are not coordinates in the ordinary sense of the word, but if we solve for μ0 in terms
of the others then the remaining independent set is a set of n ordinary coordinates
for the n-dimensional space.) An affine map is a transformation that takes lines to
lines and preserves the relative length of line segments lying on parallel lines. In
equations an affine map is a combination of a linear transformation described by a
matrix A with a translation along a constant vector b, so x = Ax + b, where A is
an invertible matrix.
By definition a subset S of an affine space is a convex set if for any pair of points
x1 and x2 belonging to the set it is true that the mixture x also belongs to the set,
where
x = λ1 x1 + λ2 x2 , λ1 + λ2 = 1, λ1 , λ2 ≥ 0. (1.3)
Here λ1 and λ2 are barycentric coordinates on the line through the given pair
of points; the extra requirement that they be positive restricts x to belong to the
segment of the line lying between the pair of points.
It is natural to use an affine space as the ‘container’ for the convex sets since
convexity properties are preserved by general affine transformations. On the other
hand it does no harm to introduce a flat metric on the affine space, turning it into
an Euclidean space. There may be no special significance attached to this notion
of distance, but it helps in visualizing what is going on. See Figures 1.1 and 1.2.
From now on, we will assume that our convex sets sit in Euclidean space, whenever
it is convenient to do so.
Intuitively a convex set is a set such that one can always see the entire set from
whatever point in the set one happens to be sitting at. Still they can come in a variety
of interesting shapes. We will need a few definitions. First, given any subset of the
1.1 Convex sets 3
Figure 1.1 Three convex sets, two of which are affine transformations of each
other. The new moon is not convex. An observer in Singapore will find the new
moon tilted but still not convex, since convexity is preserved by rotations.
Figure 1.2 The convex sets we will consider are either convex bodies (like the
simplex on the left, or the more involved example in the centre) or convex cones
with compact bases (an example is shown on the right).
affine space we define the convex hull of this subset as the smallest convex set that
contains the set. The convex hull of a finite set of points is called a convex polytope.
If we start with p + 1 points that are not confined to any (p − 1)-dimensional
subspace then the convex polytope is called a p-simplex. The p-simplex consists of
all points of the form
x = λ0 x0 + λ1 x1 + . . . + λp xp , λ0 + λ1 + . . . + λp = 1, λi ≥ 0. (1.4)
(The barycentric coordinates are all non–negative.) The dimension of a convex set
is the largest number n such that the set contains an n-simplex. When discussing
a convex set of dimension n we usually assume that the underlying affine space
also has dimension n, to ensure that the convex set possesses interior points (in the
sense of point set topology). A closed and bounded convex set that has an interior
is known as a convex body.
The intersection of a convex set with some lower dimensional subspace of the
affine space is again a convex set. Given an n-dimensional convex set S there is also
a natural way to increase its dimension with one: choose a point y not belonging
to the n-dimensional affine subspace containing S. Form the union of all the rays
(in this chapter a ray means a half line), starting from y and passing through S.
4 Convexity, colours and statistics
Figure 1.3 Left: a convex cone and its dual, both regarded as belonging to
Euclidean 2-space. Right: a self dual cone, for which the dual cone coincides with
the original. For an application of this construction see Figure 11.6.
The result is called a convex cone and y is called its apex, while S is its base. A ray
is in fact a one dimensional convex cone. A more interesting example is obtained by
first choosing a p-simplex and then interpreting the points of the simplex as vectors
starting from an origin O not lying in the simplex. Then the p + 1 dimensional set
of points
x = λ0 x0 + λ1 x1 + . . . + λp xp , λi ≥ 0 (1.5)
is a convex cone. Convex cones have many nice properties, including an inbuilt
partial order among its points: x ≤ y if and only if y − x belongs to the cone. Linear
maps to R that take positive values on vectors belonging to a convex cone form
a dual convex cone in the dual vector space. Since we are in the Euclidean vector
space En , we can identify the dual vector space with En itself. If the two cones agree
the convex cone is said to be self dual. See Figure 1.3. One self dual convex cone
that will appear now and again is the positive orthant or hyperoctant of En , defined
as the set of all points whose Cartesian coordinates are non-negative. We use the
notation x ≥ 0 to denote the fact that x belongs to the positive orthant.
From a purely topological point of view all convex bodies are equivalent to an
n–dimensional ball. To see this choose any point x0 in the interior and then for every
point in the boundary draw a ray starting from x0 and passing through the boundary
point (as in Figure 1.4). It is clear that we can make a continuous transformation
1.1 Convex sets 5
of the convex body into a ball with radius one and its centre at x0 by moving the
points of the container space along the rays.
Convex bodies and convex cones with compact bases are the only convex sets
that we will consider. Convex bodies always contain some special points that cannot
be obtained as mixtures of other points – whereas a half space does not! These
points are called extreme points by mathematicians and pure points by physicists
(actually, originally by Weyl), while non-pure points are called mixed. In a convex
cone the rays from the apex through the pure points of the base are called extreme
rays; a point x lies on an extreme ray if and only if y ≤ x ⇒ y = λx with
λ between zero and one. A subset F of a convex set that is stable under mix-
ing and purification is called a face of the convex set. What the phrase means is
that if
x = λx1 + (1 − λ)x2 , 0<λ<1 (1.6)
then x lies in F if and only if x1 and x2 lie in F. The ‘only if’ part of the definition
forces x to lie on the boundary of the set. A face of dimension k is a k-face. A 0-face
is an extreme point, and an (n − 1)-face is also known as a facet. It is interesting to
observe that the set of all faces on a convex body form a partially ordered set; we
say that F1 ≤ F2 if the face F1 is contained in the face F2 . It is a partially ordered
set of the special kind known as a lattice, which means that a given pair of faces
always have a greatest lower bound (perhaps the empty set) and a lowest greater
bound (perhaps the convex body itself).
To stem the tide of definitions let us quote two theorems that have an ‘obvious’
ring to them when they are stated abstractly but which are surprisingly useful in
practice:
Minkowski’s theorem. Any convex body is the convex hull of its pure points.
Carathéodory’s theorem. Any point in an n-dimensional convex set X can be
expressed as a convex combination of at most n + 1 pure points in X .
Thus any point x of a convex body S may be expressed as a convex combination of
pure points:
p
x= λi xi , λi ≥ 0, p ≤ n + 1, λi = 1. (1.7)
i=1 i
This equation is quite different from Eq. (1.2) that defined the barycentric coordi-
nates of x in terms of a fixed set of points xi , because – with the restriction that all
the coefficients be non-negative – it may be impossible to find a finite set of xi so
that every x in the set can be written in this new form. An obvious example is a
circular disk. Given x one can always find a finite set of pure points xi so that the
6 Convexity, colours and statistics
Figure 1.5 In a simplex a point can be written as a mixture in one and only one
way. In general the rank of a point is the minimal number of pure points needed
in the mixture; the rank may change in the interior of the set as shown in the
rightmost example. The set on the right has two non-exposed points which form
faces of their own. Note also that its insphere is not unique.
equation holds, but that is a different thing. The exact number of pure points one
needs is related to the face structure of the body, as one can see from the proof of
Carathéodory’s theorem (which we give as Problem 1.1.)
It is evident that the pure points always lie in the boundary of the convex set, but
the boundary often contains mixed points as well. The simplex enjoys a very special
property, which is that any point in the simplex can be written as a mixture of pure
points in one and only one way (as in Figure 1.5). This is because for the simplex
the coefficients in Eq. (1.7) are barycentric coordinates and the result follows from
the uniqueness of the barycentric coordinates of a point. No other convex set has
this property. The rank of a point x is the minimal number p needed in the convex
combination (1.7). By definition the pure points have rank one. In a simplex the
edges have rank two, the faces have rank three, and so on, while all the points in
the interior have maximal rank. From Eq. (1.7) we see that the maximal rank of any
point in a convex body in Rn does not exceed n + 1. In a ball all interior points have
rank two and all points on the boundary are pure, regardless of the dimension of the
ball. It is not hard to find examples of convex sets where the rank changes as we
move around in the interior of the set (see Figure 1.5).
The simplex has another quite special property, namely that its lattice of faces is
self dual. We observe that the number of k-faces in an n dimensional simplex is
n+1 n+1
= . (1.8)
k+1 n−k
Hence the set of n−k−1 dimensional faces can be put in one-to-one correspondence
with the set of k-faces. In particular, the pure points (k = 0) can be put in one-to-one
correspondence with the set of facets (by definition, the n − 1 dimensional faces).
1.1 Convex sets 7
For this, and other, reasons its lattice of subspaces will have some exceptional
properties, turning it into what is technically known as a Boolean lattice.2
There is a useful dual description of convex sets in terms of supporting hyper-
planes. A support hyperplane of S is a hyperplane that intersects the set and which is
such that the entire set lies in one of the closed half spaces formed by the hyperplane
(see Figure 1.6). Hence a support hyperplane just touches the boundary of S, and
one can prove that there is a support hyperplane passing through every point of
the boundary of a convex body. By definition a regular point is a point on the
boundary that lies on only one support hyperplane, a regular support hyperplane
meets the set in only one point, and the entire convex set is regular if all its boundary
points as well as all its support hyperplanes are regular. So a ball is regular, while
a convex polytope or a convex cone is not – indeed all the support hyperplanes of
a convex cone pass through its apex. A face is said to be exposed if it equals the
intersection of the convex set and some support hyperplane. Convex polytopes arise
as the intersection of a finite number of closed half-spaces in Rn , and any pure point
of a convex polytope saturates n of the inequalities that define the half-spaces; again
a statement with an ‘obvious’ ring that is useful in practice.
In a flat Euclidean space a linear function to the real numbers takes the form
x → a · x, where a is some constant vector. Geometrically, this defines a family of
parallel hyperplanes. We have the important
Hahn–Banach separation theorem. Given a convex body and a point x0 that
does not belong to it. Then one can find a linear function f and a constant k such
that f (x) > k for all points belonging to the convex body, while f (x0 ) < k.
This is again almost obvious if one thinks in terms of hyperplanes.
It is useful to know a bit more about dual convex sets. For definiteness let us
start out with a three dimensional vector space in which a point is represented by a
vector y. Then its dual plane is the set of vectors x such that
x · y = −1. (1.9)
2 Because it is related to what George Boole thought were the laws of thought; see Varadarajan’s book [916] on
quantum logic for these things.
8 Convexity, colours and statistics
Figure 1.7 A square is dual to another square; on the left we see how the points
on the edge at the top define a corner at the bottom of the dual square. To the right
we see a more complicated convex set with non-exposed faces (that are points). Its
dual has non-polyhedral corners. In both cases the unit circle is shown dashed.
The constant on the right hand side was set to −1 for convenience. The dual of a
line is the intersection of a one-parameter family of planes dual to the points on the
line. This is in itself a line. The dual of a plane is a point, while the dual of a curved
surface is another curved surface – the envelope of the planes that are dual to the
points on the original surface. To define the dual of a convex body with a given
boundary we change the definition slightly, and include all points on one side of the
dual planes in the dual. Thus the dual X ◦ of a convex body X is defined to be
X ◦ = {x | c + x · y ≥ 0 ∀y ∈ X}, (1.10)
where c is a number that was set equal to 1 above (and also when drawing
Figure 1.7). The dual of a convex body including the origin is the intersection
of the half-spaces defined by the pure points y of X. The dual of the dual of a
body that includes the origin is equal to the convex hull of the original body. If we
enlarge a convex body the conditions on the dual become more stringent, and hence
the dual shrinks. The dual of a sphere centred at the origin is again a sphere, so a
sphere (of suitable radius) is self dual. The dual of a cube is an octahedron. The
dual of a regular tetrahedron is another copy of the original tetrahedron, possibly
of a different size. The copy can be made to coincide with the original by means of
an affine transformation, hence the tetrahedron is a self dual body.3
We will find much use for the concept of convex functions. A real function f (x)
defined on a closed convex subset X of Rn is called convex, if for any x, y ∈ X and
λ ∈ [0, 1] it satisfies
f (λx + (1 − λ)y) ≤ λf (x) + (1 − λ)f (y). (1.11)
3 To readers who wish to learn more about convex sets – or who wish to see proofs of the various assertions that
we left unproved – we recommend the book by Eggleston [283].
1.2 High dimensional geometry 9
(a) (b)
0.5 0.5
f(x) −f(x)
0 0
−0.5 −0.5
0 0.5 1 0 0.5 1
x x
Figure 1.8 (a): the convex function f (x) = x ln x (b): the concave function
g(x) = −x ln x. The names stem from the shaded epigraphs of the functions which
are convex and concave, respectively.
The name refers to the fact that the epigraph of a convex function, that is the region
lying above the curve f (x) in the graph, is convex. Applying the inequality k − 1
times we see that
⎛ ⎞
k k
f⎝ λj xj ⎠ ≤ λj f (xj ), (1.12)
j=1 j=1
k
where xj ∈ X and the nonnegative weights sum to unity, j=1 λj = 1. If a function
f from R to R is differentiable, it is convex if and only if
asking some simple questions about convex bodies in flat space. We choose to
look at balls, cubes and simplices for this purpose. A flat metric is assumed. Our
questions will concern the inspheres and outspheres of these bodies (defined as the
largest inscribed sphere and the smallest circumscribed sphere, respectively). For
any convex body the outsphere is uniquely defined, while the insphere is not – one
can show that the upper bound on the radius of inscribed spheres is always attained
by some sphere, but there may be several of those.
Let us begin with the surface of a ball, namely the n-dimensional sphere. In
equations a sphere of radius r is given by the set
We want to compute the volume vol(Sn ) of the n-sphere, that is to say its
‘hyperarea’ – meaning that vol(S2 ) is measured in square metres, vol(S3 ) in cubic
metres, and so on. A clever trick simplifies the calculation: Consider the well known
Gaussian integral
√
I = e−X0 −X1 − ... −Xn dX0 dX1 . . . dXn = ( π)n+1 .
2 2 2
(1.17)
Using the spherical polar coordinates introduced above our integral splits into two,
one of which
∞ is related to the integral representation of the Euler Gamma function,
(x) = 0 e−t tx−1 dt, and the other is the one we want to do:
∞
−r2 n 1 n+1
I= dr dAe r = vol(Sn ). (1.18)
0 Sn 2 2
We do not have to do the integral over the angles. We simply compare these results
and obtain (recalling the properties of the Gamma function)
1.2 High dimensional geometry 11
⎧ p
⎪ 2(2π)
⎪
⎪ if n = 2p
π
n+1
2 ⎨ (2p − 1)! !
vol(S ) = 2 n+1 =
n
, (1.19)
( 2 ) ⎪
⎪
⎪ (2π)p+1
⎩ if n = 2p + 1
(2p)! !
where double factorial is the product of every other number, 5! ! = 5 · 3 · 1 and
6! ! = 6 · 4 · 2. An alarming thing happens as the dimension grows. For large x we
can approximate the Gamma function using Stirling’s formula
√ −x x− 21 1 1
(x) = 2π e x 1+ +o 2 . (1.20)
12x x
Hence for large n we obtain
n
√ 2πe 2
vol(S ) ∼ 2
n
. (1.21)
n
This is small if n is large! In fact the ‘biggest’ unit sphere – in the sense that it has
the largest hyperarea – is S6 , which has
16 3
vol(S6 ) = π ≈ 33.1. (1.22)
15
Incidentally Stirling’s formula gives 31.6, which is already rather good. We hasten
to add that vol(S2 ) is measured in square metres and vol(S6 ) in (metre)6 , so that the
direct comparison makes no sense.
There is another funny thing to be noticed. If we compute the volume of the
n-sphere without any clever tricks, simply by integrating the volume element dA
using angular coordinates, then we find that
π π π
vol(S ) = 2π
n
dθ sin θ 2
dθ sin θ . . . dθ sinn−1 θ = (1.23)
0
π 0 0
= vol(Sn−1 ) dθ sinn−1 θ.
0
As n grows the integrand of the final integral has an increasingly sharp peak close to
the equator θ = π/2. Hence we conclude that when n is high most of the hyperarea
of the sphere is confined to a ‘band’ close to the equator. What about the volume
of an n-dimensional unit ball Bn ? By definition it has unit radius and its boundary
1
is Sn−1 . Its volume, using the radial integral 0 rn−1 dr = 1/n and the fact that
(x + 1) = x(x), is
n n
vol(Sn−1 ) π2 1 2πe 2
vol(B ) =
n
= ∼√ . (1.24)
n ( 2 + 1)
n
2π n
12 Convexity, colours and statistics
Again, as the dimension grows the denominator grows faster than the numerator and
therefore the volume of a unit ball is small when the dimension is high. We can turn
this around if we like: a ball of unit volume has a large radius if the dimension is
high. Indeed since the volume is proportional to rn , where r is the radius, it follows
√
that the radius of a ball of unit volume grows like n when Stirling’s formula
applies.
The fraction of the volume of a unit ball that lies inside a radius r is rn . We
assume r < 1, so this is a quickly shrinking fraction as n grows. The curious
conclusion of this is that when the dimension is high almost all of the volume
of a ball lies very close to its surface. In fact this is a crucial observation in sta-
tistical mechanics. It is also the key property of n-dimensional geometry: when
n is large the ‘amount of space’ available grows very fast with distance from the
origin.
In some ways it is easier to see what is going on if we consider hypercubes n
rather than balls. Take a cube of unit volume. In n dimensions it has 2n corners,
and the longest straight line that we can√ draw inside the hypercube connects two
√
opposite corners. It has length L = 1 + . . . + 1 = n. Or expressed in
2 2
another way, a straight line of any length fits into a hypercube of unit volume
if the dimension is large enough. The reason why the longest line segment fit-
ting into the cube is large is clearly that we normalised the volume to one. If
√
we normalise L = 1 instead we find that the volume goes to zero like (1/ n)n .
Concerning the insphere – the largest inscribed sphere, with inradius rn – and the
outsphere – the smallest circumscribed sphere, with outradius Rn – we observe
that
√
n √
Rn = = nrn . (1.25)
2
√
The ratio between the two grows with the dimension, ζn ≡ Rn /rn = n. Inciden-
tally, the somewhat odd statement that the volume of a sphere goes to zero when
the dimension n goes to infinity can now be interpreted: since vol(n ) = 1 the real
statement is that vol(Sn )/vol(n ) goes to zero when n goes to infinity.
Now we turn to simplices, whose properties will be of some importance later
on. We concentrate on regular simplices n , for which the distance between any
pair of corners is one. For n = 1 this is the unit interval, for n = 2 a regular
triangle, for n = 3 a regular tetrahedron, and so on. Again we are interested
in the volume, the radius rn of the insphere, and the radius Rn of the outsphere.
We will also compute χn , the angle between the lines from the ‘centre of mass’
to a pair of corners. For a triangle it is arccos(−1/2) = 2π/3 = 120 degrees,
but it drops to arccos(−1/3) ≈ 110◦ for the tetrahedron. A practical way to go
about all this is to think of n as a pyramid (a part of a cone, namely the convex
1.2 High dimensional geometry 13
Figure 1.9 Regular simplices in two, three and four dimensions. For 2 we also
show the insphere, the circumsphere, and the angle discussed in the text.
hull of its apex and its base) having n−1 as its base. It is then not difficult to
show that
n 1
Rn = nrn = and rn = , (1.26)
2(n + 1) 2(n + 1)n
so their ratio grows linearly, ζn = Rn /rn = n. The volume of a pyramid is V =
Bh/n, where B is the area of the base, h is the height of the pyramid and n is the
dimension. For the simplex we obtain
1 n+1
vol(n ) = (1.27)
n! 2n
We can check that the ratio of the volume of the largest inscribed sphere to the
volume of the simplex goes to zero. Hence most of the volume of the simplex sits
in its corners, as expected. The angle χn subtended by an edge as viewed from the
centre is given by
χn 1 n+1 1
sin = = ⇔ cos χn = − . (1.28)
2 2Rn 2n n
When n is large we see that χn tends to a right angle. This is as it should be. The
corners sit on the outsphere, and for large n almost all the volume of the outsphere
lies close to the equator – hence, if we pick one corner and let it play the role of
the north pole, all the other corners are likely to lie close to the equator. Finally it is
interesting to observe that it is known for convex bodies in general that the radius
of the outsphere is bounded by
n
Rn ≤ L , (1.29)
2(n + 1)
where L is the length of the longest line segment L contained in the body. The
regular simplex saturates this bound.
Another random document with
no related content on Scribd:
PLATE LIII
“When using the ‘trumpet,’ the operator blows into the end having the smaller
diameter, with a vibratory motion of the lips....”
“arra” (one).
“kwiarra” (two).
“kwiarra arra” (three).
“kwiarra kwiarra” (four).
“kulno,” one.
“mandru,” two.
“parkulu” or “parkulintja,” three.
“mandru mandru,” four.
“mandru ja parkulintja,” five.
or “marra warra kulno,” five, i.e. hand part one (one hand).
“marra pirri kulno,” six.
“marra pirri mandru,” seven.
“marra pirri parkulintja,” eight.
“marra pirri mandru mandru,” nine.
“marra warra mandru” ten, i.e. hand parts two (two hands).
“tjinna pirri kulno,” eleven.
“marrapu,” many.
“mörla marrapu,” very many.
“ja” = and.
“marra” = hand.
“tjinna” = foot.
“warra” = part.
“pirri” = nail (toe or finger).
Thus the reading of the numerals runs: One, two, three, two-two
(four), two and three (five), or one hand part (five), hand finger-nails
one (six), hand finger-nails two (seven), hand finger-nails three
(eight), hand finger-nails four (nine), two hand parts (ten), foot toe-
nails one (eleven).
To imply a repetition or continuance of action, the frequent use of
the same verb is avoided by the Sunday Islanders, but the same
effect is achieved by reiterating the word “garra” indefinitely. The
expression of continued action is usually in the present tense and is
mostly applied to words like “running,” “walking,” “jumping,”
“sleeping,” “raining,” “blowing,” “bleeding,” etc., etc. For example, a
man from a tree or other look-out might be describing to his
companions below, the doings of their hunting party out on the plain;
while the latter keep moving, he conveys the fact to his hearers by
ejaculating “garra, garra, garra, ...” The direction of the hunters’
movements is indicated by the swing of his hand. The moment the
observer in the tree stops saying “garra,” the people below know that
the hunters have ducked or have temporarily ceased the pursuit.
The terminal syllable of a word is never uttered very clearly by an
aboriginal; and it is consequently difficult to distinguish between a
short e, a, o, or u. This is particularly true of the language used by
the old men, the defectiveness being often increased by the gaps
between the incisors resulting from the initiation ceremonies.
There are occasions, however, when the last syllable receives
especial emphasis. The Wongapitcha word for plenty is “ura”; when
the idea of plenty is great the word becomes “ura-ku,” with an
intentional intonation upon the “ku”; and when the plenty is
extraordinary, the word receives yet another syllable and becomes
the superlative “ura-ku-pu.”
The simple affirmative of the Aluridja, which has also been
adopted by the Arunndta, is “o,” less commonly “u”; but when strong
affirmation is intended the word is changed to “owa,” or “owau.”
The ordinary negative of the Arunndta is “itja”; but forcibly
expressed this may become “itjama” or “itjingalai.”
Any sudden exclamation or accidental noise, like a cough or
sneeze, might be exaggerated by the addition of a syllable at the
end. When a Sunday Islander sneezes he makes a word out of the
natural noise sounding like “Tish-e!”
An exclamation which is common practically to all Australian
tribes, and may express surprise, fear, pain, or disgust, is a very
liquid “irr.” The Arunndta have strengthened this monosyllabic cry by
giving it the suffix “ai”; nowadays the word is, however, mostly
pronounced “Yerrai!” A variety of the last-named, but not quite so
forceful, is “Yakai!” A sudden fright or the anticipation of harm might
produce a very short “i,” without any indication of the presence of the
liquid consonant.
Appreciation is indicated by “Aha!” or “Hm-hm!” practically
throughout the central and northern districts of Australia; in both
cases special emphasis is applied to the second syllable.
A central Australian exclamation calling one to order or attention in
a somewhat harsh manner is “Wai!” When one person is being
irritated by another, such as a parent by a whining child, the offender
is thus rebuked. This word may also become a suffix; it may be
combined, for instance, with the radical “irr” and produce a word
“Yirrawai,” which is perhaps the strongest in a sense of disgust and
reproach available in the Arunndta tongue.
Standing at the end of an adjective, the expression “’n-dora” in the
same dialect signifies a great or superlative degree; e.g. “marra”
means “nice,” but “marr’n-dora” a conception more like “excellent.” In
the opposite sense, “kurrina” (bad) becomes “kurrina’n-dora” (worst).
The same suffix can be attached to an adverb. When, say, an
emphatic denial is needed, “itja” (no) takes the form of “itja’n-dora.”
There are numerous other syllables, which, when fixed to the end
of a word, convey a special significance. When, in the Arunndta, the
syllable “tu” is added to a person’s name in address, it really stands
in place of a sentence like “Is it not so?” For instance, “Nani knulia
atoa utnuriraka, Endola-tu?” when literally translated reads: “This
dog man bit Endola, is it not so?” Endola is the name of a woman
who is being addressed. In place of “tu,” the longer form of “ditjekwi”
might be used.
Again, the suffix “lo” is found in daily use in the same dialect; it
stands for the phrase: “Where is?” Hence the completed sentence,
“Kwatche-lo?” stands for “Where is the water?”
When one finds “jara” added to a word, plurality is indicated, the
sense conveyed being that many of the kind specified by the noun
are assumed. The word for girl in the Arunndta is “kware,”
consequently “kwarenjara” means that a number of girls are being
considered, the “n” being simply interposed for the sake of euphony.
Any proper noun, like the name of a person, may be modified by
adding “ia” to it, and, by so doing, one makes it a term of
endearment in the vocative case. “Ware” ordinarily means “boy,” but
by altering it to “waria” (i.e. “ware-ia”), the meaning becomes “dear
boy” or “O boy.”
If the terminal “a” of a substantive is found to be changed to “inna,”
the diminutive of the original is implied. A somewhat common name
for a woman in the Arunndta is “Unnruba,” but during the years of
childhood of a bearer of this name, the appellation is always in the
form of “Unnrubinna”; in later years this changes to “Unnruba” more
or less automatically.
A diminutive sense is also conveyed by duplicating an original
word. In the eastern Aluridja dialect “kaitji” means an ordinary spear,
whereas “kaitji-kaitji” is an expression applied to a toy spear. The
Dieri recognize a Supreme Being whom they call “Mura”; any one of
their numerous demi-gods, however, from whom they trace their
descent is referred to as a “Mura-Mura.”
PLATE LV
“nunjaluki,” I break.
“minjaluki,” you break.
“iljaluki” or “n’jaluki,” he breaks.
Genitive, “-ba.”
Dative, “-lukuru.”
Accusative, “-na.”
Ablative, “-languru” (from), or “-lawana” (with).
The aspirate is very occasionally met with. The local group of the
Arunndta, residing at Arltunga, which styles itself “Herrinda,” is one
of the few exceptions. Even in the settled districts, where the natives
have adopted European names, one christened “Harry” is usually
spoken of as “Yarri.” The sounds which are most commonly
aspirated are the exclamations, e.g. “Hai!” “Aha,” and “Hm-hm!”
Used in conjunction with a vowel, usually an “a,” the consonantal
combination “ng” is common to all Australian dialects. Although to
the modern tongue a little difficult to master, the sound is strikingly
fluent with the aboriginal; indeed, its frequent inclusion in words of
fundamentally simple origin suggests a primitive, natural derivation.
Along the north-western coast, the words for “water” are “ngawa”
(Wave Hill), “ngauwa” (Humbert River), “kornga” (Sunday Island).
In the Arunndta, “nga” stands for either “Here you are!” “There!” or
“Take it!” all phrases being used in the sense of offering something to
a person, such as mother to infant. “Ngaboni” is a modified form of
the last-named, having any of the following meanings: “Here!”
“Look!” or “Behold!” And “ngarai” is yet another modification of
similar significance, usually placed immediately behind the object it
is desired to draw attention to, e.g. “Arre ngarai!” (Look at the
kangaroo).
Verbal greetings, akin to our “Good-day!” are rarely, if ever, made
use of by natives when they meet; but some of the tribes recognize
an orthodox term of salutation at partings, as indicating a friendly
feeling which the speaker bears towards the person he is on the
point of leaving. The Mulluk-Mulluk and some of their neighbours on
the Daly River have adopted the word “mummuk” to express their
farewell, whilst the Sunday Islanders’ vocabulary includes a similar
expression in the word “tchorrogu.”
An alarm, such as we would raise in the form of “Look out!” or “Get
out of the way!” is rendered by the Arunndta “Par-le!”
There are numerous examples which could be mentioned of words
appearing in an aboriginal dialect which have a striking resemblance
to European words of modern and ancient languages, which at this
stage of our knowledge of primitive tongues must be recorded as
curiosities rather than linguistic analogies. One of the most
remarkable, perhaps, is the Arunndta for head, viz. “kaputa”; but to
connect this in any way with the Latin word would be as
unreasonable as calling the “patriarchal” type of aboriginal a semite.
It is, therefore, not my intention to suggest anything beyond mere
coincidence, and it lies far from my mind to attempt theorizing upon
little evidence, yet it must be admitted that similarities in points of
evolution and culture, no matter how trivial and accidental they may
be, are not without interest. Viewed entirely in this light, there is no
harm in mentioning one or two similarities which exist between
certain words of the Australian dialects and those of other tongues,
articulate or otherwise.
Affirmation is expressed in the Cambridge Gulf district by “yau,”
“yo-au,” or “yowai.” During latter years, the same expression has
found its way south, even as far as the MacDonnell Ranges, where it
has largely replaced the original form of “owa.” In the adjoining
coastal districts of the Northern Territory, a similar word is in use as
well as one which sounds more like the German “ja.” It is of more
than passing interest to note that throughout the Malay Archipelago
“yo-au” or “yowai” is the principal form of affirmation in use among
the local primitive peoples. It is not unlikely, therefore, that the early
Macassan trepang fishers, who used to visit the north coast of
Australia long before the white man arrived, brought this word along
with them.
We find, also, that the consonantal expression, “ng,” combined
with a vowel sound, is common among other primitive languages.
Professor Garner, too, has found it included in the inarticulate
“speech” of apes in the form of a note indicating satisfaction, which
can be represented by “ngkw-a.” The speculative mind naturally
wonders whether this simian cry of satisfaction is perhaps
fundamentally of similar, though independent, phonetic origin to the
northern Australian word for water, the essential of life, which we
noted above is “ngawa.”
The di-syllabic muttering of a European child-in-arms, moreover,
which the happy parents flatter themselves sounds like “pa-pa” or
“ma-ma,” is equally characteristic of the aboriginal babe, but the
parents in the latter case interpret the note to be “nga-nga.” And,
indeed, the Arunndta mother responds by handing the infant the fluid
it needs, while she caressingly soothes it with the simple little word
“Nga,” which in our own language would be equivalent to the
sentence, “Here you are!”
INDEX
Roman numerals refer to Preface.
A
abris, 322
Acacia salicina, 157
acacia seed, 150, 151
Adelaide tribes, xii, 100, 102, 180, 195, 204, 206, 208, 212, 214,
360, 366
administration, tribal, 225 et seq.
adzes, 366, 367
Affenspalte, 38
“alangua” (native pear), Knaninja, 353
albinism, 48
“Algerrigiowumma,” sky-shying act, 238, 239
Alligator River, 197, 249
allotment of infants, 221, 222
“Altjerra,” 279, 295, 341, 342
“Altjerrajara,” Supreme Number, 271, 296
“Altjerringa,” 279, 296, 342
“Aluggurra” women, 343
Aluridja tribe, xii, 4, 47, 48, 50, 65, 78, 82, 83, 92, 95, 111, 114,
121, 147, 149, 150, 151, 153, 155, 170, 176, 178, 184, 192,
200, 205, 213, 226, 231, 236, 239, 240, 243, 255, 259, 264,
267, 269, 271, 272, 282, 295, 296, 349, 356, 361, 362, 387,
396, 398, 400, 402
amputation of finger joints, 253, 254
ancestor worship, 257
animation, in art, 320, 336, 337, 338
animal tracks, drawn in sand, 70-73
anthill burial, 206;
of bone, 214
Anthistiria (kangaroo grass), 276, 387
anthropomorphous designs, 353-358
“Antjuarra,” tooth-rapping ceremony, 235
anvil-stone, 369
approaching a stranger, 2, 3;
a camp, 105
appreciation, expression of, 399
archer fish, ochre drawing of, 328
armistice, 188
Arnhem Land, 50, 196, 197
Arrabonna Tribe, 4, 200, 237, 269, 362
“Arrarra,” a circumcision ceremony, 239 et seq.
“arrera,” kangaroo, Knaninja, 352
“Arrolmolba” (phallus), 291
art of the aboriginal, xi, xii, 297-358
artificial colouring of body, hunting, 142;
in warfare, 184;
of emissaries at initiation, 254;
for cosmetic and ceremonial purposes, 324-326
artificial warmth applied to infant, 66
Arunndta Tribe, xi, xii, 4, 6, 48, 50, 63, 65, 70, 76, 77, 78, 82,
83, 86, 92, 95, 103, 106, 114, 121, 125, 142, 147, 149, 150,
151, 152, 153, 155, 166, 170, 172, 175, 176, 178, 184, 192,
200, 205, 214, 218, 220, 226, 231, 237, 238, 263, 264, 271,
272, 274, 275, 282, 287, 291, 292, 295, 296, 308, 311, 321,
328, 330, 337, 338, 341, 343, 347, 349, 350, 352, 361, 362,
363, 386, 387, 390, 391, 398, 399, 400, 401, 402, 403, 404,
405
“Atoakwatje,” water-men, 264, 265
aspirate sounds, 403
attachment of parents to children, 65, 66
“aumba” (Brachysema), 153
Australoid, 58, 59;
migrations, 56, 57, 58
awl, stone, 365
Ayers Ranges, 75
B
bailers, 95
baldness, 50
ball-games, 77, 78
bandicoot, corrobboree of, 383
“banki,” prepuse, 243
bark, canoes, 160-164;
drawings, 323;
shield, 86, 87;
water-carriers, 92, 93
barramundi vertebræ used as spear heads, 198
barter, with ochre, 113;
with women, 222
bathing, 99
Bathurst Islanders, 76, 77, 93, 95, 96, 115, 159, 161, 163, 168,
207, 238, 309, 310, 319, 323, 373, 379, 380
beard, 50, 51, 117;
of female, 46
bees’ honey, 145
bell-bird, 386
belly, 20, 21
Berringin Tribe, 4, 11, 129, 130, 201, 374
beverages, 153
biological consanguinity between Australia and other continents,
55
bird-like attitude, 107
birds’ eggs, 125, 126
birth of aboriginal, 61-68
biting the initiate’s head, 244, 245
bleeding, stanching of, after circumcision, 243
blindness, following club hit, 37
blonde, aboriginal, 48, 49, 50
blood, drinking of, 154;
stilling the flow of, 186;
thrown at opponent during a duel, 166;
revenge, 187
Bloodwood apple, 152
Bloomfield River, 48
blue eye in aboriginal, 25
Blue Mountains, 57
boabab, carved tree, 309;
carved nuts, 311, 312, 313, 331;
nuts used as rattles, 374;
trees serving as reservoirs, 97, 98
“Bobi,” tjuringa, 270
body decoration, at completion of mourning, 214, 215;
cosmetic, artistic and ceremonial, 324, 325, 326.
See also artificial colouring of body
“Böllier” ceremony, initiation, 250
bone, carving of, 313;
“pointing” the, 174-178, 209
bone-pointed spear-thrower, 201
bones of dead man collected, 214;
radius selected for “pointing bone,” 215
“boning” to death, 174-178, 209
boomerang, practice, 86;
used as a musical instrument, 374, 383;
used in duels, 168;
used in warfare, 187, 188
“boomerang-legs,” 15
boomerang-spear duel, drawing of, 330, 338
boras, 377
boys, apportioned to old men, 65, 66;
taught use of weapons, 85, 86
Brachysema Chambersii, 153
brain of aboriginal, 37, 38
breast, 18-20;
artificial lactation of, 20;
development of, 19;
enchantment of, 19;
touched with charred mulga, 65
Bremer, Sir Gordon, 144
Buccaneer Archipelago, 50, 51
buffalo, drawing of, 325, 337;
hunt, 144, 145, and Frontispiece
bull-roarer, used at circumcision, 241, 242;
at mika operation, 246;
at tooth-rapping ceremony, 232, 233
burial and mourning customs, 203-215
burial ground, signpost erected at, 207
burial in anthill, 208
C
Calamus used as paint brush, 319, 320
Caledon Bay, 115
calvarium, Tennants Creek, 53;
used as a drinking vessel, 214