PDF Quantum Information Processing With Finite Resources Mathematical Foundations 1St Edition Marco Tomamichel Auth Ebook Full Chapter

Quantum Information Processing with
Finite Resources Mathematical

Foundations 1st Edition Marco
Tomamichel (Auth.)
Visit to download the full and correct content document:
https://textbookfull.com/product/quantum-information-processing-with-finite-resources
-mathematical-foundations-1st-edition-marco-tomamichel-auth/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...
Mathematical foundations of quantum mechanics Neumann
https://textbookfull.com/product/mathematical-foundations-of-
quantum-mechanics-neumann/
Quantum Information Processing, Quantum Computing, and

Quantum Error Correction: An Engineering Approach 2nd
Edition Ivan B. Djordjevic
https://textbookfull.com/product/quantum-information-processing-
quantum-computing-and-quantum-error-correction-an-engineering-
approach-2nd-edition-ivan-b-djordjevic/
Device-Independent Quantum Information Processing: A

Simplified Analysis 1st Edition Rotem Arnon-Friedman
https://textbookfull.com/product/device-independent-quantum-
information-processing-a-simplified-analysis-1st-edition-rotem-
arnon-friedman/
Spectral Theory and Quantum Mechanics Mathematical

Foundations of Quantum Theories Symmetries and
Introduction to the Algebraic Formulation Valter
Moretti
https://textbookfull.com/product/spectral-theory-and-quantum-
mechanics-mathematical-foundations-of-quantum-theories-
symmetries-and-introduction-to-the-algebraic-formulation-valter-
Foundations of Quantum Programming 1st Edition Ying
https://textbookfull.com/product/foundations-of-quantum-
programming-1st-edition-ying/
Introduction to the Theory of Quantum Information

Processing (Graduate Texts in Physics) 2013th Edition
Bergou
https://textbookfull.com/product/introduction-to-the-theory-of-
quantum-information-processing-graduate-texts-in-physics-2013th-
edition-bergou/
Mathematical Foundations of Computer Science 1st

Edition Satyanarayana
https://textbookfull.com/product/mathematical-foundations-of-
computer-science-1st-edition-satyanarayana/
Mathematical Image Processing Kristian Bredies
https://textbookfull.com/product/mathematical-image-processing-
kristian-bredies/
Quantum Physics: An Overview of a Weird World: A Guide

to the 21st Century Quantum Revolution 1st Edition
Marco Masi
https://textbookfull.com/product/quantum-physics-an-overview-of-
a-weird-world-a-guide-to-the-21st-century-quantum-revolution-1st-
edition-marco-masi/
SPRINGER BRIEFS IN MATHEMATIC AL PHYSICS 5
Marco Tomamichel
Quantum
Information
Processing with
Finite Resources
Mathematical
Foundations
123
SpringerBriefs in Mathematical Physics
Volume 5
Series editors
Nathanaël Berestycki, Cambridge, UK
Mihalis Dafermos, Cambridge, UK
Tohru Eguchi, Tokyo, Japan
Atsuo Kuniba, Tokyo, Japan
Matilde Marcolli, Pasadena, USA
Bruno Nachtergaele, Davis, USA
More information about this series at http://www.springer.com/series/11953
Marco Tomamichel
Quantum Information
Processing with Finite
Resources
Mathematical Foundations
123
Marco Tomamichel
School of Physics
The University of Sydney
Sydney
Australia
ISSN 2197-1757 ISSN 2197-1765 (electronic)

SpringerBriefs in Mathematical Physics
ISBN 978-3-319-21890-8 ISBN 978-3-319-21891-5 (eBook)
DOI 10.1007/978-3-319-21891-5
Library of Congress Control Number: 2015950017
Springer Cham Heidelberg New York Dordrecht London

© The Author(s) 2016
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained herein or
for any errors or omissions that may have been made.
Printed on acid-free paper
Springer International Publishing AG Switzerland is part of Springer Science+Business Media

(www.springer.com)
Acknowledgments
I was introduced to quantum information theory during my PhD studies in Renato

Renner’s group at ETH Zurich. It is from him that I learned most of what I know
about quantum cryptography and smooth entropies. Renato also got me interested
more generally in finite resource information theory as well as the entropies and
other information measures that come along with it. He shaped my views on the
topic and gave me all the tools I needed to follow my own ideas, and I am
immensely grateful for that. A particular question that grew out of my PhD studies
was whether one can understand smooth entropies in a broader framework of
quantum Rényi entropies. This book now accomplishes this.
Joining Stephanie Wehner’s group at the Centre for Quantum Technologies in
Singapore for my postdoctoral studies was probably the best thing I could have
done. I profited tremendously from interactions with Stephanie and her group about
many of the topics covered in this book. On top of that, I was also given the
freedom to follow my own interests and collaborate with great researchers in
Singapore. In particular I want to thank Masahito Hayashi for sharing a small part
of his profound knowledge of quantum statistics and quantum information theory
with me. On the other hand, Vincent Y.F. Tan taught me much of what I know
about classical information theory, and I greatly enjoyed talking to him about finite
resource effects.
Renato Renner, Mark M. Wilde, and Andreas Winter encouraged me to write
this book. It is my pleasure to thank Christopher T. Chubb and Mark M. Wilde for
carefully reading the manuscript and spotting many typos. I also want to thank
Rupert L. Frank, Elliott H. Lieb, Milán Mosonyi, and Renato Renner for many
insightful comments and suggestions on an earlier draft. While writing I also greatly
enjoyed and profited from scientific discussions with Mario Berta, Frédéric Dupuis,
Anthony Leverrier, and Volkher B. Scholz about different aspects of this book.
I also want to acknowledge support from a University of Sydney Postdoctoral
Fellowship as well as the Centre of Excellence for Engineered Quantum Systems in
Australia.
v
vi Acknowledgments
Last but not least, I want to thank my wife Thanh Nguyệt for supporting me
during this time, even when the writing continued long after the office hours ended.
I dedicate this book to the memory of my grandfathers, Franz Wagenbach and
Giuseppe B. Tomamichel, the engineers in the family.
Sydney Marco Tomamichel

Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Finite Resource Information Theory . . . . . . . . . . . . . . . . . . . . . 2
1.2 Motivating Example: Source Compression . . . . . . . . . . . . . . . . . 4
1.3 Outline of the Book . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2 Modeling Quantum Information . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2.1 General Remarks on Notation . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Linear Operators and Events . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.1 Hilbert Spaces and Linear Operators . . . . . . . . . . . . . . . . 13
2.2.2 Events and Measures . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3 Functionals and States. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3.1 Trace and Trace-Class Operators . . . . . . . . . . . . . . . . . . 19
2.3.2 States and Density Operators . . . . . . . . . . . . . . . . . . . . . 20
2.4 Multi-partite Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.4.1 Tensor Product Spaces . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.4.2 Separable States and Entanglement . . . . . . . . . . . . . . . . . 24
2.4.3 Purification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.4.4 Classical-Quantum Systems . . . . . . . . . . . . . . . . . . . . . . 25
2.5 Functions on Positive Operators . . . . . . . . . . . . . . . . . . . . . . . . 26
2.6 Quantum Channels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.6.1 Completely Bounded Maps . . . . . . . . . . . . . . . . . . . . . . 28
2.6.2 Quantum Channels . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.6.3 Pinching and Dephasing Channels . . . . . . . . . . . . . . . . . 30
2.6.4 Channel Representations . . . . . . . . . . . . . . . . . . . . . . . . 31
2.7 Background and Further Reading . . . . . . . . . . . . . . . . . . . . . . . 32
3 Norms and Metrics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

3.1 Norms for Operators and Quantum States . . . . . . . . . . . . . . . . . 33
3.1.1 Schatten Norms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.1.2 Dual Norm for States . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.2 Trace Distance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
vii
viii Contents
3.3 Fidelity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.3.1 Generalized Fidelity . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.4 Purified Distance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4 Quantum Rényi Divergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

4.1 Classical Rényi Divergence . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.1.1 An Axiomatic Approach . . . . . . . . . . . . . . . . . . . . . . . . 48
4.1.2 Positive Definiteness and Data-Processing . . . . . . . . . . . . 49
4.1.3 Monotonicity in α and Limits . . . . . . . . . . . . . . . . . . . . 51
4.2 Classifying Quantum Rényi Divergences . . . . . . . . . . . . . . . . . . 53
4.2.1 Joint Concavity and Data-Processing. . . . . . . . . . . . . . . . 54
4.2.2 Minimal Quantum Rényi Divergence . . . . . . . . . . . . . . . 55
4.2.3 Maximal Quantum Rényi Divergence . . . . . . . . . . . . . . . 56
4.2.4 Quantum Max-Divergence . . . . . . . . . . . . . . . . . . . . . . . 56
4.3 Minimal Quantum Rényi Divergence . . . . . . . . . . . . . . . . . . . . 58
4.3.1 Pinching Inequalities. . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.3.2 Limits and Special Cases. . . . . . . . . . . . . . . . . . . . . . . . 62
4.3.3 Data-Processing Inequality. . . . . . . . . . . . . . . . . . . . . . . 63
4.4 Petz Quantum Rényi Divergence . . . . . . . . . . . . . . . . . . . . . . . 67
4.4.1 Data-Processing Inequality. . . . . . . . . . . . . . . . . . . . . . . 67
4.4.2 Nussbaum–Szkoła Distributions . . . . . . . . . . . . . . . . . . . 68
5 Conditional Rényi Entropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

5.1 Conditional Entropy from Divergence . . . . . . . . . . . . . . . . . . . . 73
5.2 Definitions and Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.2.1 Alternative Expression for H α" . . . . . . . . . . . . . . . . . . . . 77
5.2.2 Conditioning on Classical Information . . . . . . . . . . . . . . 78
5.2.3 Data-Processing Inequalities and Concavity . . . . . . . . . . . 79
5.3 Duality Relations and Their Applications. . . . . . . . . . . . . . . . . . 80
5.3.1 Duality Relation for H α# . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.3.2 Duality Relation for H e α" . . . . . . . . . . . . . . . . . . . . . . . . 81

5.3.3 Duality Relation for Hα" and H e α# . . . . . . . . . . . . . . . . . . . 82
5.3.4 Additivity for Tensor Product States . . . . . . . . . . . . . . . . 83
5.3.5 Lower and Upper Bounds on Quantum Rényi Entropy . . . 84
5.4 Chain Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
6 Smooth Entropy Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

6.1 Min- and Max-Entropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
6.1.1 Semi-Definite Programs . . . . . . . . . . . . . . . . . . . . . . . . 91
6.1.2 The Min-Entropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
Contents ix
6.1.3 The Max-Entropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

6.1.4 Classical Information and Guessing Probability . . . . . . . . 97
6.2 Smooth Entropies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
6.2.1 Definition of the ε-Ball . . . . . . . . . . . . . . . . . . . . . . . . . 98
6.2.2 Definition of Smooth Entropies . . . . . . . . . . . . . . . . . . . 99
6.2.3 Remarks on Smoothing . . . . . . . . . . . . . . . . . . . . . . . . . 100
6.3 Properties of the Smooth Entropies . . . . . . . . . . . . . . . . . . . . . . 102
6.3.1 Duality Relation and Beyond . . . . . . . . . . . . . . . . . . . . . 102
6.3.2 Chain Rules. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
6.3.3 Data-Processing Inequalities. . . . . . . . . . . . . . . . . . . . . . 105
6.4 Fully Quantum Asymptotic Equipartition Property . . . . . . . . . . . 107
6.4.1 Lower Bounds on the Smooth Min-Entropy. . . . . . . . . . . 107
6.4.2 The Asymptotic Equipartition Property . . . . . . . . . . . . . . 110
7 Selected Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

7.1 Binary Quantum Hypothesis Testing . . . . . . . . . . . . . . . . . . . . . 115
7.1.1 Chernoff Bound . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
7.1.2 Stein’s Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
7.1.3 Hoeffding Bound and Strong Converse Exponent . . . . . . . 118
7.2 Entropic Uncertainty Relations . . . . . . . . . . . . . . . . . . . . . . . . . 118
7.2.1 Tripartite Uncertainty Relation . . . . . . . . . . . . . . . . . . . . 119
7.2.2 Bipartite Uncertainty Relation . . . . . . . . . . . . . . . . . . . . 120
7.3 Randomness Extraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
7.3.1 Uniform and Independent Randomness . . . . . . . . . . . . . . 121
7.3.2 Direct Bound: Leftover Hash Lemma . . . . . . . . . . . . . . . 123
7.3.3 Converse Bound. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
Appendix: Some Fundamental Results in Matrix Analysis . . . . . . . . . . 127
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
Chapter 1
Introduction
Abstract This chapter motivates the study of finite resource quantum information
theory and the mathematical framework that is required to do so. We will present a
motivating example and outline the content of the book.
As we further miniaturize information processing devices, the impact of quantum

effects will become more and more relevant. Information processing at the micro-
scopic scale poses challenges but also offers various opportunities: How much infor-
mation can be transmitted through a physical communication channel if we can
encode and decode our information using a quantum computer? How can we take
advantage of entanglement, a form of correlation stronger than what is allowed by
classical physics? What are the implications of Heisenberg’s uncertainty principle
of quantum mechanics for cryptographic security? These are only a few amongst the
many questions studied in the emergent field of quantum information theory.
One of the predominant challenges when engineering future quantum information
processors is that large quantum systems are notoriously hard to maintain in a coher-
ent state and difficult to control accurately. Hence, it is prudent to expect that there
will be severe limitations on the size of quantum devices for the foreseeable future.
It is therefore of immediate practical relevance to investigate quantum information
processing with limited physical resources, for example, to ask:
How well can we perform information processing tasks if we only have access
to a small quantum device? Can we beat fundamental limits imposed on infor-
mation processing with non-quantum resources?
This book will introduce the reader to the mathematical framework required to
answer such questions, and many others. In quantum cryptography we want to show
that a key of finite length is secret from an adversary, in quantum metrology we want
to infer properties of a small quantum system from a finite sample, and in quan-
tum thermodynamics we explore the thermodynamic properties of small quantum
© The Author(s) 2016 1

M. Tomamichel, Quantum Information Processing with Finite Resources,
SpringerBriefs in Mathematical Physics, DOI 10.1007/978-3-319-21891-5_1
2 1 Introduction
systems. What all these applications have in common is that they concern properties
of small quantum devices and require precise statements that remain valid outside
asymptopia—the idealized asymptotic regime where the system size is unbounded.
1.1 Finite Resource Information Theory
Through the lens of a physicist it is natural to see Shannon’s information theory [144]
as a resource theory. Data sources and communication channels are traditional exam-
ples of resources in information theory, and its goal is to investigate how these
resources are interrelated and how they can be transformed into each other. For
example, we aim to compress a data source that contains redundancy into one that
does not, or to transform a noisy channel into a noiseless one. Information theory
quantifies how well this can be done and in particular provides us with fundamental
limits on the best possible performance of any transformation.
Shannon’s initial work [144] already gives definite answers to the above example
questions in the asymptotic regime where resources are unbounded. This means that
we can use the input resource as many times as we wish and are interested in the
rate (the fraction of output to input resource) at which transformations can occur.
The resulting statements can be seen as a first approximation to a more realistic
setting where resources are necessarily finite, and this approximation is indeed often
sufficient for practical purposes.
However, as argued above, specifically when quantum resources are involved we
would like to establish more precise statements that remain valid even when the
available resources are very limited. This is the goal of finite resource information
theory. The added difficulty in the finite setting is that we are often not able to produce
the output resource perfectly. The best we can hope for is to find a tradeoff between
the transformation rate and the error we allow on the output resource. In the most
fundamental one-shot setting we only consider a single use of the input resource and
are interested in the tradeoff between the amount of output resource we can produce
and the incurred error. We can then see the finite resource setting as a special case of
the one-shot setting where the input resource has additional structure, for example
a source that produces a sequence of independent and identically distributed (iid)
symbols or a channel that is memoryless or ergodic.
Notably such considerations were part of the development of information the-
ory from the outset. They motivated the study of error exponents, for example by
Gallager [63]. Roughly speaking, error exponents approximate how fast the error van-
ishes for a fixed transformation rate as the number of available resources increases.
However, these statements are fundamentally asymptotic in nature and make strong
assumptions on the structure of the resources. Beyond that, Han and Verdú estab-
lished the information spectrum method [69, 70] which allows to consider unstruc-
tured resources but is asymptotic in nature. More recently finite resource information
theory has attracted considerable renewed attention, for example due to the works of
Hayashi [77, 78] and Polyanskiy et al. [133]. The approach in these works—based
1.1 Finite Resource Information Theory 3
on Strassen’s techniques [148]—is motivated operationally: in many applications we

can admit a small, fixed error and our goal is to find the maximal possible transfor-
mation rate as a function of the error and the amount of available resource.1
In an independent development, approximate or asymptotic statements were also
found to be insufficient in the context of cryptography. In particular the advent of
quantum cryptography [18, 51] motivated a precise information-theoretic treatment
of the security of secret keys of finite length [99, 139]. In the context of quantum
cryptography many of the standard assumptions in information theory are no longer
valid if one wants to avoid any assumptions on the eavesdropper’s actions. In partic-
ular, the common assumption that resources are iid or ergodic is hardly justified. In
quantum cryptography we are instead specifically interested in the one-shot setting,
where we want to understand how much (almost) secret key can be extracted from a
single use of an unstructured resource.
The abstract view of finite resource information theory as a resource theory also
reveals why it has found various applications in physical resource theories, most
prominently in thermodynamics (see, e.g., [30, 47, 52] and references therein).
Rényi and Smooth Entropies
The main focus of this book will be on various measures of entropy and informa-
tion that underly finite resource information theory, in particular Rényi and smooth
entropies. The concept of entropy has its origins in physics, in particular in the works
of Boltzmann [28] and Gibbs [66] on thermodynamics. Von Neumann [170] gen-
eralized these concepts to quantum systems. Later Shannon [144]—well aware of
the origins of entropy in physics—interpreted entropy as a measure of uncertainty
of the outcome of a random experiment. He found that entropy, or Shannon entropy
as it is called now in the context of information theory,2 characterizes the optimal
asymptotic rate at which information can be compressed. However, we will soon see
that it is necessary to consider alternative information measures if we want to move
away from asymptotic statements.
Error exponents can often be expressed in terms of Rényi entropies [142] or
related information measures, which partly explains the central importance of this
one-parameter family of entropies in information theory. Rényi entropies share many
mathematical properties with the Shannon entropy and are powerful tools in many
information-theoretic arguments. A significant part of this book is thus devoted to
exploring quantum generalizations of Rényi entropies, for example the ones proposed
by Petz [132] and a more recent specimen [122, 175] that has already found many
applications.
The particular problems encountered in cryptography led to the development of
smooth entropies [141] and their quantum generalizations [139, 140]. Most impor-
tantly, the smooth min-entropy captures the amount of uniform randomness that
can be extracted from an unstructured source if we allow for a small error. (This
1 Thetopic has also been reviewed recently by Tan [151].

2 Notwithstanding the historical development, we follow the established tradition and use Shannon
entropy to refer to entropy. We use von Neumann entropy to refer to its quantum generalization.
4 1 Introduction
example is discussed in detail in Sect. 7.3.) The smooth entropies are variants of
Rényi entropies and inherit many of their properties. They have since found various
applications ranging from information theory to quantum thermodynamics and will
be the topic of the second part of this book.
We will further motivate the study of these information measures with a simple
example in the next section.
Besides their operational significance, there are other reasons why the study of
information measures is particularly relevant in quantum information theory. Many
standard arguments in information theory can be formulated in term of entropies, and
often this formulation is most amenable to a generalization to the quantum setting. For
example, conditional entropies provide us with a measure of the uncertainty inherent
in a quantum state from the perspective of an observer with access to side information.
This allows us to circumvent the problem that we do not have a suitable notion
of conditional probabilities in quantum mechanics. As another example, arguments
based on typicality and the asymptotic equipartition property can be phrased in terms
of smooth entropies which often leads to a more concise and intuitive exposition.
Finally, the study of quantum generalizations of information measures sometimes
also gives new insights into the classical quantities. For example, our definitions
and discussions of conditional Rényi entropy also apply to the classical special case
where such definitions have not yet been firmly established.
1.2 Motivating Example: Source Compression
We are using notation that will be formally introduced in Chap. 2 and concepts that
will be expanded on in later chapters (cf. Table 1.1). A data source is described
probabilistically as follows. Let X be a random variable with distribution ρ X (x) =
Pr[X = x] that models the distribution of the different symbols that the source emits.
The number of bits of memory needed to store one symbol produced by this source so
that it can be recovered with certainty is given by H0 (X )ρ , where H0 (X )ρ denotes
the Hartley entropy [72] of X , defined as

H0 (X )ρ = log2 {x : ρ X (x) > 0}. (1.1)
The Hartley entropy is a limiting case of a Rényi entropy [142] and simply measures
the cardinality of the support of X . In essence, this means that we can ignore symbols
Table 1.1 Reference to detailed discussion of the quantities and concepts mentioned in this section
Concept To be discussed further in
Hα Rényi entropy Chapters 4 and 5
Δ(·, ·) Variational distance Section 3.1, as generalized trace distance
ε
Hmax Smooth Rényi entropy Chapter 6, as smooth max-entropya
Entropic AEP Section 6.4, entropic asymptotic
equipartition property
a We will use a different metric for the definition of the smooth max-entropy
1.2 Motivating Example: Source Compression 5
that never occur but otherwise our knowledge of the distribution of the different
symbols does not give us any advantage.
As an example, consider a source that outputs lowercase characters of the English
alphabet. If we want to store a single character produced by this source such that it
can be recovered with certainty, we clearly need log2 26 = 5 bits of memory as a
resource.
Analysis with Rényi Entropies
More interestingly, we may ask how much memory we need to store the output of
the source if we allow for a small probability of failure, ε ∈ (0, 1). To answer this
we investigate encoders that assign codewords of a fixed length log2 m (in bits) to
the symbols the source produces. These codewords are then stored and a decoder is
later used to compute an estimate of X from the codewords. If the probability that
this estimate equals the original symbol produced by the source is at least 1 − ε, then
we call such a scheme an (ε, m)-code. For a source X with probability distribution
ρ X , we are thus interested in finding the tradeoff between code length, log2 m, and
the probability of failure, ε, for all (ε, m)-codes.
Shannon in his seminal work [144] showed that simply disregarding the most
unlikely source events (on average) leads to an arbitrarily small failure probability
if the code length is chosen sufficiently long. In particular, Gallager’s proof [63, 64]
implies that (ε, m)-codes always exist as long as
α 1 1
log2 m ≥ Hα (X )ρ + log2 for some α ∈ ,1 . (1.2)
1−α ε 2
Here, Hα (X )ρ is the Rényi entropy of order α, defined as

1 α
Hα (X )ρ = log2 ρ X (x) . (1.3)
1−α x
For all α ∈ (0, 1) ∪ (1, ∞) and as the respective limit for α ∈ {0, 1, ∞}. The Rényi
entropies are monotonically decreasing in α. Clearly the lower bound in (1.2) thus
constitutes a tradeoff: larger values of the order parameter α lead to a smaller Rényi
α
entropy but will increase the penalty term 1−α log2 1ε . Statements about the existence
of codes as in (1.2) are called achievability bounds or direct bounds.
This analysis can be driven further if we consider sources with structure. In partic-
ular, consider a sequence of sources that produce n ∈ N independent and identically
distributed (iid) symbols X n = (Y1 , Y2 , . . . , Yn ), where each Yi is distributed accord-
ing to the law τY (y). We then consider a sequence of (ε, 2n R )-codes for these sources,
6 1 Introduction
where the rate R indicates the number of memory bits required per symbol the source
produces. For this case (1.2) reads
1 α 1 α 1
R≥ Hα (X n )ρ + log2 = Hα (Y )τ + log2 (1.4)
n n(1 − α) ε n(1 − α) ε
where we used additivity of the Rényi entropy to establish the equality. The above
inequality implies that such a sequence of (ε, 2n R )-codes exists for sufficiently large
n if R > Hα (X )ρ . And finally, since this holds for all α ∈ [ 21 , 1), we may take the
limit α → 1 in (1.4) to recover Shannon’s original result [144], which states that
such codes exists if

R > H (X )ρ , where H (X )ρ = H1 (X )ρ = − ρ X (x) log2 ρ X (x) (1.5)
x
is the Shannon entropy of the source. This rate is in fact optimal, meaning that
every scheme with R < H (X )ρ necessary fails with certainty as n → ∞. This is an
example of an asymptotic statement (with infinite resources) and such statements can
often be expressed in terms of the Shannon entropy or related information measures.
Analysis with Smooth Entropies
Another fruitful approach to analyze this problem brings us back to the unstructured,
one-shot case. We note that the above analysis can be refined without assuming any
structure by “smoothing” the entropy. Namely, we construct an (ε, m) code for the
source ρ X using the following recipe:
• Fix δ ∈ (0, ε) and let ρ̃ X be any probability distribution that is (ε − δ)-close to ρ X
in variational distance. Namely we require that Δ(ρ̃ X , ρ X ) ≤ ε − δ where Δ(·, ·)
denotes the variational distance.
• Then, take a (δ, m)-code for the source ρ̃ X . Instantiating (1.2) with α = 21 , we
find that there exists such a code as long as log2 m ≥ H1/2 (X )ρ̃ + log2 1δ .
• Apply this code to a source with the distribution ρ X instead, incurring a total error
of at most δ + Δ(ρ X , ρ̃ X ) ≤ ε. (This uses the triangle inequality and the fact
that the variational distance contracts when we process information through the
encoder and decoder.)
Hence, optimizing this over all such ρ̃ X , we find that there exists a (ε, m)-code if
ε−δ 1 ε
log2 m ≥ Hmax (X )ρ + log2 , where Hmax (X )ρ := min H1/2 (X )ρ̃
δ ρ̃ X :Δ(ρ X ,ρ̃ X )≤ε
(1.6)
is the ε -smooth max-entropy, which is based on the Rényi entropy of order 21 .

Furthermore, this bound is approximately optimal in the following sense. It can
ε (X ) . Such bounds
be shown [138] that all (ε, m)-codes must satisfy log2 m ≥ Hmax ρ
1.2 Motivating Example: Source Compression 7
that give restrictions valid for all codes are called converse bounds. Rewriting this,
we see that the minimal value of m for a given ε, denoted m ∗ (ε), satisfies
1
ε ε−δ
Hmax (X )ρ ≤ log2 m ∗ (ε) ≤ inf Hmax (X )ρ + log2 . (1.7)
δ∈(0,ε) δ
We thus informally say that the memory required for one-shot source compression
is characterized by the smooth max-Rényi entropy.3
Finally, we again consider the case of an iid source, and as before, we expect that
in the limit of large n, the optimal compression rate n1 m ∗ (ε) should be characterized
by the Shannon entropy. This is in fact an expression of an entropic version of the
asymptotic equipartition property, which states that
1 ε
lim H (X n )ρ = H (Y )τ for all ε ∈ (0, 1). (1.8)
n→∞ n max
Why Shannon Entropy is Inadequate
To see why the Shannon entropy does not suffice to characterize one-shot source
compression, consider a source that produces the symbol ‘’ with probability 1/2
and k other symbols with probability 1/2k each. On the one hand, for any fixed
failure probability ε 1, the converse bound in (1.7) evaluates to approximately
log2 k. This implies that we cannot compress this source much beyond its Hartley
entropy. On the other hand, the Shannon entropy of this distribution is 21 (log2 k + 2)
and underestimates the required memory by a factor of two.
1.3 Outline of the Book
The goal of this book is to explore quantum generalizations of the measures encoun-
tered in our example, namely the Rényi entropies and smooth entropies. Our expo-
sition assumes that the reader is familiar with basic probability theory and linear
algebra, but not necessarily with quantum mechanics. For the most part we restrict
our attention to physical systems whose observable properties are discrete, e.g. spin
systems or excitations of particles bound in a potential. This allows us to avoid mathe-
matical subtleties that appear in the study of systems with observable properties that
are continuous. We will, however, mention generalizations to continuous systems
where applicable and refer the reader to the relevant literature.
The book is organized as follows:
Chapter 2 introduces the notation used throughout the book and presents the
mathematical framework underlying quantum theory for general (potentially con-
tinuous) systems. Our notation is summarized in Table 2.1 so that the remainder of
3 Thesmoothing approach in the classical setting was first formally discussed in [141]. A detailed
analysis of one-shot source compression, including quantum side information, can be found in [138].
8 1 Introduction
the chapter can easily be skipped by expert readers. The exposition starts with intro-
ducing events as linear operators on a Hilbert space (Sect. 2.2) and then introduces
states as functionals on events (Sect. 2.3). Multi-partite systems and entanglement
is then discussed using the Hilbert space tensor product (Sect. 2.4) and finally
quantum channels are introduced as a means to study the evolution of systems in
the Schrödinger and Heisenberg picture (Sect. 2.6). Finally, this chapter assembles
the mathematical toolbox required to prove the results in the later chapters, includ-
ing a discussion of operator monotone, concave and convex functions on positive
operators (Sect. 2.5). Most results discussed here are well-known and proofs are
omitted. We do not attempt to provide an intuition or physical justification for
the mathematical models employed, but instead highlight some connections to
classical information theory.
Chapter 3 treats norms and metrics on quantum states. First we discuss Schatten
norms and a variational characterization of the Schatten norms of positive oper-
ators that will be very useful in the remainder of the book (Sect. 3.1). We then
move on to discuss a natural dual norm for sub-normalized quantum states and the
metric it induces, the trace distance (Sect. 3.2). The fidelity is another very promi-
nent measure for the proximity of quantum states, and here we sensibly extend
its to definition to cover sub-normalized states (Sect. 3.3). Finally, based on this
generalized fidelity, we introduce a powerful metric for sub-normalized quantum
states, the purified distance (Sect. 3.4). This metric combines the clear operational
interpretation of the trace distance with the desirable mathematical properties of
the fidelity.
Chapter 4 discusses quantum generalizations of the Rényi divergence. Diver-
gences (or relative entropies) are measures of distance between quantum states
(although they are not metrics) and entropy as well as conditional entropy can con-
veniently be defined in terms of the divergence. Moreover, the entropies inherit
many important properties from corresponding properties of the divergence. In
this chapter, we first discuss the classical special case of the Rényi divergence
(Sect. 4.1). This allows us to point out several properties that we expect a suitable
quantum generalization of the Rényi divergence to satisfy. Most prominently we
expect them to satisfy a data-processing inequality which states that the divergence
is contractive under application of quantum channels to both states. Based on this,
we then explore quantum generalizations of the Rényi divergence and find that
there is more than one quantum generalization that satisfies all desired properties
(Sect. 4.2).
We will mostly focus on two different quantum Rényi divergences, called the
minimal and Petz quantum Rényi divergence (Sects. 4.3 and 4.4). The first quan-
tum generalization is called the minimal quantum Rényi divergence (because it
is the smallest quantum Rényi divergence that satisfies a data-processing inequal-
ity), and is also known as “sandwiched” Rényi relative entropy in the literature.
It has found operational significance in the strong converse regime of asymmet-
ric binary hypothesis testing. The second quantum generalization is Petz’ quan-
tum Rényi relative entropy, which attains operational significance in the quantum
1.3 Outline of the Book 9
generalization of Chernoff’s and Hoeffding’s bound on the success probability in

binary hypothesis testing (cf. Sect. 7.1).
Chapter 5 generalizes conditional Rényi entropies (and unconditional entropies as
a special case) to the quantum setting. The idea is to define operationally relevant
measures of uncertainty about the state of a quantum system from the perspective
of an observer with access to some side information stored in another quantum
system. As a preparation, we discuss how the conditional Shannon entropy and
the conditional von Neumann entropy can be conveniently expressed in terms of
relative entropy either directly or using a variational formula (Sect. 5.1). Based on
the two families of quantum Rényi divergences, we then define four families of
quantum conditional Rényi entropies (Sect. 5.2). We then prove various properties
of these entropies, including data-processing inequalities that they directly inherit
from the underlying divergence. A genuinely quantum feature of conditional Rényi
entropies is the duality relation for pure states (Sect. 5.3). These duality relations
also show that the four definitions are not independent, and thereby also reveal a
connection between the minimal and the Petz quantum Rényi divergence. Further-
more, even though the chain rule does not hold with equality for our definitions,
we present some inequalities that replace the chain rule (Sect. 5.4).
Chapter 6 deals with smooth conditional entropies in the quantum setting. First,
we discuss the min-entropy and the max-entropy, two special cases of Rényi
entropies that underly the definition of the smooth entropy (Sect. 6.1). In particular,
we show that they can be expressed as semi-definite programs, which means that
they can be approximated efficiently (for small quantum systems) using standard
numerical solvers. The idea is that these two entropies serve as representatives
for the Rényi entropies with large and small α, respectively. We then define the
smooth entropies (Sect. 6.2) as optimizations of the min- and max-entropy over a
ball of states close in purified distance. We explore some of their properties, includ-
ing chain rules and duality relations (Sect. 6.3). Finally, the main application of
the smooth entropy calculus is an entropic version of the asymptotic equiparti-
tion property for conditional entropies, which states that the (regularized) smooth
min- and max-entropies converge to the conditional von Neumann entropy for iid
product states (Sect. 6.4).
Chapter 7 concludes the book with a few selected applications of the mathemati-
cal concepts surveyed here. First, we discuss various aspects of binary hypothesis
testing, including Stein’s lemma, the Chernoff bound and the Hoeffding bound as
well as strong converse exponents (Sect. 7.1). This provides an operational inter-
pretation of the Rényi divergences discussed in Chap. 4. Next, we discuss how the
duality relations and the chain rule for conditional Rényi entropies can be used
to derive entropic uncertainty relations—powerful manifestations of the uncer-
tainty principle of quantum mechanics (Sect. 7.2). Finally, we discuss randomness
extraction against quantum side information, a premier application of the smooth
entropy formalism that justifies its central importance in quantum cryptography
(Sect. 7.3).
10 1 Introduction
What This Book Does Not Cover

It is beyond the scope of this book to provide a comprehensive treatment of the
many applications the mathematical framework reviewed here has found. However,
in addition to Chap. 7, we will mention a few of the most important applications in
the background section of each chapter. Tsallis entropies [162] have found several
applications in physics, but they have no solid foundation in information theory and
we will not discuss them here. It is worth mentioning, however, that many of the
mathematical developments in this book can be applied to quantum Tsallis entropies
as well. There are alternative frameworks besides the smooth entropy framework that
allow to treat unstructured resources, most prominently the information-spectrum
method and its quantum generalization due to Nagaoka and Hayashi [124]. These
approaches are not covered here since they are asymptotically equivalent to the
smooth entropy approach [45, 157]. Finally, this book does not cover Rényi and
smooth versions of mutual information and conditional mutual information. These
quantities are a topic of active research.
Chapter 2
Modeling Quantum Information
Abstract Classical as well as quantum information is stored in physical systems, or

“information is inevitably physical” as Rolf Landauer famously said. These physical
systems are ultimately governed by the laws of quantum mechanics. In this chapter
we quickly review the relevant mathematical foundations of quantum theory and
introduce notational conventions that will be used throughout the book.
In particular we will discuss concepts of functional and matrix analysis as well as

linear algebra that will be of use later. We consider general separable Hilbert spaces
in this chapter, even though in the rest of the book we restrict our attention to the
finite-dimensional case. This digression is useful because it motivates the notation
we use throughout the book, and it allows us to distinguish between the mathematical
structure afforded by quantum theory and the additional structure that is only present
in the finite-dimensional case.
Our notation is summarized in Sect. 2.1 and the remainder of this chapter can
safely be skipped by expert readers. The presentation here is compressed and we omit
proofs. We instead refer to standard textbooks (see Sect. 2.7 for some references) for
a more comprehensive treatment.
2.1 General Remarks on Notation
The notational conventions for this book are summarized in Table 2.1. The table
includes references to the sections where the corresponding concepts are introduced.
Throughout this book we are careful to distinguish between linear operators (e.g.
events and Kraus operators) and functionals on the linear operators (e.g. states), which
are also represented as linear operators (e.g. density operators). This distinction is
inspired by the study of infinite-dimensional systems where these objects do not
necessarily have the same mathematical structure, but it is also helpful in the finite-
dimensional setting.1
1 Forexample, it sheds light on the fact that we use the operator norm for ordinary linear operators
and its dual norm, the trace norm, for density operators.
© The Author(s) 2016 11
M. Tomamichel, Quantum Information Processing with Finite Resources,
SpringerBriefs in Mathematical Physics, DOI 10.1007/978-3-319-21891-5_2
12 2 Modeling Quantum Information
Table 2.1 Overview of notational conventions

Symbol Variants Description Section
R, C R+ Real and complex fields (and non-negative reals)
N Natural numbers
log, exp ln, e Logarithm (to unspecified basis), and its inverse, the
exponential function (natural logarithm and Euler’s
constant)
H H AB , H X Hilbert spaces (for joint system AB and system X ) 2.2.1
·|, |· Bra and ket
Tr(·) Tr A Trace (partial trace) 2.3.1
⊗ (·) ⊗n Tensor product (n-fold tensor product) 2.4.1
⊕ Direct sum for block diagonal operators 2.2.2
AB A is dominated by B, i.e. kernel of A contains kernel of B
A⊥B A and B are orthogonal, i.e. AB = B A = 0
L L (A, B) Bounded linear operators (from H A to H B ) 2.2.1
L† L † (B) Self-adjoint operators (acting on H B )
P P (C D) Positive semi-definite operators (acting on HC D )
{A ≥ B} Projector on subspace where A − B is non-negative
· Operator norm 2.2.1
L• L• (E) Contractions in L (acting on H E )
P• P• (A) Contractions in P (corresponding to events on A) 2.2.2
I IY Identity operator (acting on HY )
·, · Hilbert-Schmidt inner product 2.3.1
T T ≡L ‡ Trace-class operators representing linear functionals
S S ≡ P ‡ Operators representing positive functionals
· ∗ Tr | · | Trace norm on functionals 2.3.1
S• S• (A) Sub-normalized density operators (on A) 2.3.2
S◦ S◦ (B) Normalized density operators, or states (on B)
π πA Fully mixed state (on A), in finite dimensions 2.3.2
ψ ψ AB Maximally entangled state (between A and B), in finite 2.4.2
dimensions
CB CB(A, B) Completely bounded maps (from L (A) to L (B)) 2.6.1
CP Completely positive maps 2.6.2
CPTP CPTNI Completely positive trace-preserving
(trace-non-increasing) map
· + · p Positive cone dual norm (Schatten p-norm) 3.1
(·, ·) Generalized trace distance for sub-normalized states 3.2
F(·, ·) F∗ (·, ·) Fidelity (generalized fidelity for sub-normalized states) 3.3
P(·, ·) Purified distance for sub-normalized states 3.4
‡ This equivalence only holds if the underlying Hilbert space is finite-dimensional
2.1 General Remarks on Notation 13
We do not specify a particular basis for the logarithm throughout this book, and
simply use exp to denote the inverse of log.2 The natural logarithm is denoted by ln.
We label different physical systems by capital Latin letters A, B, C, D, and E,
as well as X , Y , and Z which are specifically reserved for classical systems. The
label thus always determines if a system is quantum or classical. We often use these
labels as subscripts to guide the reader by indicating which system a mathematical
object belongs to. We drop the subscripts when they are evident in the context of an
expression (or if we are not talking about a specific system). We also use the capital
Latin letters L, K , H , M, and N to denote linear operators, where the last two
are reserved for positive semi-definite operators. The identity operator is denoted I .
Density operators, on the other hand, are denoted by lowercase Greek letters ρ, τ ,
σ , and ω. We reserve π and ψ for the fully mixed state and the maximally entangled
state, respectively. Calligraphic letters are used to denote quantum channels and other
maps acting on operators.
2.2 Linear Operators and Events
For our purposes, a physical system is fully characterized by the set of events that
can be observed on it. For classical systems, these events are traditionally modeled
as a σ -algebra of subsets of the sample space, usually the power set in the discrete
case. For quantum systems the structure of events is necessarily more complex, even
in the discrete case. This is due to the non-commutative nature of quantum theory:
the union and intersection of events are generally ill-defined since it matters in which
order events are observed.
Let us first review the mathematical model used to describe events in quantum
mechanics (as positive semi-definite operators on a Hilbert space). Once this is done,
we discuss physical systems carrying quantum and classical information.
2.2.1 Hilbert Spaces and Linear Operators
For concreteness and to introduce the notation, we consider two physical systems
A and B as examples in the following. We associate to A a separable Hilbert space
H A over the field C, equipped with an inner product ·, · : H A × H A → C. In the
finite-dimensional case, this is simply a complex inner product space, but we will
follow a tradition in quantum information theory and call H A a Hilbert space also in
this case. Analogously, we associate the Hilbert space H B to the physical system B.
2 The reader is invited to think of log(x) as the binary logarithm of x and, consequently, exp(x) = 2x ,
as is customary in quantum information theory.
Linear Operators
Our main object of study are linear operators acting on the system’s Hilbert space.
We consistently use upper-case Latin letters to denote such linear operators. More
precisely, we consider the set of bounded linear operators from H A to H B , which
we denote by L (A, B). Bounded here refers to the operator norm induced by the
Hilbert space’s inner product.
The operator norm on L (A, B) is defined as

· : L → sup Lv, Lv B : v ∈ H A , v, v A ≤ 1 . (2.1)
For all L ∈ L (A, B), we have L < ∞ by definition. A linear operator is

continuous if and only if it is bounded.3 Let us now summarize some important
concepts and notation that we will frequently use throughout this book.
• The identity operator on H A is denoted I A .
• The adjoint of a linear operator L ∈ L (A, B) is the unique operator L † ∈
L (B, A) that satisfies w, Lv B = L † w, v A for all v ∈ H A , w ∈ H B . Clearly,
(L † )† = L.
• For scalars α ∈ C, the adjoint corresponds to the complex conjugate, α † = α.
• We find (L K )† = K † L † by applying the definition twice.
• The kernel of a linear operator L ∈ L (A, B) is the subspace of H A spanned by
vectors v ∈ H A satisfying Lv = 0. The support of L is its orthogonal complement
in H A and the rank is the cardinality of the support. Finally, the image of L is the
subspace of H B spanned by vectors w ∈ H B such that w = Lv for some v ∈ H A .
• For operators K , L ∈ L (A) we say that L is dominated by K if the kernel of K
is contained in the kernel of L. Namely, we write L K if and only if
K |v A = 0 =⇒ L |v A = 0 for all v ∈ H A . (2.3)
• We say K , L ∈ L (A) are orthogonal (denoted K ⊥ L) if K L = L K = 0.

• We call a linear operator U ∈ L (A, B) an isometry if it preserves the inner
product, namely if U v, U w B = v, w A for all v, w ∈ H A . This holds if U † U =
IA.
3 Relationto Operator Algebras: Let us note that L (A, B) with the norm · is a Banach space
over C. Furthermore, the operator norm satisfies
L 2
= L† 2
= L† L and LK ≤ L · K . (2.2)
for any L ∈ L (A, B) and K ∈ L (B, A). The inequality states that the norm is sub-multiplicative.
The above properties of the norm imply that the space L (A) is (weakly) closed under multipli-
cation and the adjoint operation. In fact, L (A) constitutes a (Type I factor) von Neumann algebra
or C ∗ algebra. Alternatively, we could have started our considerations right here by postulating a
Type 1 von Neumann algebra as the fundamental object describing individual physical systems,
and then deriving the Hilbert space structure as a consequence.
2.2 Linear Operators and Events 15
• An isometry is an example of a contraction, i.e. an operator L ∈ L (A, B) sat-

isfying L ≤ 1. The set of all such contractions is denoted L• (A, B). Here the
bullet ‘•’ in the subscript of L• (A, B) simply illustrates that we restrict L (A, B)
to the unit ball for the norm · .
For any L ∈ L (A), we denote by L −1 its Moore-Penrose generalized inverse

or pseudoinverse [130] (which always exists in finite dimensions). In particular, the
generalized inverse satisfies L L −1 L = L and L −1 L L −1 = L −1 . If L = L † , the
generalized inverse is just the usual inverse evaluated on the operator’s support.
Bras, Kets and Orthonormal Bases
We use the bra-ket notation throughout this book. For any vector v A ∈ H A , we use
its ket, denoted |v A , to describe the embedding
|v A : C → H A , α → αv A . (2.4)
Similarly, we use its bra, denoted v| A , to describe the functional
v| A : H A → C, w A → v, w A . (2.5)
It is natural to view kets as linear operators from C to H A and bras as linear

operators from H A to C. The above definitions then imply that
|Lv A = L |v A , Lv| A = v| A L † , and v| A = |v†A . (2.6)
Moreover, the inner product can equivalently be written as w, Lv B = w| B L|v A .
Conjugate symmetry of the inner product then corresponds to the relation
w| B L|v A = v| A L † |w B . (2.7)
As a further example, we note that |v A is an isometry if and only if v|v A = 1.

In the following we will work exclusively with linear operators (including bras
and kets) and we will not use the underlying vectors (the elements of the Hilbert
space) or the inner product of the Hilbert space anymore.
We now restrict our attention to the space L (A) := L (A, A) of bounded linear
operators acting on H A . An operator U ∈ L (A) is unitary if U and U † are isome-
tries. An orthonormal basis (ONB) of the system A (or the Hilbert space H A ) is a
set of vectors {ex }x , with ex ∈ H A , such that

1 x=y
ex e y A = δx,y := and |ex ex | A = I A . (2.8)
0 x = y x
We denote the dimension of H A by d A if it is finite and note that the index x ranges
over d A distinct values. For general separable Hilbert spaces x ranges over any
countable set. (We do not usually specify such index sets explicitly.) Various ONBs
exist and are related by unitary operators: if {ex }x is an ONB then {U ex }x is too,
and, furthermore, given two ONBs there always exists a unitary operator mapping
one basis to the other, and vice versa.
Positive Semi-Definite Operators
A special role is played by operators that are self-adjoint and positive semi-definite.
We call an operator H ∈ L (A) self-adjoint if it satisfies H = H † , and the set of all
self-adjoint operators in L (A) is denoted L † (A). Such self-adjoint operators have
a spectral decomposition,
H= λx |ex ex | (2.9)

x
where {λx }x ⊂ R are called eigenvalues and {|ex }x is an orthonormal basis with
eigenvectors |ex . The set {λx }x is also called the spectrum of H , and it is unique.
Finally we introduce the set P(A) of positive semi-definite operators in L (A).
An operator M ∈ L (A) is positive semi-definite if and only if M = L † L for some
L ∈ L (A), so in particular such operators are self-adjoint and have non-negative
eigenvalues. Let us summarize some important concepts and notation concerning
self-adjoint and positive semi-definite operators here.
• We call P ∈ P(A) a projector if it satisfies P 2 = P, i.e. if it has only eigenvalues
0 and 1. The identity I A is a projector.
• For any K , L ∈ L † (A), we write K ≥ L if K − L ∈ P(A). Thus, the relation
‘≥’ constitutes a partial order on L (A).
• For any G, H ∈ L † (A), we use {G ≥ H } to denote the projector onto the subspace
corresponding to non-negative eigenvalues of G − H . Analogously, {G < H } =
I − {G ≥ H } denotes the projector onto the subspace corresponding to negative
eigenvalues of G − H .
Matrix Representation and Transpose
Linear operators in L (A, B) can be conveniently represented as matrices in
Cd A × Cd B . Namely for any L ∈ L (A, B), we can write
L= | f y f y | B L|ex ex | A = f y |L|ex · | f y ex |, (2.10)

x,y x,y
where {ex }x is an ONB of A and { f y } y an ONB of B. This decomposes L into

elementary operators | f y ex | ∈ L• (A, B) and the matrix with entries [L] yx =
f y |L|ex .
Moreover, there always exists a choice of the two bases such that the resulting
matrix is diagonal. For such a choice of bases, we find the singular value decompo-
sition L = x sx | f x ex |, where {sx }x with sx ≥ 0 are called the singular values
of L. In particular, for self-adjoint operators, we can choose | f x = |ex and recover
the eigenvalue decomposition with sx = |λx |.
2.2 Linear Operators and Events 17
The transpose of L with regards to the bases {ex } and { f y } is defined as
L T := f y |L|ex · |ex f y |, L T ∈ L (B, A). (2.11)

x,y
Importantly, in contrast to the adjoint, the transpose is only defined with regards to
a particular basis. Also contrast (2.11) with the matrix representation of L † ,
† T
L† = f y |L|ex · |ex f y | = ex |L † | f y · |ex f y | = L . (2.12)
x,y x,y
Here, L denotes the complex conjugate, which is also basis dependent.
2.2.2 Events and Measures
We are now ready to attach physical meaning to the concepts introduced in the
previous section, and apply them to physical systems carrying quantum information.
Observable events on a quantum system A correspond to operators in the unit

ball of P(A), namely the set
P• (A) := {M ∈ L (A) : 0 ≤ M ≤ I }. (2.13)
(The bullet ‘•’ indicates that we restrict to the unit ball of the norm · .)
Two events M, N ∈ P• (A) are called exclusive if M + N is an event in P• (A)

as well. In this case, we call M + N the union of the events M and N . A complete set
of mutually exclusive events that sum up to the identity is called a positive operator
valued measure (POVM). More generally, for any measurable space (X , Σ) with
Σ a σ -algebra, a POVM is a function
O A : Σ → P• (A) with O A (X ) = I A (2.14)
that is σ -additive, meaning that O A ( i Xi ) = i O A (Xi ) for mutually disjoint

subsets Xi ⊂ X . This definition is too general for our purposes here, and we will
restrict our attention to the case where X is discrete and Σ the power set of X . In
that case the POVM is fully determined if we associate mutually exclusive events to
each x ∈ X .
A function x → M A (x) with M A (x) ∈ P• (A), x M A (x) = I A is called a

positive operator valued measure (POVM) on A.
We assume that x ranges over a countable set for this definition, and we will
in fact not discuss measurements with continuous outcomes in this book. We call
x → M A (x) a projective measure if all M A (x) are projectors, and we call it rank-one
if all M A (x) have rank one.
Structure of Classical Systems
Classical systems have the distinguishing property that all events commute.
To model a classical system X in our quantum framework, we restrict P• (X ) to
a set of events that commute. These are diagonalized by a common ONB, which we
call the classical basis of X . For simplicity, the classical basis is denoted {x}x and
the corresponding kets are |x X. (To avoid confusion, we will call the index y or z
instead of x if the systems Y and Z are considered instead.)
Every M ∈ P• (X ) on a classical system can be written as

M= M(x) |xx| X = M(x), where 0 ≤ M(x) ≤ 1. (2.15)
x x
Instead of writing down the basis projectors, |xx|, we sometimes employ the
direct sum notation to illustrate the block-diagonal structure of such operators. In
the following, whenever we introduce a classical event M on X we also implicitly
introduce the function M(x), and vice versa.
This definition of “classical” events still goes beyond the usual classical formalism
of discrete probability theory. In the usual formalism, M represents a subset of the
sample space (an element of its σ -algebra), and thus corresponds to a projector in
our language, with M(x) ∈ {0, 1} indicating if x is in the set. Our formalism, in
contrast, allows to model probabilistic events, i.e. the event M occurs at most with
probability M(x) ∈ [0, 1] even if the state is deterministically x.4
2.3 Functionals and States
States of a physical system are functionals on the set of bounded linear operators
that map events to the probability that the respective event occurs.
4 This
generalization is quite useful as it, for example, allows us to see the optimal (probabilistic)
Neyman-Pearson test as an event.
2.3 Functionals and States 19
Continuous linear functionals can be represented as trace-class operators. This

then allows us to introduce states for quantum and classical systems.
2.3.1 Trace and Trace-Class Operators
The most fundamental linear functional is the trace. For any orthonormal basis {ex }x
of A, we define the trace over A as
Tr A (·) : L (A) → C, L → ex | L |ex A . (2.16)

x
Note that Tr(L) is finite if d A < ∞ or more generally if L is trace-class, as we will

see below. The trace is cyclic, namely we have
Tr A (K L) = Tr B (L K ) (2.17)
for any two operators L ∈ L (A, B), K ∈ L (B, A) when K L and L K are trace-
class. Thus, in particular, for any L ∈ L (A), we have Tr A (L) = Tr B (ULU† ) for
any isometry U ∈ L (A, B), which shows that the particular choice of basis used for
the definition of the trace in (2.16) is irrelevant. Finally, we have Tr(L † ) = Tr(L).
Trace-Class Operators
Using the trace, continuous linear functionals can be conveniently represented as
elements of the dual Banach space of L (A), namely the space of linear operators
on H A with bounded trace norm.
The trace norm on L (A) is defined as

· ∗ : ξ → Tr |ξ | = Tr ξ †ξ . (2.18)
Operators ξ ∈ L (A) with ξ ∗ < ∞ are called trace-class operators.
We denote the subspace of L (A) consisting of trace-class operators by T (A) and

we use lower-case Greek letters to denote elements of T (A). In infinite dimensions
T (A) is a proper subspace of L (A). In finite dimensions L (A) and T (A) coincide,
but we will use this convention to distinguish between linear operators and linear
operators representing functionals nonetheless.
Another random document with
no related content on Scribd:
held. This is fetched from the bush by the men, who, singing a
certain song, carry it in procession into the arena, where the director
of the mysteries stands, in the attitude of a sacrificing priest. He now
kills a fowl, the blood of which is caught in a bowl, while some
charcoal is pounded to powder in a second vessel, and some red clay
crushed in a third—the branch is then encircled with a triple band of
the three substances—red, black, and red. Meanwhile some men
have been digging a hole, in which is laid a charm made out of pieces
of bark tied together. The hole is then filled up and the earth heaped
over it in a mound on which the forked branch, called lupanda, is
planted. A second mound is then made, which, as well as the first,
was still clearly recognizable in the ring of huts at Akundonde’s. This
second mound is the seat for the unyago boy who is considered of
highest rank, the others being grouped around him, on stumps,
which, if the director of the proceedings has the slightest sense of
beauty, are arranged in two regular, concentric circles similar to
those which I saw in the bush near Chingulungulu. “The cromlech of
the tropics!” was the idea which occurred to me at the time, and even
now I cannot resist the impression that this arrangement of tree-
stumps resembles our prehistoric stone circles, not only in form but
perhaps also in the object for which it is designed. If our Neolithic
megaliths were, really used by assemblies for ritual purposes, there
seems no reason why these venerable stones should not have served
as seats for our ancestors. The negro, too, would no doubt dispense
with wooden seats, if stone ones had been obtainable in his country.
If I were at all given to imaginative speculations, I could easily
prove that the Makonde are fire-worshippers. As soon as the men
have built the likumbi, i.e., a hut of the kind we saw at Mangupa, all
scatter to look for medicine in the bush. In the evening of the same
day, they give the roots they have collected to an old woman who
pounds them in a mortar. The resulting paste is dabbed in spots on
the arms of some five or six men by the high priest or doctor. When
this is done all sit inactive till midnight, when the munchira (doctor)
begins to beat his drum. As the deep sound of this instrument
thunders out through the dark tropical night, all the people, adults
and children, stream out of the huts, and dancing and gun-firing are
kept up till the following afternoon, when they distribute presents to
each other and to the boys’ instructor. Thereupon the munchira
delivers an address. The six men above referred to are, he says,
sacred; if they should take it into their heads to steal, or commit
violent assaults, or interfere with their neighbours’ wives, no one
must do anything to them, their persons are inviolable. The six, for
their part, are now informed by him that it is their duty to beat the
drum at midnight for the next three months.
When the three months are ended, the village is all stir and bustle.
Men go into the bush to collect dead wood, and in the evening carry
it in perfect silence to the open space near the likumbi hut. The
women, meanwhile, have been preparing enormous quantities of
beer, which also finds its way to the likumbi. In this hut stands a
small round covered basket (chihero), containing medicines, into
which (and on the medicines) every one of the wood-gatherers spits
out a little of the specially prepared beer. Beside the chihero stands
the old woman who pounded the medicines in the mortar, who then
puts the basket on her head, seizes in one hand the end of a whole
piece of calico, specially bought for the ceremony, and leaves the hut
with a slow and solemn step, dragging the cloth behind her. The first
of the wood-gatherers quickly takes hold of it, so as not to let it touch
the ground; as it unrolls from the bale a second takes it, then a third,
and a fourth, till at last it passes along a little above the ground, like a
train borne by pages. The munchira walks in front next to the
woman, and they circumambulate the likumbi, after which the
munchira takes the end of the calico and wraps it round the chihero.
This he then holds to his right ear; after a short pause, he places it on
his shoulder, again keeping it there for a few moments; then it is
lowered to the hip, the knee, and finally to the outside of the ankle.
At the close of the ceremony the venerable man takes both cloth and
chihero as his well-earned fee.
Again it is night—the outline of the great wood-pile is just
recognizable in the faint light. About an hour after midnight, a tall,
gaunt figure rises from the circle of prostrate figures wrapped in
their sleeping-mats. Silently it glides up to the pile, a little flame
flashes up, to disappear again; but soon there is a fresh crackling; the
flame, in the draught produced by the rhythmic pulsations of a fan,
grows and strengthens. Now we recognize the figure—it is that of the
munchira. In a few minutes the whole large pile is a sheet of flame,
its flickering, quivering lights dancing on the shining faces of the
men standing round in a circle. The fire having now burnt up
brightly, the munchira walks quickly round it, and, his face turned to
the pile, utters the following words:—“Let the wounds of the boys
heal soon and painlessly, and let the chief who is keeping the likumbi
this year find the boys do him credit in after life.” At the same time
he ties a white rag to a pole, and fans the fire with powerful strokes.
The men remain standing round it, watching it as it dies down, till
broad daylight.
Fire, as the central point in a ceremony which cuts so deeply into
social life as do the celebrations of puberty among these tribes, is so
far as I know quite an isolated phenomenon among the peoples of
Africa. Have we here a case of genuine fire-worship, or are the walk
round the fire and the address to it only the last unconscious
survivals of a cult prevalent in ancient times? I do not know, and, to
speak frankly, cannot even say where the answer to this question
may be looked for. We must not a priori assume it to be impossible
that the Makonde should once have been fire-worshippers; we know
far too little as yet of their social evolution. The abundant results of
my inquiries up to this point are the best proof that unexpected
discoveries are yet in store for us.
MASKED DANCE AT THE GIRLS’ UNYAGO, NIUCHI
In the male sex the transition from childhood to the status of fully-
qualified maturity is a single, definite process, though extending over
a long period. The memory of rejoicings and sufferings experienced
in common is preserved henceforth among the men by means of a
free, voluntary association known as the “age-class.” All those who
have passed through the unyago in the same year stand by each
other till death severs the connection. This connection, however,
must be thought of in terms of African conditions; there is no society
or club, or the like, and the sole obligation incurred by the old friends
is that every one of them is bound to offer hospitality to any of the
others who may come to his village. Secret societies no longer
consciously influence the character of the age-classes here in the
East, though the reverse is the case in West Africa where the two
things go hand in hand, acting and reacting on each other as cause
and effect, and both finding their common outward expression in
great festivals with masked dances and other mysterious accessories
calculated to terrify the women and the uninitiated men. Here on the
Makonde plateau, the three phenomena—the age-classes, the
festivals and the masked dances—are at the present day not very
closely connected together; yet everything leads to the conclusion
that the masked dance now in use among the Makonde was
originally the outcome of a long-forgotten system of secret societies,
similar to the quite analogous institutions of Kamerun, Upper
Guinea, and Loango. There is many a knotty problem yet to be solved
in this department of African ethnography.
The girls’ unyago is a graduated series of courses of instruction. I
have purposely emphasized the word instruction, as there is nothing
here in the nature of a surgical operation, with a single exception in
the case of the Makua. In all the tribes each girl is given for the whole
period of the unyago into the charge of a special teacher, who
remains her friend through life. Under the guidance of these older
women, the novices in the first place go through a curriculum very
much resembling that of the boys. The children are unreservedly
enlightened as to all sexual relations, and have to learn everything
connected with married life. They are also taught all the rules which
govern intercourse between members of the same tribe, and above all
of the same family.
There is an opening and a closing ceremony for this first course of
the girls’ initiation. I was able personally to observe the revels which
take place on such occasions, at all three of the places where I had
the opportunity of making the chiputu (or echiputu) illustrious by my
presence. The phenomenal thirst shown is quite explained by the
amount of dancing gone through.
WOMAN OF THE MAKONDE TRIBE
After the mysteries, both boys and girls in due course become
marriageable, but I have not succeeded in ascertaining, even
approximately, the age at which this is the case. Individuals are
always out of measure astonished when asked their age, and their
relatives are profoundly indifferent on the subject. In general,
marriage takes place very early, as is proved by the very young
mothers who may be seen in any large assemblage of people, and
who are mostly no further developed than German girls at their
confirmation. Matola tells me that the form of marriage known as
masange was formerly very prevalent, in which young children of
from five to seven were united, huts being built for them to live in.
This custom is said still to be practised occasionally.[58] The same
informant states that it is very common for one woman, who has just
had a child to say to a neighbour expecting a like event, “I have a son
—if you have a daughter, let him marry her”; and this, in due course,
is done.
The African native is a peasant, not only in his avocation, but in
the way in which he sets about his courting. In no other department
is his mental kinship with our own rustics so startlingly shown. To
express it briefly: the native youth in love is too shy to venture a bold
stroke for his happiness in person; he requires a go-between quite in
the style of our own rural candidates for matrimony. This office is
usually undertaken by his own father, who, under some pretext or
other, calls on the parents of the bride-elect, and in the course of
conversation touches on his son’s projects. If the other side are
willing to entertain the proposition, the negotiations are soon
brought to a satisfactory conclusion—that is to say, if the maid, too,
is willing. Girls are not in reality so passive in the matter as we are
apt to assume, but most certainly expect to have their wishes
consulted; and many a carefully-planned match has come to nothing
merely because the girl loved another man. In this respect there is
not the slightest difference between white and black. Of course, not
every native girl is a heroine of constancy and steadfastness; here
and there one lets herself be persuaded to accept, instead of the
young man she loves in secret, an elderly wooer who is indifferent to
her, but in that case she runs the risk of incurring—as happens
elsewhere—the ridicule of her companions. The old bridegroom,
moreover, may be pretty certain that he will not enjoy a monopoly of
his young wife’s society.
Marriage is a matter of business, thinks the African, quite
consistently with his general character, and the contract is only
looked upon as concluded when the two fathers have come to an
agreement as to the amount of the present to be paid by the
bridegroom. The people here in the south are poor—they have
neither large herds of horned cattle, nor abundance of sheep and
goats; the whole purchase—were it correct, which it is not, to call the
transaction by that name—is effected by handing over a moderate
quantity of calico.
Much more interesting from an ethnographic point of view than
the Yao wooing just sketched, are the customs of the Makua and
Makonde. In their case, too, negotiations are opened by the fathers;
but this is, in reality, only a skirmish of outposts,—the main action is
afterwards fought by the mothers, each supported by her eldest
brother, or perhaps by all her brothers. The fact that the
matriarchate is still flourishing here explains the part they take in
the matter.
Nils Knudsen, by the way, can tell a pretty story—of which he is
himself the hero—illustrating the constancy of native girls. During
the years of his lonely life at Luisenfelde, he so completely adapted
himself to native ways as to take a wife from among the Wayao. Even
now, after the lapse of years, he never grew tired of praising the
virtues of this chipini wearer;—she was pretty, and domestic, and a
first-rate cook—she could make excellent ugali, and had all the other
good qualities which go to make up a good housewife in the bush.
One day he went off to the Rovuma on a hunting expedition; he was
only absent a few days, but on his return she had disappeared. On
the table lay a knotted piece of bark-string. He counted the knots and
found that there were seventy; the meaning of the token, according
to the explanation given by the wise men of the tribe being this:
—“My kinsfolk have taken me away; they do not like me to live with
the white man, and want me to marry a black man who lives far away
on the other side of the Rovuma. But even if I should live as many
years as there are knots on this string, I will not take him, but remain
faithful to you, the white man.” This was Knudsen’s story, and he
added, with emotion not untouched by the pride of a man who feels
himself to be greatly sought after, the further statement that the girl
was in fact keeping her vow. She was living far away, in the heart of
the Portuguese territory, and near the man for whom she was
destined, but even the strongest pressure brought to bear by her
family could not make her give way. After all, there is such a thing as
faithfulness in love.
The native wedding is a very tame affair—one might almost say
that there is no such thing. Betrothal and marriage, if we may say so,
coincide in point of time. When once the wooer has obtained the
approval of the rightful authorities, there is no further hindrance to
the union of the couple than the delay necessary for erecting a new
hut for them. When this is done and they have taken up their abode
in it, the young husband begins to work for his mother-in-law, in the
manner aforesaid, which appears so strange to our European ideas,
though we cannot deny that there is room for improvement in our
manners in this respect.
Now, however, we have to consider the question of who may marry
whom, or, in other words, the table of forbidden degrees. This
question has its importance even in Europe—how much more among
people so much nearer the primitive conditions of society. If it is for
the wise men of an Australian tribe one of the highest problems of
social science to determine with absolute correctness which girl
among the surrounding families the young man A may marry, and
who is eligible for the young man B, so neither are the matrimonially
disposed in the Rovuma valley free to indulge their inclination in any
direction they may choose.
It is late in the afternoon. In the baraza at Newala fifteen natives of
respectable age are squatting, as they have done for some weeks past,
on the big mat. From time to time one of these seniors rises, and
leaves the building to stretch his cramped legs, but always returns
after a short time. The place is hot, a fetid vapour hangs over the
assembly, so that the European in khaki, writing so assiduously at his
folding table, presses his hands again and again to his aching
forehead. The company are obviously tired, but they have to-day
been occupied with a very exhausting subject. Hour after hour, I—for
I am the man with the headache—have been trying, in the first place,
to make clear to Nils Knudsen the principles of human marriage
customs, of the various tribal divisions, of totemism, of father-right
and mother-right—in short, a whole series of points in sociology, but
with no very satisfactory result, as is clearly shown by every question
I put. Now the task before me is to elicit from the fifteen wise elders,
with his help and that of the usually acute Sefu, everything they know
on these subjects. All my small failures have made me quite savage,
besides wearying me to the point of exhaustion; and it costs me an
appreciable effort to fling a question into the midst of the learned
assembly.
“Well, old Dambwala, lazy one, you have a son, have you not?”
“Yes, sir.”
“And you, Nantiaka, you have a daughter?”
“Yes, sir.”
“Very good. Now, Dambwala, can your son marry Nantiaka’s
daughter?”
“No.”
“And why not?” I must have been very tired, indeed, for even the
surprise audible in this decided negative raised no particular
expectations in my mind. I only began to listen more attentively
when, among the reasons for the negative then alleged, my ear
caught the word litawa. “Nini litawa? What is a litawa?” I ask, now
quite fresh and lively. Well, it appears, a litawa is a litawa. Then
comes a long shauri, in which the wits of the natives, who, like us
have been half asleep, awaken to full activity, and all three languages
—Makonde, Yao, and Makua—are heard at once with a clatter of
tongues like that conventionally attributed to a woman’s tea-party.
At last the definition is found. Translated into technical language
litawa means the matriarchal exogamic kin, including all descended
from one common ancestress. A man’s inheritance does not descend
to his son, but to the son of his sister, and a young Makonde takes his
wife, not from his own litawa, but in one of the numerous matawa
outside his own. The Makua have exactly the same arrangement, but
the word they use instead of litawa is nihimu.
The evening of this day—the twenty-first of September—was
cheered by the feeling that it had been among the most successful of
my whole journey. In order to celebrate it in a worthy fashion,
Knudsen and I, instead of the one bottle of beer which we had been
in the habit of sharing between us, shared two.
The reader, especially after my declaration in Chapter II, will
wonder how we suddenly became possessed of this beverage. It is
true that, in the heat of the plains the mere thought of it was
intolerable, but, up here, close to the clouds, especially when the east
wind blows cold of an evening, a glass of German beer is very
welcome. A few weeks ago I had occasion to send a dozen cases of
specimens down to Lindi. The twelve carriers left early one morning,
and were expected back in a fortnight. On all previous occasions of
this sort, their absence had left me cold; this time, to be honest, we
two white men counted the days of that fortnight, and, when, on a
Sunday morning, the unmistakable sound of Wanyamwezi porters
approaching their journey’s end was heard far out in the bush, we
hurried to meet the great case containing many long-forgotten
comforts—not only the heavy German stout from the Dar es Salam
brewery, but above all, the milk we had so greatly missed, and which
in our present state of emaciation was an absolute necessity.
On that memorable afternoon, however, the close of which I have
thus been anticipating, I had no leisure to think of such material
delights as these.
“So your son, friend Dambwala, cannot marry Nantiaka’s
daughter, because both belong to the same litawa—what is the name
of your litawa?”
“Waniuchi.”
“And where do you live?”
“In and around Niuchi.”
“And you, Kumidachi,” I went on, turning to another old man, in a
new embroidered fez, which marked him as a headman, “to what
litawa do you belong?”
“Nanyanga,” was the prompt reply. Instantly the name is written
down, and my eye rests questioningly on the next wise man. He, one
of the quickest, already knows what is wanted, and does not wait to
be asked, but calls out, “Wamhwidia.”
But I cannot go on in this way—I must find out, not only the names
but their meanings. I have already discovered, in my study of
personal names, how fond the natives are of discussing etymologies,
and here, too, only a slight hint is needed to get the meaning of the
clan-name as well as the name itself. I had translated Waniuchi as
“the people of Niuchi;” but this interpretation did not satisfy these
black philologists,—niuchi was “a bee,” they said, and the Waniuchi
were people who sought honey in hollow trees. The Nanyanga were
flute players in time of war, nanyanga being the name of the
Makonde flute. The Wamhidia, they said, had their name derived
from the verb muhidia, “to strike down,” from their warlike
ancestors, who were continually fighting, and had beaten down
everything before them.
That afternoon, the old men, in spite of their weariness, had to
keep on much longer than usual: I had tasted blood and pumped
them, till, about sunset, their poor brains, unaccustomed to such
continued exertion, could do no more. They, however, received an
extra tip, in return for their self-sacrificing help in this difficult
subject. Even Moritz, the finance-minister, had to-day quite lost his
usual hang-dog expression, and grinned all over his brown face when
he came, after we had struck work, to hand my assistants their bright
new silver pieces. Since then I have devoted all my efforts to the
study of the clan system, and do not know what most excites my
astonishment, the social differentiation of the tribes, their
subdivision into innumerable matawa and dihimu (plural of
nihimu), or the fact that, as I am forced to assume, none of my
predecessors in this field of study has had his attention called to this
arrangement. However, when I come to think it over, I have no
reason to be surprised, for in the first place, I had been travelling
about the country for months without suspecting the existence of the
clan system, and in the second, it was a mere accident that, in the
discussion just described, the answer happened to take just the form
it did. Men are to a certain extent at the mercy of the unforeseen—
the scientific traveller most of all.
Needless to say, immediately after this momentous discovery, I
came back to the problem of the Yaos. After my Makua and Makonde
men had for some time been dictating name after name with the
most interesting explanations into my note-book, Nils Knudsen
suddenly said, “The Yaos have something of that sort, too.” Ten
minutes later, swift messengers were already on the way to fetch up
from the plain any men of that tribe who had the slightest
pretensions to intelligence. They all came up—Zuza, and Daudi, and
Masanyara and the rest. Even now the examination was no easy task,
either for me or for the subjects, but after honestly doing my best, I
got enough out of them to be able to say, “Nils Knudsen is right, the
Yaos, too, have something of the sort.” Not only so, but in their case I
ascertained without much difficulty that there is a second division
into large groups, quite independent of the system of matriarchal,
exogamous clans.
Of the great groups of the Yao tribe, which is now spread over an
extraordinarily large region of East Africa, since it extends from Lake
Chilwa in the south almost to the gates of Lindi in the north, the
following are known to us,—the Amakale, near the sources of the
Rovuma, the Achinamataka or Wamwembe at Mataka’s, between the
Rovuma and the Lujende; the Amasaninga, originally at the south
end of Lake Nyasa; the Achinamakanjira, or Amachinga, on the
Upper Lujende; the Mangoche in the neighbourhood of Blantyre. The
indication of the residences of these great groups, as here given, has
now merely a historic value. Through the gradual migrations already
alluded to, the old limits of the groups are now quite effaced, and can
no longer be definitely laid down on the map. The clans, too (here
called ngosyo, plural of lukosyo), cannot possibly have any definite
position assigned them on the map; and this is also true of the other
tribes. Some clans, indeed, may have a recognizable centre of
distribution, but in general, the same confusion prevails here as in
the case of the larger divisions.
It was not merely curiosity which made me so persistent in
inquiring into the meaning of clan names, but the desire to ascertain
whether they convey any indications of totemism. It may not be
superfluous to say that the word totem comes from North America,
and was originally applied to the drawings of animals appended by
the Iroquois chiefs to their treaties with the white man by way of
signature, the animal represented being that from which the clan of
the signatory traced its descent. Totemism was first studied among
these North American Indians, but was afterwards discovered to
exist in Australia, apparently, also, in Melanesia, and in a very
marked form among the older populations of India, as well as in
various other parts of the world. In most cases, the clans trace their
descent from some animal, which is reckoned sacred and
invulnerable and must not be hunted or eaten. In some isolated
instances it is even considered the height of good fortune for a man
to be eaten by his totem animal. Small and harmless creatures, as
well as plants, are also chosen as totems—otherwise it would scarcely
be possible to find enough; as, for example, in Southern India, where
the totems are innumerable. I cannot here give the whole long series
of clan names collected by me for all three tribes, but must refer the
reader for this part of my results to the official publication. But it was
interesting to find that though totemism no longer consciously exists
among the natives, many a small trait witnesses to its former
prevalence. To point out these traits in detail will be the task of later
inquirers, I will here give only a few specimens of the clan names.
PHONOGRAPHIC RENDERING OF A NATIVE SONG
Matola and his cousin, our common friend, Daudi, belong to the
lukosyo of the Achemtinga, but at the same time to the group of the
Amachinga.[59] The prefix Che, as already stated, is an honorific title
for both men and women:—Chemtinga, according to Daudi, was once
a great chief in the region of the upper Lujende. The Masimbo lived
in Zuza’s district. These take their name from the pitfalls (lisimbo,
plural masimbo) in which their forefathers used to catch game. The
Amiraji, who lived near Mwiti, derive their name from the character
of the country where they formerly lived, which abounded in bamboo
(mlasi).[60] Another Yao clan are the Achingala, who take their name
from the ngala, a kind of mussel, found in the Rovuma and its
tributaries, the shells of which are still used as spoons; the reason for
the name is said to be that their ancestors chiefly lived on this
mollusc.
In the same category as these last we may place the Makua clan of
the Wamhole, whose forefathers fed on the wild manioc (mhole), a
root still eaten in time of famine. The Makonde clan of the
Wambunga derive their name from the tradition that their ancestors
ate the nambunga, or fruit of the bamboo. The Wantanda formerly
had the custom of cutting the flesh of the game they killed into long
strips (nantanda). The Wamunga[61] are rice-planters, the ancestors
of the Alamande lived on a small locust of that name, and the
Wutende are people famous throughout the country on account of a
quality for which we are little disposed to give the natives credit—
they are always working (kutenda).
Even in the cool climate of Europe it is not altogether easy for the
mind to grasp the marriage laws of these clans. Here in tropical
Africa, with its perpetual alternations of heat and cold, I find it
almost impossible to follow the expositions of old Mponda, my
principal lecturer on Civil Law. Moreover, it is very much of a shock
to our customary ways of thinking, to hear, for example, the
following:—After the Makonde boy has been circumcised he does not
return to his parents’ house, but remains in that of his maternal
uncle. There he has nothing further to do but grow up and wait till
his girl cousins are grown up likewise. If the uncle has no daughters,
the nephew first waits till one is born, and, after this event has taken
place, he has again to wait. It must be understood that the young
man is not supposed to get his board for nothing all this time; he is
expected to work pretty hard, like Jacob serving seven years for
Rachel. When at last the goal is reached and the cousin is
marriageable, the suitor, meanwhile arrived at years of discretion,
goes away somewhere where he can earn a rupee’s worth of calico,
hands this to his uncle, and takes home his wife. He is not, however,
free to live where he likes, but remains at his uncle’s village, and
works for him like a bondsman, as before. If, in due course, he has a
son, this son, according to Mponda, must again marry a cousin—the
daughter of his father’s sister. In the old man’s own concise words:
“If I have a sister and she has a daughter, and I have a son, my son
can marry that girl. But if I have a brother and he has a daughter, my
son cannot marry his daughter, because she is numbuwe—his sister.”
We took our leave of the young girl at the moment when, after
passing through the months of the chiputu with their formalities and
festivities, she has taken her place among the initiated. According to
some of my informants the child’s marriage takes place very soon
after this epoch—certainly before the period which we in Europe
consider as the beginning of maturity, viz., the first menstruation.
I have no means of checking these statements, so cannot say
whether this is so or not; in any case we are just now more interested
in the treatment of girls on the occasion alluded to—the more so that
this treatment is analogous to that practised in a whole series of
other regions. As on the Lower Guinea coast, (in Loango,[62] on the
Gabun, and on the Ogowe) and in various parts of Melanesia, the girl
is lodged in a separate hut, where she remains entirely alone; her
friends come and dance, uttering the shrill cry of the ntungululu
outside the hut, but otherwise keep at a distance. Her mother, her
instructress during the unyago, and the other wise women, however,
impart to her the rules of conduct and hygiene:—she must keep at a
distance from every one; she must be particular as to cleanliness,
must wash herself and bathe, but above all, must have intercourse
with no one. This is repeated over and over again, while at the same
time eating, singing and dancing go on incessantly.
At the first pregnancy of a young wife, also, various ceremonies
take place. At bottom, however, these are only a pleasant setting for a
number of rules and prohibitions inculcated on this occasion by the
older women. In the fifth month the young woman has her head
shaved, and a month later the women make a feast for themselves,
and roast some maize for her. Some more maize is then soaked in
water and pounded and the resulting paste smeared on her head.
Then the husband goes to the bush, accompanied by a near relation
of his wife’s, the woman wearing nothing but a small waist-cloth. The
man cuts down a suitable tree and prepares a piece of bark-cloth in
the way already described, while the girl sings in time to the strokes
of his mallet “Nalishanira wozewa neakutende.” The fabric when
finished is ornamented with beads, and the instructress hangs it
round her protegée’s neck as a charm. This is called mare ndembo,
and the same name is henceforth applied to the expectant mother.
Next morning all the people are again assembled for the dance—the
inevitable ntungululu inseparable from all joyful feelings or festive
occasions, mingling, of course, with the singing and hand-clapping.
All, however, do not take part in these rejoicings; the wise women
and the instructress stand apart from the crowd, in a group round
the young wife. “You must not sit on other people’s mats,” says one
toothless old woman, “it would injure both you and the child—you
would be prematurely confined.”
“You must not talk to your friends, men or women,” says another
woman, whose utterance is impeded by an enormous pelele, “that,
too, would be bad for the child.”
“You must not go out much after this,” says a third. “If possible let
no one see you but your husband, or the baby might resemble
someone else. But if you do go out, you must get out of people’s way,
for even the smell of them might hurt the child.”
There is, after all, something in these rules and warnings. We in
Europe are quite familiar with the idea that a pregnant woman must
not see anything unpleasant or terrifying, and ought not, if she can
possibly help it, to let herself be impressed by any other face than
that of her husband. The other prescriptions belong to the region of
sympathetic magic, or action by analogy—the mere possibility of
coming within the atmosphere of people who have recently had
sexual intercourse with one another may endanger the coming life.
But this is not all,—the most important points are yet to come.
“You must not eat eggs, or your child will have no hair.”
“You must not eat the flesh of monkeys, or the child will have no
more sense than a monkey.”
“You must not eat what is left over in the cooking-pot from the day
before, or the baby will be ill.”
“If you go to the garden or the well, and anyone salutes you, you
must not thank him or answer him in any way, for then the birth of
the child will be long delayed.”
The conclusion of the whole lecture which, in contrast to the
system pursued in our Universities, is simultaneously delivered by
many teachers to one unhappy student, is the very urgent and
serious warning to have nothing to do with any other man than her
husband, or she will infallibly die. On the other hand, if her husband
were to forget himself and go after another woman, she would have a
miscarriage, resulting in her death. She must, therefore, be very good
to him and cook his porridge as he likes it.
This is the last word. With the peculiar gait of the native woman,
which has an inimitable twist in it, not to be described in words, the
dispensers of wise counsel hasten, as fast as their dignity will allow,
across the open space and join the rest of the throng, “Lu-lu-lu-lu-lu-
lu,”—the shrill vibrations again agitate the air, the drums, beaten by
the men’s strong hands, strike up afresh, a mighty cloud of dust rises
and veils the whole scene, everything is in motion and full of genuine
African mirth, all unconscious of life’s daily miseries. One alone sits
by in silence, the young woman herself who, according to the
instructions just received, is entirely interdicted from taking any part
in the festivity. Her brown eyes—which would deserve to be called
beautiful were their effect not marred by the white being
interspersed with yellowish-brown specks—are fixed musingly on
one point. Is she thinking of the dark hour she will have to encounter
in a few months’ time? The Scripture, “In sorrow shalt thou bring
forth children,” is true for the black race also. But, personally, I do
not think that the young thing is looking so far ahead; it is not in any
case natural for youth to do so, and African youth, in particular, sees
no occasion to be anxious about the future. The race is truly happy,
in the enviable facility with which it lives for to-day, leaving to-
morrow’s cares entire and untouched for to-morrow.
Note.—The system of kinship among the Yaos and neighbouring tribes has not
been so entirely overlooked by inquirers as Dr. Weule supposes. The subject has
been investigated by Archdeacon Johnson, the late Bishop Maples, and the Rev. H.
B. Barnes among others, though, unfortunately, many of their notes are buried in
little-known periodicals. Some valuable information is also to be found in Mr. R.
Sutherland Rattray’s Some Folk-lore, Songs and Stories in Chinyanja. We think
Dr. Weule is mistaken in distinguishing the “larger groups” of the Yao tribe from
the ngosyo: they are probably identical with the latter in origin: e.g., the Machinga
would be the descendants of a single (female) ancestor, who in the course of
generations became numerous and powerful, and perhaps increased their
consequence by incorporating weaker clans who placed themselves under their
protection and adopted their name. But there is a second system of descent, which
may be what Dr. Weule is referring to. This is called by the Anyanja chilawa, and
descends through the father; marriage within it is prohibited. “A man may not
marry any woman who is of his kamu (Yao, lukosyo) or of his chilawa. Thus the
daughters of his mother’s sisters are excluded because they are of the same kamu,
and daughters of his father’s brothers are excluded because they are of the same
chilawa; but the daughters of his mother’s brothers or of his father’s sisters are
eligible, because they are neither of the same kamu nor of the same chilawa” (Rev.
H. B. Barnes). This tallies with the information given to Dr. Weule about the
Makonde marriage laws (p. 314). Mr. Barnes doubts whether the clan names
explained to Dr. Weule are really connected with totems, and thinks the customs
they refer to are “perhaps more likely to be traceable to individual peculiarities of
some ancestor than to any religious totemistic restriction,” and that the chilawa
names, whose significance appears to be lost, are the real totem names. But the
subject is too wide to be discussed in a note. [Tr.]
CHAPTER XV
LAST DAYS AT NEWALA
Newala, October 10, 1906.

“Morgen muss ich fort von hier
Und muss Abschied nehmen....”
The words of the German students’ song rise to my lips, now that I
am thinking of bringing our stay here to a close—though, as a rule, I
am anything but musical, and Knudsen, for his part, can never get
beyond the first line of Gamle Norge. The mention of music suggests
my experiences with the phonograph. When laying in my stock of
blank cylinders at Berlin, it was a happy inspiration of mine to take
half-a-dozen records as well, in the hope that they might serve to
charm the savage breast of the African. I have no sort of
responsibility for the choice of these pieces, as I left it entirely to the
girl who served me at the shop where I bought them. What
determined her selection I cannot tell, but it is a fact that the greater
number of the six records, though not all, are immensely popular. An
American march—quite rightly—produces no impression whatever,
and a selection of songs fails to attract my public: it seems to suggest
nothing at all to them. The next item on the programme, the
arrangement of which I always leave to Knudsen, so that he may
learn to work the instrument,—is “Die beiden kleinen Finken” (“The
Two Little Finches”). Here and there an eye lights up with
intelligence when the twittering of the birds begins, and many sets of
white teeth are seen flashing behind the parapet which shuts off our
baraza from the outer passage. Then comes the well-known
xylophone solo, “Der Specht” (“The Woodpecker”). As the deep bass
voice announcing the title of the piece issues from the funnel, the
whole audience leans over the wall in feverish excitement, one might
almost say with ears erect. A few of the experienced elders, who have
been on the coast and therefore have the right to appear blasés,
laugh ostentatiously to show that they understand. But this laughter
dies away when the pure tones of my instrument, unmixed with any
adventitious sound, begin to reproduce in the most striking way the
unmistakable notes of the xylophone. One can see that these people
have an ear and enjoy the harmony of sounds perhaps as much as we
do. Besides, the sounds are not in this case unfamiliar—for the
mgoromondo, the straw xylophone already described, has exactly the
same timbre. By the time the final tapping duet begins, everything
about them is shining—their eyes, their teeth, their whole faces—in
fact they shine all over, for they keep crowding together more and
more closely, and it is by no means cool. “Die Schmiede im Walde”
(“The Forge in the Forest”) scarcely heightens their pleasure; it is
true that the enjoyment is great and general, but the blacksmith is a
familiar figure of everyday life, and the rhythm of his hammer as well
known to them as it is to us. Now, however, comes our aria di
bravura. It has been my experience that when a white man, after
long residence among savages, declines more or less from the level of
civilized society, music is the first thing to stimulate the endeavour
towards recovery. Nils Knudsen can listen to the Fledermaus
seventeen times running without getting enough of it. He winds up
the apparatus over and over again and remarks that this is real music
—the right sort. The natives, too, are delighted with the merry,
audacious tunes, and if the mood of the moment is such that I feel
moved to execute a few waltz or polka steps and float, like a fairy
weighing some thirteen stone, round the table on which the
phonograph is placed, their delight becomes indescribable rapture.
This is the right moment for turning the tables and calling on the
audience to become performers in their turn. The Newala natives are
very reluctant to oblige in this respect; the men can only be induced
to come up to the phonograph when under the influence of the
ecstasy just alluded to, but the women are off like the wind whenever
I want them.
The men, too, here at Newala, would not come near me for a time.
I had become so absorbed in the linguistic studies which had been
occupying me more and more during the last few weeks, that my
growing isolation did not at first strike me. Only when Knudsen and I
found that we scarcely ever saw any one besides my three teachers,
the akida Sefu, the Yao Akuchigombo (which is, being interpreted,
Mr. Toothbrush), and the Makua Namalowe (Mr. Echo), it became
clear to me that some circumstance unknown to me must be the
cause of this boycotting. Neither Sefu nor the other two could or
would explain matters. Mr. Echo had only been resident a short time
at Newala, having recently come to be trained as a teacher under his
older colleague at the Universities’ Mission, so that his ignorance was
not surprising; but it annoyed me greatly that the other two would
give no answer to all my inquiries beyond “Si jui” (“I don’t know”).
However, I was forced to admit that even these two did not really
belong to the place, Sefu being a coast man, and in his capacity of
akida, probably more feared than loved, while Akuchigombo was
educated at Zanzibar, and through his position as teacher of the
Mission School, separated by a great gulf from the illiterate mass of
the population. This school, with a rusty tube of an artesian well and
a small church-bell, hung according to the custom of this country in
the first convenient tree, are the only relics of the once flourishing
station of New Newala.
Only within the last few days has Knudsen been able to get out of
an old friend from the plains the reason why we have been left so
severely alone. The explanation, strange as it may seem to a
European, is genuinely African: it is nothing more nor less than the
suspicion—indeed the certainty—that I am a dangerous sorcerer.
Somehow the belief had gained ground that in photographing people
I deprived them of whatever clothes they were wearing. “Have you
not seen,” some individual whose name is as yet unknown to me, is
reported as saying to his countrymen, “how the white man gets
under his great black cloth? It is then that he bewitches you. You are
standing there with all your clothes on, but he goes and stands for
hours in his tent overnight, working his charms, and next day, when
he gets out his glasses, there you are on them quite naked. And if you
are foolish enough to go and stand in front of the other machine, he
will take away your voices, too. He is a great wizard, and his
medicines are stronger than even our chisango (divination oracle).
We made war against the Wadachi (the Germans), but what fools we
were to do so, for this white man is one of them!”
The comic aspect of the situation struck me far more forcibly than
the annoying one, and we both laughed heartily. I had not before

PDF Quantum Information Processing With Finite Resources Mathematical Foundations 1St Edition Marco Tomamichel Auth Ebook Full Chapter

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

PDF Quantum Information Processing With Finite Resources Mathematical Foundations 1St Edition Marco Tomamichel Auth Ebook Full Chapter

Uploaded by

Copyright:

Available Formats

Quantum Information Processing with

Finite Resources Mathematical

Mathematical foundations of quantum mechanics Neumann

Quantum Information Processing, Quantum Computing, and

Device-Independent Quantum Information Processing: A

Spectral Theory and Quantum Mechanics Mathematical

Introduction to the Theory of Quantum Information

Mathematical Foundations of Computer Science 1st

Mathematical Image Processing Kristian Bredies

Quantum Physics: An Overview of a Weird World: A Guide

ISSN 2197-1757 ISSN 2197-1765 (electronic)

Library of Congress Control Number: 2015950017

Springer Cham Heidelberg New York Dordrecht London

Printed on acid-free paper

Springer International Publishing AG Switzerland is part of Springer Science+Business Media

I was introduced to quantum information theory during my PhD studies in Renato

Sydney Marco Tomamichel

2 Modeling Quantum Information . . . . . . . . . . . . . . . . . . . . . . . . . . 11

3 Norms and Metrics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

4 Quantum Rényi Divergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

5 Conditional Rényi Entropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

6 Smooth Entropy Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

6.1.3 The Max-Entropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

7 Selected Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

Appendix: Some Fundamental Results in Matrix Analysis . . . . . . . . . . 127

As we further miniaturize information processing devices, the impact of quantum

© The Author(s) 2016 1

1.1 Finite Resource Information Theory

on Strassen’s techniques [148]—is motivated operationally: in many applications we

1 Thetopic has also been reviewed recently by Tan [151].

1.2 Motivating Example: Source Compression

Here, Hα (X )ρ is the Rényi entropy of order α, defined as

is the ε -smooth max-entropy, which is based on the Rényi entropy of order 21 .

1.3 Outline of the Book

generalization of Chernoff’s and Hoeffding’s bound on the success probability in

What This Book Does Not Cover

Abstract Classical as well as quantum information is stored in physical systems, or

In particular we will discuss concepts of functional and matrix analysis as well as

2.1 General Remarks on Notation

Table 2.1 Overview of notational conventions

2.2 Linear Operators and Events

2.2.1 Hilbert Spaces and Linear Operators

The operator norm on L (A, B) is defined as

For all L ∈ L (A, B), we have L < ∞ by definition. A linear operator is

K |v A = 0 =⇒ L |v A = 0 for all v ∈ H A . (2.3)

• We say K , L ∈ L (A) are orthogonal (denoted K ⊥ L) if K L = L K = 0.

• An isometry is an example of a contraction, i.e. an operator L ∈ L (A, B) sat-

For any L ∈ L (A), we denote by L −1 its Moore-Penrose generalized inverse

Similarly, we use its bra, denoted v| A , to describe the functional

v| A : H A → C, w A → v, w A . (2.5)

It is natural to view kets as linear operators from C to H A and bras as linear

|Lv A = L |v A , Lv| A = v| A L † , and v| A = |v†A . (2.6)

w| B L|v A = v| A L † |w B . (2.7)

As a further example, we note that |v A is an isometry if and only if v|v A = 1.

H= λx |ex ex | (2.9)

L= | f y f y | B L|ex ex | A = f y |L|ex · | f y ex |, (2.10)

where {ex }x is an ONB of A and { f y } y an ONB of B. This decomposes L into

The transpose of L with regards to the bases {ex } and { f y } is defined as

L T := f y |L|ex · |ex f y |, L T ∈ L (B, A). (2.11)

Here, L denotes the complex conjugate, which is also basis dependent.

2.2.2 Events and Measures

Observable events on a quantum system A correspond to operators in the unit

P• (A) := {M ∈ L (A) : 0 ≤ M ≤ I }. (2.13)

v| A : H A → C, w A → v, w A . (2.5)

A function x → M A (x) with M A (x) ∈ P• (A), x M A (x) = I A is called a

Tr A (·) : L (A) → C, L → ex | L |ex A . (2.16)