Instant Download Quantum Stochastic Thermodynamics: Foundations and Selected Applications (Oxford Graduate Texts) Philipp Strasberg PDF All Chapter

Full download test bank at ebook ebookmass.
com
Quantum Stochastic Thermodynamics:

Foundations and Selected
Applications (Oxford Graduate
Texts) Philipp Strasberg
CLICK LINK TO DOWLOAD
https://ebookmass.com/product/quantum-
stochastic-thermodynamics-foundations-and-
selected-applications-oxford-graduate-texts-
philipp-strasberg/
ebookmass.com
More products digital (pdf, epub, mobi) instant
download maybe you interests ...
Discrete Communication Systems (Oxford Graduate Texts)

Stevan Berber
https://ebookmass.com/product/discrete-communication-systems-
oxford-graduate-texts-stevan-berber/
Thermodynamics and Applications in Hydrocarbon Energy

Production Abbas Firoozabadi
https://ebookmass.com/product/thermodynamics-and-applications-in-
hydrocarbon-energy-production-abbas-firoozabadi/
Quantum Mechanics 3rd Edition Nouredine Zettili
https://ebookmass.com/product/quantum-mechanics-3rd-edition-
nouredine-zettili/
Introduction to Quantum Field Theory with Applications

to Quantum Gravity 1st Edition Iosif L. Buchbinder
https://ebookmass.com/product/introduction-to-quantum-field-
theory-with-applications-to-quantum-gravity-1st-edition-iosif-l-
buchbinder/
The Historical and Physical Foundations of Quantum
Mechanics Robert Golub
https://ebookmass.com/product/the-historical-and-physical-
foundations-of-quantum-mechanics-robert-golub/
Optimizations and Programming: Linear, Non-linear,

Dynamic, Stochastic and Applications with Matlab
Abdelkhalak El Hami
https://ebookmass.com/product/optimizations-and-programming-
linear-non-linear-dynamic-stochastic-and-applications-with-
matlab-abdelkhalak-el-hami/
Quantum Materials, Devices, and Applications Mohamed

Henini
https://ebookmass.com/product/quantum-materials-devices-and-
applications-mohamed-henini/
ChatGPT for Beginners: Features, Foundations, and

Applications 1st Edition Eric Sarrion
https://ebookmass.com/product/chatgpt-for-beginners-features-
foundations-and-applications-1st-edition-eric-sarrion/
Quantum Dots: Fundamentals, Synthesis and Applications

Rakshit Ameta
https://ebookmass.com/product/quantum-dots-fundamentals-
synthesis-and-applications-rakshit-ameta/
Quantum Stochastic Thermodynamics
Quantum Stochastic Thermodynamics
Foundations and Selected Applications
Philipp Strasberg
Universitat Autònoma de Barcelona
3
This book is dedicated to the memory of my PhD supervisor, Tobias Brandes
Preface
Recent decades have seen much progress in our understanding of thermodynamic pro-
cesses at the nanoscale. But nanoscale systems are very small, in most applications
far from equilibrium, often subject to strong fluctuations and sometimes even char-
acterized by exotic quantum properties—so it seems that these features rule out any
possibility of finding a consistent thermodynamic description for them.
It is the primary objective of this book to show that this is not the case. There is a
thermodynamic framework, characterized by a remarkable internal consistency, that
is able to describe nanoscale systems even under extreme conditions. Moreover, this
framework not only reaffirms common folklore around thermodynamics (there is no
perpetual motion machine, etc.), but also provides a wealth of beautiful results beyond
the traditional scope of thermodynamics—opening up the possibility of understanding
a plethora of different physical situations from a unified perspective.
The main title Quantum Stochastic Thermodynamics suggests that the present
book is about a synthesis of two research fields: classical stochastic thermodynamics
and quantum thermodynamics. Both have pushed the boundaries of the applicability of
the laws of thermodynamics by explaining and supplementing them with microscopic
considerations. For a considerable large class of nanoscale systems and processes, I
believe that most foundational questions have been settled by now. The present book
is intended to fill a gap in the literature by justifying this claim in detail for a large
variety of situations. I am also convinced that its content will prove important in
exploring new territories at the rapidly evolving frontiers of this field.
The subtitle Foundations and Selected Applications emphasizes that the reader can
mainly expect explanations about the basic theoretical pillars. These explanations are
intended to be pedagogically accessible. However, the book is also driven by the desire
to introduce a general and versatile framework characterized by conceptual clarity—
in complete awareness of the fact that this poses additional technical obstacles for
the beginner. To remedy this, a considerable effort has been made to transparently
explain common ‘jargon’ in the community (non-equilibrium entropies, local detailed
balance, Landauer’s principle, entropy production, time-reversal symmetry, the arrow
of time, etc.), which often appears unnecessarily mystifying (a problem that seems to
have a tradition in thermodynamics and statistical mechanics). The reader will often
find the same (or closely related) results derived in different ways in order to generate
confidence and trust in the framework.
The field of quantum stochastic thermodynamics is, however, fascinating, not only
because it allows us to addresss foundational questions about the nature of heat,
entropy or the second law, but also because it might have direct practical applications
in a world with increasing nanotechnological abilities. These applications could come
in the form of efficient thermoelectric devices, powerful energy harvesters, fast cooling
How to Read this Book vii
strategies, or energy-efficient computers, among other more exotic applications. To

form a close connection between foundational and practical problems, this book treats
a few selected applications in detail.
Unfortunately, because of a lack of time and understanding on the author’s side,
this book cannot cover all possible directions that are currently under investigation.
The selected material is obviously biased and I ask for forbearance from my many
colleagues who feel that the present book misses some important ideas.
How to Read this Book

This book is written for graduate students who know the basics of quantum mechanics
and equilibrium statistical mechanics and who cannot wait to combine these fields to
understand non-equilibrium phenomena. As a rule of thumb, you are ready to delve
into the book if the following equations do not scare you off:
∂ i
ρ(t) = − [H, ρ(t)],
∂t Xh̄
ρ= λn |nihn|,
n
Z(β) = tr{e−βH },
∂
U(β) = − ln Z(β),
∂β
SB = kB ln V,
1
f (ϵ) = β(ϵ−µ) .
e +1
Of course, this book is also written for more experienced researchers. I hope that they
will have no trouble in jumping between different sections of this book (though it might
help to first look at the Basic Notation section below). To become further acquainted
with the book, I summarize here its most important features.
Structure of the book: To get a complete picture, I believe one should sooner or
later read the entire book and, then, it would perhaps be most beneficial to go through
it in linear order. However, I also believe that the impatient reader should not be afraid
to skip sections or chapters. For instance, Chapter 1 (Quantum Stochastic Processes)
appears to be the most abstract one, in particular its second half. While I believe that
this more abstract point of view helps to view the entire field in a clear and unified
way, it is certainly not necessary to reach an understanding of Chapter 2 (Classical
Stochastic Thermodynamics), which solely requires some basic background knowledge
of the theory of classical stochastic processes. Likewise, readers with some familiarity
with open quantum system theory can directly start reading Chapter 3 (Quantum
Thermodynamics Without Measurements). To understand Chapter 4 (Quantum Fluc-
tuation Theorems), some background information from previous chapters is required.
For some sections of Chapter 5 (Operational Quantum Stochastic Thermodynamics)
it is necessary to have also read and understood the end of Chapter 1. Finally, three
appendices complement this book by providing information about topics that appear
viii Preface
at various points in the main text, but whose detailed exposition requires a longer
detour, which would blur the main narrative.
Structure of the sections: Typically, I have tried to start each section with
a small paragraph motivating its content and to end each section with a small sum-
mary or outlook. To facilitate orientation, some sections (in particular longer ones) are
divided into subsections using unnumbered subtitles. Furthermore, important state-
ments are distinguished by longer italic text and boxed equations highlight important
definitions or results.
Index: I have tried to make a long and informative index list. Words or phrases
appearing in this list are printed in a boldface font in the main text at the point
where they are first introduced or explained. However, as I said above, the book is
characterized by presenting similar concepts in different contexts and from different
perspectives. Thus, the book is not written as an encyclopaedia, but tries to maintain
a narrative that is most beneficial for pedagogical purposes.
Exercises: Various exercises are scattered throughout the text. These exercises, some-
times supplemented by (hopefully) helpful hints on how to solve them, should be rather
simple because I believe there is no benefit in torturing the reader. Exercises fall, how-
ever, into two categories. The first category of exercises are the short ones. They are
supposed to supply simple cross-checks for the reader or, by asking for derivations of
some equations, to acquaint the reader with standard mathematical manipulations in
the field. Then, there are also various longer exercises, which might even require some
simple symbolic programming. These longer exercises are typically meant to introduce
ideas, concepts or results whose detailed exposition would probably bore the more ex-
perienced reader in the field. Thus, the exercises help to keep the book more concise,
while allowing me at the same time to cover a wider range of topics. I remark that all
exercises appear during the text at the point where they best fit the overall narrative.
References: In contrast to the exercises, all references are relegated to a special

Further reading section at the end of each chapter. Having the overall pedagogical
purpose of the book in mind, I indeed believe that there is little benefit from mentioning
references during the main exposition of the material. Moreover, it ought to be clear
that, given the breadth and scope of the present book, it is impossible to give credit
to all contributions and, seen again from a pedagogical perspective, I believe there is
little benefit from trying to do so here. Hence, this book should not be confused with
a conventional ‘review article’. Most citations are given to the latest work, where the
main concepts or equations I rely on were first introduced or derived. Other citations
typically refer to expositions that go beyond the material presented here in a wider
sense (i.e. books, reviews or introductory articles about fields related to but not part of
quantum stochastic thermodynamics). Finally, some citations are added for historical
clarity. Thus, the overall idea is that the list of references provides a first orientation
for the newcomer, not an exhaustive list of contributions to the field.
Basic Notation ix
Basic Notation
I tried to keep abbreviations to a minimum and used only those that are widely
used in the literature. The three most important abbreviations, which are scattered
throughout the text, are CP (completely positive), CPTP (completely positive and
trace-preserving) and BMS (Born–Markov secular). In some equations, I write h.c. to
denote the Hermitian conjugate. Also the abbreviation POVM (positive operator-
valued measure) is used occasionally.
Below, I further provide a non-exhaustive list of the most important notation used
throughout the book.
General mathematics: Matrices are denoted by capital letters (such as M ). Bold-

face letters are used for vectors (e.g. a vector of probabilities p) and sequences (e.g.
a sequence of measurement results r), but their elements are written as, for exam-
ple, pj or rn . Multiplication of vectors and matrices is written without a dot (e.g.
M p). [A, B] = AB − BA and {A, B} = AB + BA denote the commutator and anti-
commutator, respectively. A superscript ∗, T or † denotes the complex conjugate,
transpose or the conjugate transpose, respectively. The imaginary unit is denoted i
and the symbol O(x) denotes that limx→0 |O(x)/x| < ∞. Total and partial derivatives
(e.g. with respect to time t) appearing in in-line equations are written as dt and ∂t ,
respectively. Frequently used functions include the Heaviside step function Θ(x), the
Kronecker delta δm,n and the Dirac delta function δ(x − y).
Quantum dynamics: I use Dirac notation for states |ψi and their conjugate transpose
hψ| that live in a Hilbert space H and its dual, respectively. The scalar product is
written as hϕ|ψi. To avoid unwanted technicalities, I assume Hilbert spaces are (or
can be approximated to be) finite dimensional: d = dim H < ∞. The density matrix
is typically denoted by ρ, whereas most other operators are denoted by capital letters
such as H, P, X, . . . and I is the identity. Exceptions are the familiar bosonic and
fermionic creation and annihilation operators (denoted typically by a(†) , b(†) , c(†) , d(†) )
and the Pauli matrices:

01 0 −i 1 0
σx = , σy = , σz = .
10 i 0 0 −1
The trace of some operator O is denoted tr{O}. Superoperators (also simply called
maps), which map operators onto operators, are always denoted by calligraphic let-
ters such as C, D, L, . . . . The tensor product is denoted ⊗ and additional subscripts
A, B, S . . . are used to indicate on which subspace some operator is acting (e.g. ρS ) or
to denote the partial trace (e.g. trB {. . . }).
Statistical mechanics: Equilibrium concepts are denoted by calligraphic letters

such as the equilibrium internal energy U or equilibrium entropy S. Exceptions are
parameters such as temperature T or chemical potential µ. Out-of-equilibrium quan-
tities are denoted by, for example, U and S if they refer to some expectation value
or ensemble average. Thermodynamic quantities defined along single stochastic tra-
jectories are denoted by small Latin letters (e.g. u and s). The canonical ensemble (or
x Preface
Gibbs state) is denoted by the Greek letter π. For instance, πS (β) denotes the Gibbs
state of some system S at inverse temperature β.
Finally, every process starts at the initial time t0 = 0. Moreover, I do not set Boltz-
mann’s and Planck’s constants kB and h̄ to 1. I believe that this makes the physical
content of many equations more insightful. For practical manipulations, this choice is,
of course, not the most convenient one, but since the majority of the literature sets
kB ≡ 1 and h̄ ≡ 1, I thought it would be good to keep them explicit here.
Acknowledgements
This book is dedicated to the memory of Tobias Brandes because—long before I en-
joyed being his PhD student—his inspiring and unprecedented lectures were the reason
why I actually turned towards theoretical physics. In my opinion, Tobias’s approach
to physics was dual, tackling deep and conceptual problems while having at the same
time an eye on experimentally well-grounded approaches and models, which also work
‘in practice’. I view my own research, and in particular this book, in the tradition of
his philosophy, perhaps with some bias more towards conceptual and general ideas. I
hope Tobias would have enjoyed reading it.
Concerning the topics exposed here I further owe much of my detailed knowledge
to Massimiliano Esposito and Gernot Schaller. Neither ever hesitated to share their
insights and ideas with me about various topics in statistical mechanics, stochastic
thermodynamics, non-equilibrium physics and open quantum systems. Much of their
knowledge is reflected here; yet, I believe I succeeded in also adding my own twist.
A special thanks goes to Kavan Modi. A short, but in retrospect important, dis-
cussion at the Kavli Institute in Santa Barbara in 2018 inspired me to look at the
problem from a different angle and part of the material presented in Chapter 5, and
in some sense the motivation to write this book, is a consequence of that discussion.
Many more colleagues have shared their insights and thoughts with me. Those on
whom I could particularly rely concerning the topics presented in this book include
Robert Alicki, Janet Anders, Felipe Barra, Victor Bastidas, Javier Cerrillo, Luis Cor-
rea, María García Díaz, David Gelbwaser-Klimovsky, John Goold, Giacomo Guarnieri,
Géraldine Haack, Christopher Jarzynski, Matteo Lostaglio, Mark Mitchison, Kavan
Modi, Wolfgang Muschik, Juan Parrondo, Martí Perarnau-Llobet, Matteo Polettini,
Andreu Riera-Campeny, Felix Ritort, Àngel Rivas, Dominik S̆afránek, Rafael Sánchez,
Anna Sanpera, Udo Seifert, Michalis Skotiniotis, Christopher Wächtler and Andreas
Winter. It makes me a bit sad to know that I probably forgot to correctly acknowledge
all the people who contributed to my actual understanding of this topic. There were
clearly many more people who influenced my thinking.
Furthermore, since writing a book is quite a ‘mammoth project’, I am grateful to
Javier Cerrillo, Marco Merkli, Kavan Modi, Andreu Riera-Campeny, Àngel Rivas and
Rafael Sánchez, who read (parts of) early drafts of the book and provided valuable
comments. In particular, a special thanks goes to Teresa Reinhard for making sure
that various parts of the book do not sound too cryptic to the outsider.
I also want to thank the team at Oxford University Press. In particular, I much
appreciated the uncomplicated manner of communicating with them, which clearly
contributed to the fact that the writing of this book was an overall very joyful expe-
rience.
xii Acknowledgements
Of course, the constant support I receive from my family and friends, including the
little lion, are invaluable.
Most parts of the work reported here were financially supported by the DFG
(project STR 1505/2-1). For further support I thank the Spanish Agencia Estatal
de Investigación, projects IJC2019-040883-I and PID2019-107609GB-I00, the Span-
ish MINECO FIS2016-80681-P (AEI/FEDER, UE) and the Generalitat de Catalunya
CIRIT 2017-SGR-1127.
Finally and almost needless to say, all mistakes and opinions expressed here are entirely
my own. I am always grateful to receive further comments, questions and feedback (to
get in contact with me I suggest you type my name into your preferred search engine
and look up my up-to-date email address). Errata will be announced on my homepage.
Contents
1 Quantum Stochastic Processes 1

1.1 Isolated Quantum Systems 1
1.2 System–Bath Theories and the Origin of Noise 3
1.3 Equilibrium States of Open Quantum Systems 8
1.4 Quantum Measurement Theory 12
1.5 Operations, Interventions and Instruments 17
1.6 Classical Stochastic Processes 22
1.7 Quantum Stochastic Processes 25
1.8 Quantum Markov Processes and Dynamical Maps 33
1.9 Classical Quantum Stochastic Processes 36
Further reading 41
2 Classical Stochastic Thermodynamics 43
2.1 Phenomenological Non-Equilibrium Thermodynamics 43
2.2 From Equilibrium Entropies to Landauer’s Principle 47
2.3 Classical Markov Processes and Local Detailed Balance 54
2.4 Fluctuating Internal Energy, Heat, Work and Entropy 61
2.5 Coarse-Graining and Time-Scale Separation 69
2.6 Time-Reversed Trajectories and Fluctuation Theorems 75
2.7 Work Fluctuation Theorems 81
2.8 Strong Coupling Corrections 89
2.9 Measuring Free Energies of Complex Molecules 97
Further reading 101
3 Quantum Thermodynamics Without Measurements 104
3.1 Mechanical Work in the Quantum Regime 104
3.2 Quantum Master Equations in the Weak Coupling Regime 108
3.3 Properties of the Born–Markov Secular Master Equation 116
3.4 Thermodynamics in the Born–Markov Secular Approximation 119
3.5 Work Extraction from Small Systems 124
3.6 Correlations and the Zeroth Law of Thermodynamics 130
3.7 Exact Dissipation Inequalities 135
3.8 Nonequilibrium Entropy, Entropy Production and Finite Baths 140
3.9 Non-Equilibrium Resources and Repeated Interactions 149
3.10 Particle Transport and Thermoelectric Devices 158
Further reading 172
4 Quantum Fluctuation Theorems 175
4.1 Two-Point Measurement Scheme 175
4.2 Work Fluctuation Theorems for Quantum Systems 179
xiv Contents
4.3 Exchange and Further Fluctuation Theorems 183

4.4 Counting Field Methods for Quantum Master Equations 189
4.5 Stochastic Quantum Jump Trajectories 196
4.6 Thermodynamic Uncertainty Relations 206
4.7 (Impossibility of) Carnot Efficiency at Finite Power 212
4.8 Third Law of Thermodynamics 215
Further reading 221
5 Operational Quantum Stochastic Thermodynamics 223
5.1 Classicality of Quantum Fluctuation Theorems 223
5.2 A No-Go Theorem 225
5.3 Thermodynamics of Quantum Measurements 229
5.4 Stochastic Thermodynamics of Quantum Markov Processes 235
5.5 Maxwell’s Demon and the Thermodynamics of Feedback Control 245
5.6 Autonomous Approach and Strong Coupling Corrections 252
5.7 Application to Photon Number Stabilization Experiments 262
Further reading 271
6 Outlook 273
Appendix A Concepts from Information Theory 277
A.1 Basic Concepts 277
A.2 Advanced Inequalities 281
Further reading 282
Appendix B Superoperators 283
B.1 Numerically Convenient Superoperator Mapping 283
B.2 The Choi–Jamiołkowski Isomorphism 285
B.3 Matrix Representations of the Process Tensor 287
Further reading 291
Appendix C Time-Reversal Symmetry 292
C.1 Time-Reversal Symmetry in Classical Mechanics 293
C.2 Time-Reversal Symmetry in Quantum Mechanics 297
C.3 The Arrow of Time 302
Further reading 305
References 306
Index 316
1
Quantum Stochastic Processes
Summary. This chapter describes the basic features of open quantum systems, i.e.
quantum systems that are affected by noise due to uncontrollable degrees of freedom
of an environment or bath. This noise is responsible for effects such as dissipation,
decoherence and irreversibility. We study the equilibrium states of open quantum
systems and review tools from quantum measurement theory, which describe how
to extract information from an (open) quantum system. We generalize these tools
to multi-time statistics and define the notion of a quantum stochastic process and
a quantum Markov process. Finally, we study in which cases a quantum stochastic
process looks classical.
1.1 Isolated Quantum Systems

Time-independent case
A quantum system is described by a Hilbert space H and a state ρ called the density
matrix, which acts on that space. The density matrix is characterized by the facts
that it is Hermitian, ρ† = ρ (with † denoting the Hermitian conjugate), has unit trace,
tr{ρ} = 1, and is positive, ρ ≥ 0. Here, the notation ρ ≥ 0 is shorthand for hψ|ρ|ψi ≥ 0
for all |ψi ∈ H.
The state of an isolated quantum system, i.e. a quantum system which is not in
contact with any other part of the world, obeys the Liouville–von Neumann equation
∂ i
ρ(t) = − [H, ρ(t)]. (1.1)
∂t h̄
Here, H is the Hamiltonian operator characterizing the total energy of the system and
[A, B] ≡ AB −BA is the commutator. Furthermore, i and h̄ are the familiar imaginary
unit and Planck’s constant, respectively. Note that we will also use the notation ∂t ρ(t)
to denote a partial derivative with respect to time.
Isolated quantum systems have some important characteristics:
(i) There exists a unitary time evolution operator U (t) ≡ exp(−iHt/h̄), which means
that U (t)U (t)† = U (t)† U (t) = I, where I denotes the identity matrix. This time
evolution operator propagates the system state according to
ρ(t) = U (t)ρ(0)U (t)† , (1.2)
where ρ(0) denotes the initial state of the quantum system.
Quantum Stochastic Thermodynamics. Philipp Strasberg, Oxford University Press.

© Philipp Strasberg (2021). DOI: 10.1093/oso/9780192895585.003.0001
2 Quantum Stochastic Processes
(ii) The spectrum of the state ρ(t), i.e. its eigenvalues λk , do not change in time:
X
ρ(t) = λk |ψk (t)ihψk (t)|. (1.3)
k
Here, λk is time independent and |ψk (t)i belongs to an orthonormal basis of

wave functions, i.e. hψk (t)|ψℓ (t)i = δk,ℓ with the Kronecker delta δk,ℓ . Note
that the spectrum can be degenerate, i.e. it is possible that λk = λℓ for some
k 6= ℓ. Equation (1.3) follows from the fact that any unitary transformation of the
form (1.2) leaves the spectrum invariant since the characteristic polynomial does
not change:
det{ρ(t) − λI} = det{U (t)[ρ(0) − λI]U (t)† } = det{ρ(0) − λI}. (1.4)
Here, det{. . . } denotes the determinant of a matrix.

(iii) The purity of ρ(t), which measures the ‘mixedness’ of a state and is defined as
tr{ρ2 }, is conserved. This follows immediately from point (ii) above or, alter-
natively, from eqn (1.1) and the fact that the trace is cyclic, i.e. tr{ABC} =
tr{CAB}. In particular, if the initial state is pure, i.e. tr{ρ(0)2 } = 1, we have
ρ(t) = |ψ(t)ihψ(t)| for some wave function |ψ(t)i. This wave function evolves in
time according to the familiar Schrödinger equation:
∂ i
|ψ(t)i = − H|ψ(t)i. (1.5)
∂t h̄
Since the Schrödinger equation can be applied to compute the time evolution of
any of the |ψk (t)i in eqn (1.3), it is equivalent to the Liouville–von Neumann
equation (1.1).
(iv) The von Neumann entropy of the system, which is defined as
SvN (ρ) ≡ −tr{ρ ln ρ}, (1.6)
is P
constant in time. This follows again from point (ii) above because SvN [ρ(t)] =
− k λk ln λk , which implies that the von Neumann entropy quantifies the clas-
sical uncertainty about the state of the system. Alternatively, the conservation of
von Neumann entropy follows by using

d dρ(t)
SvN [ρ(t)] = −tr ln ρ(t) (1.7)
dt dt
and eqn (1.1). Note that eqn (1.7) holds only if the rank of the density operator
does not change during the evolution, which is guaranteed by point (ii) above. If
the rank changes in time, the von Neumann entropy is not differentiable. Note
that we use a subscript ‘vN’ for the von Neumann entropy throughout the book
because we want to distinguish it from the notion of thermodynamic entropy
introduced later on. Readers unfamiliar with information-theoretic concepts of
entropy can find an overview in Appendix A.
System–Bath Theories and the Origin of Noise 3
Time-dependent case
Quantum systems are often subjected to time-dependent fields in a laboratory, and
these fields can be treated semiclassically (e.g. laser light). In this case, the Hamil-
tonian becomes time dependent and is denoted by H(λt ), where the time-dependent
parameter λt specifies the external fields. We prefer the notation H(λt ) instead of
H(t) to avoid any possible confusion with the Heisenberg picture. In the following, λt
is called a driving or control protocol and the state of such a driven system evolves
in time according to the Liouville–von Neumann equation (1.1) with H replaced by
H(λt ).
The evolution ρ(t) = U (t, 0)ρ(0)U † (t, 0) is still described by a unitary operator
U (t, 0), but its explicit computation is now more complicated. Formally, we have
∞
X n Z t Z t1 Z tn−1
i
U (t, 0) = − dt1 H(λ1 ) dt2 H(λ2 )· · · dtn H(λn )
n=0
h̄ 0 0 0
X∞ n Z t Z t
1 i
= − dt1 · · · dtn [H(λ1 ) . . . H(λn )]+ (1.8)
n=0
n! h̄ 0 0
Z
i t
≡ exp+ − dsH(λs ) ,
h̄ 0
where we abbreviated λtj ≡ λj and the subscript + denotes time ordering. By dividing
the time interval [0, t] into n = t/δt small steps δt, we can also write
Y
n−1
U (t, 0) ≈ e−iH(λj )δt/h̄ , (1.9)
j=0
which becomes exact in the limit n → ∞.

Strictly speaking, a driven quantum system is not isolated as it is in contact with
the external driving field. However, it is easy to check that points (i)–(iv) above are still
satisfied for such a driven system. In contrast, the time evolution of a quantum system
which interacts with another quantum system is markedly different, as we will see in
the rest of this chapter. Therefore, it seems wise to call a quantum system ‘isolated’
as long as it evolves in time according to the Liouville–von Neumann equation (1.1),
whether the Hamiltonian is time independent or not.
1.2 System–Bath Theories and the Origin of Noise

In reality, nobody has ever observed an isolated quantum system at the very end
because the mere act of ‘observing’ a quantum system requires it to be coupled with an
external detector. We will, however, postpone the discussion of quantum measurement
theory to Section 1.4 and are here rather concerned with the fact that many quantum
systems interact with uncontrollable degrees of freedom of a so-called environment or
bath. One example is an atom (the ‘system’) interacting with the many electromagnetic
modes of the surrounding space (the ‘environment’). Another example is a single spin,
e.g. an impurity in a metal or crystal, interacting with many remaining spins and
Fig. 1.1 Open quantum systems. (a) A rough sketch of a system S in contact with a bath B
with which it can exchange energies, particles, entropy, etc. (b) Diagram of the Newcomen at-
mospheric steam engine as an example of an ancient ‘open quantum system’: thermodynamics
had already divided the universe into a system part and reservoirs. (c) (False-coloured) scan-
ning electron microscope image of a modern open quantum system: a quantum dot formed by
a nanowire (thin green line) in contact with electron reservoirs (yellow) and heaters (blue and
red). Such setups will be treated in further detail in Section 3.10. Reprinted by permission
from Springer Nature: Springer Nature, Nature Nanotechnology (A quantum-dot heat engine
operating close to the thermodynamic efficiency limits, M. Josefsson et al.), Copyright by
Springer Nature (2018).
phonons (lattice vibrations) in the surroundings. Finally, the historical origin of the
theory of thermodynamics is rooted in the desire to understand, for example, steam
in a container in contact with hot air produced by burning coal on the one side and
cold water on the other side. Here, the system is defined by the container, which
contains the so-called working medium or working fluid, whereas the outside hot air
and cold water are called a heat bath or reservoir. We will also use these words from
Chapter 2 onwards. In this chapter, however, we use the broader term ‘environment’
or ‘bath’ to refer to any external, uncontrollable part of the world, not necessarily
described by a thermodynamic variable such as temperature. It will become clear
throughout this book that the predictive power of the second law comes from an
efficient description of these uncontrollable degrees of freedom about which we have
only very little information.
Quantum systems which are not isolated but in contact with a bath or environ-
ment are called open quantum systems; see Fig. 1.1 for sketches. Conceptually, many
different approaches exist to describe them theoretically and a very powerful one is
to model the bath itself as another quantum system such that the system and the
bath (which we sometimes also call the universe) constitute one big isolated quantum
system. This is the origin of system–bath theory and we will see below that this is
indeed not an assumption: any open quantum system can be seen as being part of a
larger isolated quantum system.
Mathematically, the system–bath composite is a bipartite quantum system de-
scribed by the tensor product of the system and bath Hilbert space: HS ⊗ HB . The
dimension of that space is dim(HS ⊗ HB ) = dim HS · dim HB . For many applications
the dimension of HS is very small, e.g. dim HS = 2 for a single spin, whereas the
dimension of HB is often very large, e.g. dim HB = 2NA , where the Avogadro number
NA is of the order of 1023 . The dynamics of the system and the bath is governed by
the Hamiltonian HSB = HS ⊗ IB + IS ⊗ HB + VSB . Here, HS and HB denote the
Hamiltonian of the isolated system or bath, respectively, and VSB denotes their in-
teraction. They sum up to the total Hamiltonian HSB . Whereas the system and bath
Hamiltonian commute (since they live on different Hilbert spaces) we have in general
6 0 and [VSB , HB ] 6= 0. In the following, we suppress tensor products with
[VSB , HS ] =
the identity for notational simplicity and write the system–bath Hamiltonian as
HSB = HS + HB + VSB . (1.10)
For simplicity, we assumed no driving λt here, but this will change in later chapters.
We remark that the tensor product structure of the system–bath composite assumes
the system and bath to be distinguishable objects and implies a particular choice of
gauge made when identifying what is the ‘system’ and what is the ‘bath’.
Since the system–bath composite is isolated, its global state described by the den-
sity operator ρSB (t) evolves according to the Liouville–von Neumann equation (1.1)
with H replaced by HSB . The system evolution is obtained by taking the partial trace:
ρS (t) = trB {ρSB (t)}. (1.11)

In contrast to the isolated case, the evolution of an open quantum system is signifi-
cantly different. None of the four points mentioned in Section 1.1 remain true:
(i′ ) The time evolution of ρS (t) is not described by a unitary operator.
(ii′ ) The eigenvalues of the system state change in time, i.e. eqn (1.3) is replaced by
X
ρS (t) = λk (t)|ψk (t)iS hψk (t)|. (1.12)
k
′
(iii ) Since the eigenvalues of ρS (t) change, the purity of the state can change too. In
particular, an initially pure state becomes mixed in general.
′
(iv ) The von Neumann entropy is no longer conserved. Note that the von Neumann
entropy of the system can become larger or smaller during the evolution, without
violating the second law of thermodynamics. The second law states only that
the thermodynamic entropy of the universe, i.e. the system and the bath, cannot
become smaller in time. It does not imply that the von Neumann entropy of the
system state cannot decrease.
Illustrative example
We illustrate the above arguments by considering an assembly of n interacting spins.
Readers unfamiliar with open quantum systems are invited to explicitly follow this
example by doing their own numerics for it. We assume that the spins are described
by the following global Hamiltonian:
h̄Ω X (i) h̄ X X
n n
HSB = σ + gij σx(i) σx(j) , (1.13)
2 i=1 z 2 i=1 j>i
(i)
where σα (α = x, y, z) are the familiar Pauli matrices acting on spin i. The first term
describes n isolated spins with energy gap h̄Ω. The second term describes the interac-
tion between two spins with coupling strength gij . As our system we now choose one of
(1)
the spins, say the first, such that HS = h̄Ωσz /2. The bath Hamiltonian consequently
Pn (i) Pn P (i) (j)
becomes HB = h̄Ω i=2 σz /2 + h̄ i=2 j>i gij σx σx /2 and the remaining part
defines the interaction Hamiltonian VSB . The initial state of the system and bath is
assumed to be decorrelated:
ρSB (0) = ρS (0) ⊗ ρB (0). (1.14)
√ simulations, we take ρS (0) = |+ih+|S to be a pure state. Here, |+i ≡ (|0i +

In the
|1i)/ 2 denotes a coherent superposition of the eigenstates of σz (with σz |0i = −|0i
and σz |1i = +|1i), which coincide with the energy eigenstates of HS . The bath instead
is taken to be a canonical equilibrium (Gibbs) ensemble with respect to the inverse
temperature β, denoted by
e−βHB
ρB (0) = πB ≡ , ZB ≡ trB {e−βHB }, (1.15)
ZB
where ZB is the partition function of the bath. Therefore, we assume the following
situation: previous to the initial time the first spin is decoupled from the others and
prepared in a superposition of energy eigenstates. Then, at time t = 0 we suddenly
switch on the interaction with the bath, which is assumed to be thermalized.
We aim at a numerical exact simulation of the system–bath dynamics based on the
Liouville–von Neumann equation (1.1). For n spins the total density matrix as well
as the unitary time evolution operator are 2n × 2n matrices. This exponential growth
in size necessarily limits us to consider only a few spins, precisely we consider seven
spins in total (i.e. the bath consists of six spins). Note that one is often interested in a
bath that is much larger in size. Computing the exact dynamics of an open quantum
system then quickly becomes impossible, which motivates the need for efficient and
reliable approximation schemes. For the moment, however, a small bath suffices to
illustrate our points above. To continue with the discussion of our model, let us choose
the spin–spin interactions gij random from a uniform distribution over [0, 1]. We do
not explicitly write down the values for gij here because the dynamics are qualitatively
similar for most choices.
Numerical results for an initial (dimensionless) bath temperature of βh̄Ω = 10 are
shown in Fig. 1.2. Note that this bath temperature is relatively cold, i.e. the energy
gap h̄Ω of each single spin is 10 times larger than the typical energy kB T = β −1 of
a thermal excitation. Figure 1.2a shows the time evolution of the expectation value
(1) (1)
hσx i(t) = tr1 {σx ρS (t)}. For better comparison we also plot its time evolution in
the case that the spin was isolated, i.e. for VSB = 0 (thin grey line). Their difference
is quite striking; in particular, it is difficult to recognize any structure or pattern for
the open quantum system case. Figure 1.2b shows the purity, which decreases as
expected. Note that the minimal value of the purity for a two-level system is 1/2.
Figure 1.2c shows the evolution of the von Neumann entropy, which is no longer
conserved. Note that the maximum value for the von Neumann entropy of a two-level
Fig. 1.2 Exemplary time evolution of an open quantum system for a ‘cold’ bath. The thin
grey line in (a) shows the time evolution of an isolated system with an identical system Hamil-
tonian. The thin grey line in (d) shows the time evolution of the quantum relative entropy
with respect to a reference state, which describes a refined equilibrium state introduced in
Section 1.3.
system is ln 2 ≈ 0.7. In Fig. 1.2d we show the time evolution of a quantity that
plays an important role throughout the book. It is known as the quantum relative
entropy, which is defined in general as
D(ρ|σ) ≡ tr{ρ(ln ρ − ln σ)} ≥ 0 (1.16)
for two arbitrary density matrices ρ and σ. The relative entropy is non-negative and
only zero if ρ = σ. It can be regarded as a measure of statistical ‘distance’ between ρ
and σ; further information is provided in Appendix A. In particular, we plot

e−βHS
D ρS (t) , (1.17)
ZS
i.e. the distance between the reduced system state and its associated Gibbs state at the
inverse temperature of the bath. One could naively expect that this difference should
become very small for long times in accordance with equilibrium statistical mechanics.
This is not the case for two important reasons. First, a bath of only six spins is still
too small to induce equilibration of a small quantum system. Second, even if the bath
were larger, equilibrium statistical mechanics in fact predicts a state different from
the Gibbs state, as we will explain in detail in the next section. The time evolution of
the relative entropy with respect to this different state is shown by the thin grey line,
which is much closer to zero than eqn (1.17).
Finally, Fig. 1.3 shows the same plots as Fig. 1.2, but for a different initial tempera-
ture of the bath. This time we chose βh̄Ω = 1, such that the thermal excitation energies
Fig. 1.3 Exemplary time evolution of an open quantum system for a ‘hot’ bath.
are comparable with the energy gap of each single spin. Equivalently, one could say
that we have more ‘noise’ in the bath. The difference compared with the cold case
βh̄Ω = 10 is quite striking. Oscillations are much less pronounced and eqn (1.17) be-
comes quite small after some transient time. This means that the standard equilibrium
Gibbs ensemble describes the system state quite well. In fact, the system is well de-
scribed by a completely mixed state ρS (t) ≈ IS /2, which follows by considering the
time evolution of the purity.
To conclude, an open quantum system behaves very differently from an isolated
quantum system. Even for a very small bath of only six spins effects such as equilibra-
tion and thermalization start to become visible. Next, we consider equilibrium states
of open quantum systems in the case that the bath is much larger.
1.3 Equilibrium States of Open Quantum Systems

Before we analyse equilibrium states of open quantum systems, it might be worth
asking whether an open quantum system can actually reach an equilibrium state.
In fact, we have introduced open quantum systems via the system–bath paradigm,
i.e. a composite system that evolves unitarily in time. This implies that, as long
as the system–bath composite can be described by a finite-dimensional Hilbert space,
the dynamics is quasi-periodic. By this we mean the following. Let |ψ(0)i denote
the initial state of an arbitrary isolated system, which is here assumed to be pure
for simplicity. We denote the eigenvalues and eigenstates of its Hamiltonian
P H by En
and |ni, respectively. Then, if we
P expand the initial state as |ψ(0)i = n cn |ni with
complex coefficients cn obeying n |cn |2 = 1, we can write the state at time t as
X
|ψ(t)i = cn e−iEn t/h̄ |ni, (1.18)
n
Equilibrium States of Open Quantum Systems 9
i.e. it is a sum of periodic functions with frequencies ωn ≡ En /h̄. If we wait long

enough, we might wonder whether there exists some time t and a set of natural numbers
{kn } such that ωn t = 2πkn for all n. This would then imply that hψ(0)|ψ(t)i = 1, i.e.
the system returned back to its initial state. If all frequencies are rational numbers, we
can straightforwardly find such a time t. By assumption we can then Q write ωn = qn /rn
for some qn ∈ N and rn ∈ N. In particular, if we define N ≡ n rn , we can write
ωn = mn /N for all n and some mn ∈ N. Thus, by choosing t = 2πkN , we obtain
hψ(0)|ψ(t)i = 1 for all k ∈ N, i.e. the quantum system returns infinitely often to its
initial state.
Things start to become more subtle if the ratio ωn /ωn′ of some pair of frequencies
becomes irrational. Indeed, it then never happens that hψ(0)|ψ(t)i = 1. However,
irrational numbers can be approximated arbitrarily well by rational numbers. Hence,
it seems reasonable to expect from the foregoing argument that the system returns
after a sufficiently long time to its initial state up to a very small error. Indeed, one can
prove that an isolated finite-dimensional quantum system returns arbitrarily close to
its initial state infinitely often, i.e. for any ϵ > 0 there exist infinitely many times t such
that |hψ(0)|ψ(t)i| = 1 − ϵ. This is known as quasi-periodicity. This result is also known
from classical mechanics as Poincaré’s recurrence theorem. Obviously, since an
open quantum system is a part of a larger isolated system, this also implies that any
open quantum system state returns infinitely often arbitrarily close to its initial state.
This seems to imply that we should not expect an open quantum system to equilibrate
at all and Poincaré’s recurrence theorem has played an important historical role in
the debate on whether it is possible to derive the laws of thermodynamics from an
underlying microscopic perspective: If all states return arbitrarily close to their initial
state, why do we observe any irreversibility?
The key insight to resolve this question lies in the fact that the next time t for which
|hψ(0)|ψ(t)i| = 1−ϵ happens grows incredibly fast with the number of coefficients cn in
eqn (1.18). This number is expected to scale with the number of particles in the system
because it is experimentally extremely unlikely to prepare a many-body system in a
state |ψ(0)i, which contains only a few energy eigenstates |ni. Then, if the isolated
system is composed of N particles, e.g. the N spins in the previous example, one
generically expects the average recurrence time to scale double exponentially with N ,
i.e. t = O{exp[exp(N )]}. If one assumes that the number of particles is typically of the
order of 1023 , the problem of Poincaré recurrences becomes irrelevant for all human
time scales.
Therefore, we assume for now that the system equilibrates and reaches a stationary
state limt→∞ ρS (t), where the notation t → ∞ means that we consider times much
larger than typical time scales of the open system evolution, but still, of course, smaller
than the Poincaré recurrence time. The next question is then: What is the equilibrium
state of the open quantum system? Finding a complete answer to that question turns
out to be difficult and is still the subject of intense research. Since this is not the
topic of the book, we here restrict ourselves to invoking arguments from equilibrium
statistical mechanics, which work well in many cases.
For this purpose, we consider the familiar Gibbs state or canonical ensemble,
defined for any system with Hamiltonian H by
e−βH
π≡ , Z ≡ tr{e−βH }. (1.19)
Z
Here, β = (kB T )−1 is the inverse temperature and Z is the partition function. If the
internal energy U = tr{Hπ} is fixed, the Gibbs state is characterized by the fact that it
maximizes the von Neumann entropy with the inverse temperature implicitly defined
through the relation U = −∂β ln Z. Vice versa, for a fixed von Neumann entropy
SvN (π), the Gibbs state minimizes the energy with the inverse temperature implicitly
defined through the relation SvN (π) = −β 2 ∂β (β −1 ln Z).
Now, suppose that the system–bath composite is well described by a global Gibbs
state πSB = e−βHSB /ZSB with ZSB ≡ trSB {e−βHSB }. What does the reduced system
state trB {πSB } look like? Perhaps surprisingly, it turns out that in general trB {πSB } 6=
πS = e−βHS /ZS . Only in the case of very weak coupling between the system and the
bath, i.e. if VSB is negligible compared with HS and HB , does the reduced system state
equal the conventional canonical ensemble. For non-negligible coupling VSB , however,
we define πS∗ ≡ trB {πSB } and we have πS∗ 6= πS .
Nevertheless, it is still possible to write πS∗ in an apparent Gibbs form with respect
to an effective Hamiltonian H̃S :
e−β H̃S
πS∗ = trB {πSB } = . (1.20)
Z̃S
In fact, this is possible for every density matrix ρS by defining H̃S ≡ −β −1 ln(Z̃S ρS ),
and not only for πS∗ . Notice that H̃S is fixed only up to an arbitrary choice of the
positive normalization constant Z̃S . For the reduced state πS∗ of a canonical equilibrium
ensemble there is one convenient choice for the effective partition function, which we
denote by ZS∗ = Z̃S and which is obtained by setting
ZSB trSB {e−βHSB }

ZS∗ ≡ = . (1.21)
ZB trB {e−βHB }
That is, the effective partition function ZS∗ is the ratio of the partition functions of the
system–bath composite and of the bath alone. The corresponding effective Hamiltonian
is known as the Hamiltonian of mean force and reads explicitly
1 1 trB {e−βHSB }
HS∗ = − ln(ZS∗ πS∗ ) = − ln . (1.22)
β β ZB
∗
One quickly verifies that trB {πSB } = e−βHS /ZS∗ . The Hamiltonian of mean force pro-
vides a neat concept used various times in the following chapters. Physically speaking,
it can be seen as an effective free energy landscape for the system, a claim that becomes
clearer in Chapter 2. Notice that the Hamiltonian of mean force depends explicitly on
the inverse temperature β. For weak coupling VSB , HS∗ reduces to HS . Thus, the
familiar canonical ensemble only emerges in the weak coupling regime.
It is natural to ask: Does the effective Gibbs state (1.20) represent a good approx-
imation of the equilibrated open quantum system state, i.e. is πS∗ ≈ limt→∞ ρS (t)?
Equilibrium States of Open Quantum Systems 11
If that is the case, we say that the system thermalizes, which is a stronger require-
ment than equilibration alone, which only assumes the system state to become time
independent for long times. Paralleling the objections raised by Poincaré’s recurrence
theorem above that equilibration can never strictly happen, similar doubts may rise for
the case of thermalization. In particular, no initial system–bath state ρSB (0) different
from πSB will ever reach the latter state during the evolution.
Exercise 1.1 Show that, if ρSB (0) 6= πSB and if HSB is time independent, ρSB (t) 6= πSB
for all t.
However, similar to the case of equilibration, thermalization should also be re-

garded as a convenient illusion by recalling that our experimental capabilities are
limited. For a bath with its prosaic 1023 degrees of freedom, the full system–bath state
23
ρSB (t), which is defined on a Hilbert space of dimension of the order of 1010 , remains
experimentally inaccessible. Instead, one typically has access only to a restricted set
of observables, for instance those determined by looking at the open quantum system.
Thermalization to eqn (1.20) is then very likely to happen as long as one can mean-
ingfully associate some macroscopic temperature T with the bath. Exceptions are, for
instance, a bath consisting of two parts kept at different temperatures (a scenario that
becomes relevant in later chapters when studying transport processes) or a bath pre-
pared in a giant Schrödinger cat state as a superposition of two very different energies
(an unlikely scenario). Furthermore, also in the presence of additional conservation
laws, i.e. if some observables commute with the global Hamiltonian, thermalization
to eqn (1.20) will typically not happen. However, if these scenarios can be excluded,
many open quantum systems thermalize to eqn (1.20) regardless of the initial state
ρS (0). In fact, this observation is confirmed by our example at the end of Section 1.2.
The grey line in Figs. 1.2d and 1.3d shows the time evolution of D[ρS (t)|πS∗ ]. We see
that, even for a bath of only six spins, πS∗ can represent a good approximation to
the open system state, even at low temperatures, i.e. in the regime where quantum
fluctuations dominate thermal fluctuations.
The last observation raises the question of whether the deviation of πS∗ from the
standard Gibbs state πS is mainly caused by quantum effects. In general, this does not
need to be the case, but we will return to this question more rigorously in Section 3.6
when we discuss the zeroth law of thermodynamics. For now, we conclude this section
by studying an important class of models for which there is a clear difference between
πS∗ and πS in the quantum and classical regime.
Exercise 1.2 Consider a harmonic oscillator with frequency ω and Hamiltonian (in mass-
weighted coordinates) HS = 21 (p2S +ω 2 x2S ) coupled to a ‘bath’ of one other harmonic oscillator
with the same frequency ω. The global Hamiltonian is
2
1 2 c
HSB = HS + pB + ω 2 xB − 2 xS . (1.23)
2 ω
Here, the coupling strength is c and we have written the Hamiltonian in a manifestly positive
form such that HSB ≥ 0 for any choice of ω and c. Now, show first that, if treated as a classical
system, there is no correction to the thermal state of the system: πS∗ = πS = e−βHS /ZS . Hint:
Remember that the trace over the bath degrees of freedom becomes classically an integral
over the phase-space coordinates (xB , pB ) of the bath.
Then, show that quantum mechanically this is no longer true. Hint: You are not asked
to directly compute HS∗ , but only to show that πS∗ 6= πS . This can be done in various ways.
One way is to compute the equilibrium variance of the system coordinate tr{x2S πSB } by
changing
√ to normal modes. Denote the (squared) eigenfrequencies of HSB by Ω2± ≡ [c2 +
2ω ± c + 4c ω ]/2ω and show that
4 4 2 4 2
eβh̄(Ω+ +Ω− ) − 1
tr{x2S πSB } = h̄ . (1.24)
(Ω+ + Ω− )(eβh̄Ω+ − 1)(eβh̄Ω− − 1)
Confirm that in the classical limit (h̄ → 0) this result is independent of the coupling strength
c. Also confirm that for c → 0 you recover the result trS {x2S πS } = h̄ coth(βh̄ω/2)/2ω, showing
that even at zero temperature quantum fluctuations give rise to a non-zero variance.
The above result can be generalized to a system–bath Hamiltonian of the form
" 2 #
1X 2 ck
HSB = HS + pk + ωk xk − 2 S
2
. (1.25)
2 ωk
k
This describes a system with arbitrary Hamiltonian HS coupled via an arbitrary system
coupling operator S to a bunch of harmonic oscillators. Such system–bath models, which are
well justified whenever the statistics of the bath are (approximately) Gaussian, are known as
Caldeira–Leggett models and they play an important role in the theory of open quantum
systems. Convince yourself of the fact that for a classical system the Hamiltonian of mean
force is identical to the system Hamiltonian: HS∗ = HS . Obviously, this no longer holds
for quantum systems. Computing the exact reduced equilibrium state of the open quantum
system is possible, but it is no longer a triviality.
1.4 Quantum Measurement Theory

We now start to focus on the question of how to obtain information about a quantum
system by considering measurements at a single point in time. For that purpose it is
sufficient to simply consider an arbitrary system state ρS , neglecting for a moment
the possible presence of a bath. We will come back to the question of how to treat
multi-time measurements under the influence of a bath in Section 1.6 onwards.
Projective measurements
In quantum mechanics the measurement Pn of a system observable XS is described by
the projection postulate. Let XS = x=1 λ(x)ΠS (x) be the spectral decomposition
of the observable with n different eigenvalues λ(x), which are labelled by the index
x. Furthermore,P{ΠS (x)} is a set of projection operators satisfying ΠS (x)ΠS (x′ ) =
n
δx,x′ ΠS (x) and x=1 ΠS (x) = IS . If the state of the system is ρS , then the probability
of observing result x is p(x) = trS {ΠS (x)ρS }. The post-measurement state ρ′S (x)
conditional on observing x becomes
ΠS (x)ρS ΠS (x)
ρ′S (x) = . (1.26)
p(x)
The average or unconditional post-measurement state ρ′S of the system follows as

Quantum Measurement Theory 13
X
n X
n
ρ′S = p(x)ρ′S (x) = ΠS (x)ρS ΠS (x). (1.27)
x=1 x=1
Notice that this state is in general different from the pre-measurement state, ρ′S 6=
ρS , which reflects the well-known fact that quantum measurements are disturbing.
This measurement backaction is absent only if the initial state commutes with the
observable: [ρS , XS ] = 0. Even in this case, however, the conditional state (1.26) is
in general different from the pre-measurement state, i.e. ρ′S (x) 6= ρS . This is not a
quantum effect but a consequence of the fact that the observer updates its state of
knowledge about the system. Readers who feel unfamiliar with this description are
advised to do the following exercise.
Exercise 1.3 Consider the case where all projectors ΠS (x) = |xihx|S are of rank 1 and the
system state is pure: ρS = |ψihψ|S . Convince yourself of the fact that the above framework
then reduces to the conventional picture known from introductory textbooks in quantum
mechanics. What corresponds to the Born rule? Which equation describes the ‘collapse’ of
the wave function? When does the measurement reveal no information, i.e. when do we have
ρ′S (x) = ρS ? Can you think about physical examples where the projectors are not of rank 1?
Remarkably, it is possible to derive eqn (1.27), which describes the average effect
of a quantum measurement, using only unitary dynamics and no projection postulate.
This works as follows. Consider a second n-dimensional ‘auxiliary’ quantum system
A, called the ancilla in the following, which interacts with our system S of interest.
Imagine that the initial system–ancilla state is decorrelated, ρSA = ρS ⊗ ρA , and the
ancilla is prepared in some fixed pure state ρA = |1ih1|A . Furthermore, assume that
the overall system–ancilla interaction is described by the following unitary operator:
X
n X
n
V = ΠS (x) ⊗ |r + x − 1ihr|A , (1.28)
x=1 r=1
where we interpret r + x − 1 modulo n whenever r + x − 1 > n. It is easy to verify

that V is unitary, satisfying, V V † = V † V = ISA . The joint system–ancilla state after
the interaction consequently reads
ρ′SA = V ρS ⊗ ρA V † . (1.29)
The reduced state of the system is obtained by tracing out the ancilla,
X
n
ρ′S = trA {V ρS ⊗ ρA V } = †
ΠS (x)ρS ΠS (x), (1.30)
x=1
which is identical to eqn (1.27). Thus, the average effect of a system measurement arises
in this picture from interactions of the system with an outside ancilla. By tracing out
the ancilla, we lose information and the reduced dynamics is no longer unitary.
To derive the conditional post-measurement state (1.26) in this picture, assume
that we subject the ancilla to a projective measurement of the observable RA =
Fig. 1.4 Circuit representation of an ancilla-assisted measurement with time running from
left to right. The input states to the process (left, triangles) are a system and ancilla state ρS
and ρA . Both interact via a unitary transformation, denoted here with a calligraphic symbol
V. The joint state ρ′SA after the interaction is in general correlated. To finally infer something
about the system, the experimentalist measures the state of the ancilla, here with projector
|rihr|A . The conditional post-measurement state of the system is denoted ρ′S (r).
Pn
r=1 r|rihr|A . Suppose this projective measurement of RA reveals outcome r, then the
conditional post-measurement state reads ρ′SA (r) = |rihr|A V ρS ⊗ ρA V † |rihr|A . Keep-
ing the information about r but tracing out the ancilla degrees of freedom reveals
ρ̃′S (r) ≡ trA {|rihr|A V ρS ⊗ ρA V † } = ΠS (r)ρS ΠS (r). (1.31)

After identifying x ≡ r, we obtain eqn (1.26) up to normalization. In fact, the trace
of the non-normalized state ρ̃′S (r) equals the probability of obtaining the measure-
ment result r: p(r) = trS {ρ̃′S (r)} = trS {ΠS (r)ρS }. Hence, the normalized state reads
ρ′S (r) = ρ̃′S (r)/p(r), in agreement with eqn (1.26).
As we will see throughout the remainder of this book, ancillas provide us with
a flexible mathematical tool to think about quantum measurements and much more.
They can be seen as comprising the essential features of an environment in an abstract
and minimal way. Physically, the use of ancillas can be motivated as follows. In order to
find out something about a system, an experimentalist prepares an external probe and
puts it into contact with the system. Then, both start to interact and, depending on
the state of the system, the system imprints some information onto the probe. Finally,
the experimentalist reads out the state of the probe in order to infer something about
the system. This picture is often justified by recognizing that an experimentalist has
precise control about the detectors in the laboratory—how they are prepared, how they
intervene with the system and how they are read off—whereas the system itself is given
by nature and can be only indirectly accessed via the detectors found in the laboratory.
An experiment in which this picture becomes particularly transparent will be treated
in detail at the end of this book in Section 5.7. A pictorial representation of this
process is provided in Fig. 1.4. For the rest of this section we focus on investigating the
process in Fig. 1.4 in greater detail, finding that it can describe generalized quantum
measurements beyond the projection postulate.
Generalized measurements
We generalize the picture abovePby starting with an arbitary initial ancilla state,
written in its eigenbasis as ρA = j pj |jihj|A , an arbitrary unitary V , not necessarily
Quantum Measurement Theory 15
the one specified P in eqn (1.28), and a final projective measurement of some ancilla
n
observable RA = r=1 r|rihr|A . As above, we now take a look at the conditional state
of the system given the measurement result r of RA . Tracing out the ancilla, we get
the non-normalized system state
X
ρ̃′S (r) = trA {|rihr|A V ρS ⊗ ρA V † } = pj hr|A V |jiA ρS hj|A V † |riA . (1.32)
j
Next, we introduce the operators

√
Kj (r) ≡ pj hr|A (V |jiA ) . (1.33)
Notice that the Kj (r) are still operators acting on the system Hilbert space. Using
them, we can express eqn (1.32) more compactly as
X
ρ̃′S (r) = Kj (r)ρS Kj† (r). (1.34)
j
This equation makes it more transparent how the system state actually gets trans-
formed by computing the operators Kj (r) via eqn (1.33). In terms of them, we can
also express the probability of obtaining measurement result r as
X n o
p(r) = trS {ρ̃′S (r)} = trS Kj† (r)Kj (r)ρS . (1.35)
j
This equation suggests introducing the following probability operators:

X †
M (r) ≡ Kj (r)Kj (r), (1.36)
j
which are always positive, M (r) ≥ 0, which follows from the fact that any operator of
the form A† A is positive. Furthermore, they satisfy the completeness relation
X
M (r) = IS . (1.37)
r
Moreover, the set of operators {M (r)} completely fixes the measurement statistics
and therefore plays an important role in quantum measurement theory, where they
are known as a positive operator-valued measure or, in short, a POVM.
Exercise 1.4 Show that any set of operators {M (r)}r satisfying the completeness relation
and M (r) ≥ 0 for all r gives rise to a set of well-defined probabilities p(r) = trS {M (r)ρS }
for any state ρS . Hint: By ‘well defined’ we mean probabilities that are positive and sum up
to one.
We now study two examples to illustrate how the projection postulate can be
generalized. The first
Pnexample describes projective measurement of the initially studied
observable XS = x=1 λ(x)ΠS (x), but it includes classical measurement errors. For
p we relabel j ≡ x in eqn (1.33) and model the measurement with operators

this purpose
Kx (r) = p(r|x)ΠS (x), giving rise to the probability operator
X
M (r) = p(r|x)ΠS (x). (1.38)
x
For {M (r)} to be a POVM,

P we demand that the p(r|x) are non-negative numbers
normalized according to r p(r|x) = 1. Using eqn (1.34), the non-normalized post-
measurement state of the system becomes
X
ρ̃′S (r) = p(r|x)ΠS (x)ρS ΠS (x). (1.39)
x
P
Its norm equals the probablity of obtaining result r, p(r) = x p(r|x)trS {ΠS (x)ρS },
which has a transparent interpretation: since trS {ΠS (x)ρS } equals the probability of
obtaining result x in an error-free projective measurement, p(r|x) is the conditional
probability of obtaining result r given that an error-free measurement had given result
x. Thus, owing to some errors in the classical data processing, erroneous detection
events of r 6= x can occur. The case of error-free measurements is recovered if p(r|x) =
δr,x , where eqn (1.39), after normalization, reduces to eqn (1.26).
The second example also describes a measurement of XS with errors, but this time
the errors
P cannot
p be explained classically. For this purpose, let us consider operators
K(r) = x p(r|x)ΠS (x) (without any label j), which gives rise to
X
M (r) = p(r|x)ΠS (x), (1.40)
x
which coincides with eqn (1.38). Thus, they give rise to the same measurement statis-
tics. However, the post-measurement state looks very different:
Xp Xp
ρ̃′S (r) = K(r)ρS K(r) = p(r|x)ΠS (x)ρS p(r|x′ )ΠS (x′ ). (1.41)
x x′
This teaches us an important lesson, namely that the probablity operators M (r) do not
uniquely fix how the state changes as a result of the measurement. The next exercise
elucidates the difference between eqns (1.39) and (1.41) further.
Exercise 1.5 Consider an initially pure state ρS = |ψihψ|S . Then, show that the post-
measurement state given by eqn (1.41) is always pure, i.e. [ρ′S (r)]2 = ρ′S (r), whereas this is
in general not the case for eqn (1.39). Thus, eqn (1.41) preserves the ‘quantum character’ of
the state, whereas eqn (1.39) inevitably introduces classical noise.
Summary
We summarize what we have found out in this chapter. By shifting the description of
the projective measurement from the system itself to an external ancilla, with which the
system interacts in a unitary way, we obtained a more general description of quantum
measurements expressed by system state changes of the form (1.34). It includes the
Operations, Interventions and Instruments 17
case of projective measurements, but as eqns (1.39) and (1.41) have shown also more
general transformations. The generalization of the average unconditional system state
after a projective measurement, eqn (1.27), follows from eqn (1.34) as
X X XX
ρ′S = p(r)ρ′S (r) = ρ̃′S (r) = Kj (r)ρS Kj† (r). (1.42)
r r r j
To simplify the notation, we introduce the multi-index α = (r, j) and write

X
ρ′S = Kα ρS Kα† . (1.43)
α
Owing to trace conservation, we know that the operators Kα must satisfy the com-
pleteness relation X
Kα† Kα = 1. (1.44)
α
Apart from this relation, it turns out that the operators Kα are indeed arbitrary. We
will learn more about them from an abstract perspective, which is not necessarily
related to quantum measurements, in the next section.
It is instructive to point out the similarities and differences between the system–
ancilla picture and the system–bath picture of Section 1.2. In both cases the system is
open because of its coupling to some external degrees of freedom. However, we did not
call the ancilla a ‘bath’ here because its interpretation is quite different in the present
context. Whereas the bath was introduced in Section 1.2 as an uncontrollable and
typically large object, the small and controllable ancilla of this section was introduced
to probe the state of a quantum system. The physical interpretation of this situation
is different, but—as the next sections will show in greater detail—the mathematical
description is essentially the same. Before proceeding, we conclude this section with a
final exercise to elucidate the difference between classical and quantum measurements.
Exercise 1.6 Construct the classical counterpart of the theory above. To this end, replace
the notion of the density matrix ρS by a vector p with elements p(x) denoting the probability
of finding the classical system in state x. Let p(r|x) be the conditional probability of obtaining
the measurement result r given that the system is in state x and define the matrix M (r) with
elements Mxx′ (r) = δx,x′ p(r|x). What do {M (r)} and a POVM have in common? What is
the
P quantum counterpart of the classical expression M (r)p? Relate the probability p(r) =
x p(r|x)p(x) to obtain outcome r to the expression M (r)p. Show that the (normalized) post-
measurement state of the system given result r obeys Bayes’ rule p′ (x|r) = p(r|x)p(x)/p(r).
Finally, verify that the average post-measurement state does not change: p′ = p. Thus, Bayes’
rule describes the most general non-disturbing classical measurement.
1.5 Operations, Interventions and Instruments

The goal of this section is to review the formalism of quantum measurement theory
from an abstract perspective and to introduce some terminology and powerful math-
ematical results. From now on, we call the map (1.43) in general a (control) operation
or intervention. These notions emphasize that an experimentalist has a large amount
of freedom to manipulate a quantum system and the intention might not always be
to ‘measure’ the system in a literal sense. We further abbreviate eqn (1.43) as
X
ρ′S = CρS ≡ Kα ρS Kα† . (1.45)
α
The map C, which takes an operator and maps it to another operator, is also known as
a superoperator to distinguish it from the notion of a ‘usual’ operator, which ‘only’
maps vectors onto vectors. Mathematically speaking, this distinction is superfluous:
the space of matrices itself forms a vector space and, therefore, one can represent
any superoperator C also by a (large) matrix. For physics applications it is, however,
advantageous to use a terminology and notation that distinguishes whether an operator
acts on states |ψi or density matrices ρ. From now on, we frequently use superoperators
as an efficient bookkeeping tool. By convention, superoperators are always written
with a calligraphic letter such as C and they act on all operators to their right, i.e.
C2 C1 ρ = C2 [C1 (ρ)]. Note that the action of different superoperators does not commute
in general, i.e. C2 C1 ρ 6= C1 C2 ρ. Readers who are exposed to this for the first time are
advised to consult Appendix B before proceeding.
We also introduce the superoperator C(r) corresponding to eqn (1.34):
X
ρ̃′S (r) = C(r)ρS ≡ Kαr ρS Kα† r , (1.46)
αr
which is obtained from eqn (1.45) by using only a subset of the operators Kα . The
subsets of Pindices {αr } are obtained from the total set {α} by a disjoint decomposition
such that r C(r) = C. In the following we P call a set of maps {C(r)} that decomposes
an operation C of the form (1.45) such that r C(r) = C an instrument. Notice that,
given only eqn (1.45), there are many instruments that decompose C.
Now, our claim is that eqn (1.45) describes the most general control operation
or intervention we can implement in a laboratory on a quantum system at a single
time. To verify this, let us find the minimial conditions we would like to have satisfied
by an instrument {C(r)} such that it describes a physically allowed control operation
in a laboratory. Clearly, if the state of the system was previously described by a
valid density matrix ρS , the minimal requirement is that also the final set of states
{ρ̃′S (r) = C(r)ρS } can be interpreted in a legitimate way as the state of a physical
system. Recall that any density matrix ρ is positive, has unit trace and if ρ1 and ρ2
are two valid density matrices, then their convex combination λρ1 + (1 − λ)ρ2 for any
λ ∈ [0, 1] is another valid density matrix. Therefore, we postulate that any physically
legitimate instrument {C(r)}r should at least satisfy the following:
(i) C(r) is positive: if ρS ≥ 0, then also C(r)ρS ≥ 0.
(ii) C(r) is trace non-increasing: if tr{ρS } = 1, then p(r) = tr{C(r)ρS } is the prob-
ability of applying the map C(r) (in view of a quantum measurement, it is the
probability of obtaining result r). If p(r) = 1, C(r) = C is called trace-preserving.
P (i)
(iii) C(r) is convex linear:
P if i λi ρS is a statistical mixture of quantum states with
λi ≥ 0 satisfying i λi = 1, then
X (i)
X (i)
C(r) λ i ρS = λi C(r)ρS . (1.47)
i i
Now, to make things even a little more complicated, it turns out that desideratum
(i) is not enough. Remember that C(r) could describe, for instance, a measurement of
a system S coupled to a bath B. Prior to the measurement the system–bath state ρSB
can be arbitrary and we would like to ensure that the post-measurement state also
describes a legitimate quantum state. This means that we actually want the global
state ρ̃′SB (r) = [C(r) ⊗ IB ]ρSB ≥ 0 to be positive, where IB denotes the identity
operation acting on the bath defined via IB ρB ≡ ρB for any state ρB . This is not
necessarily guaranteed by the notion of a positive map. Therefore, we replace (i) by
the stronger condition:
(i′ ) C(r) is completely positive: consider the composite system HS ⊗HB with HB an
arbitrary Hilbert space. If ρSB ≥ 0 is an arbitrary positive state of the composite
system, then [C(r) ⊗ IB ]ρSB ≥ 0 is also positive.
Exercise 1.7 An example for a positive but not completely positive map is the transpose
operation T . Let ρ = ρ00 |0ih0| + ρ01 |0ih1| + ρ10 |1ih0| + ρ11 |1ih1| be an arbitrary density
matrix of a qubit, then the transpose operation with respect to the basis {|0i, |1i} is defined
via T ρ ≡ ρ00 |0ih0| + ρ10 |0ih1| + ρ01 |1ih0| + ρ11 |1ih1|. Show that this map is positive, but not
completely positive. Hint: As a counterexample consider the transpose operation√acting on
the first qubit of the pure and maximally entangled bipartite state (|00i + |11i)/ 2.
We remark that the distinction between positivity and complete positivity is a

quantum phenomenon: classically, all postive maps are also completely positive. Fur-
thermore, the next exercise shows that any convex linear map acting on density ma-
trices can be extended to a linear map acting on arbitrary matrices. We therefore use
the notions convex linearity and linearity in the following synonymously.
Exercise 1.8 Let C be a convex linear map acting on density matrices ρ of a Hilbert space
with dimension dim H = d. Show that every complex d × d matrix A can be written P as a
linear combination of density matrices ρi with complex coefficients ci ∈ C, i.e. A = i ci ρi .
Hint: To do so, you can use the two fundamental results that, first, every A can be written
as A = A1 + iA2 with A1 = A†1 and A2 = A†2 being Hermitian and, second, every Hermitian
A can be decomposed as A = A+ − A− with A± ≥ 0 being positive.
˜ ≡ P ci Cρi and in the following, tacitly
The extension C˜ of C is then defined via CA i
assuming this extension, we identify C ≡ C.
˜
To conclude, physically legitimate operations are mathematically described by an

instrument that is a collection of completely positive, trace non-increasing and linear
maps, which we call CP maps. All CP maps of an instrument have to add up to a
completely positive and trace-preserving map, which we call a CPTP map. There
are two important theorems to characterize CP(TP) maps mathematically.
Operator–sum representation. A map C(r) satisfies desiderata (i′ ), (ii) and (iii)
if and only if it can be written as in eqn (1.46) for some operators Kαr satisfying
P P
Kα† r Kαr ≤ IS . If αr Kα† r Kαr = IS , then C(r) = C is trace-preserving. Equa-
αr
tions (1.45) or (1.46) are called the operator–sum representation of a CP(TP) map.
P
It is not too hard to show that a map of the form (1.46) with αr Kα† r Kαr ≤ IS
satisfies points (i′ ), (ii) and (iii). The other direction is harder to show and not done
here. Instead, we only note that the operator–sum representation (1.46) is not unique,
as explicitly shown by the next exercise.
Pd † P
Exercise 1.9 Consider the map Cρ = α=1 Kα ρK and define Kα ≡ α Uαβ K̃β for an
Pαd †
arbitrary d × d unitary matrix U . Show that Cρ = α=1 K̃α ρK̃α .
It turns out that there is another important representation theorem for an instru-
ment, which brings us back to our ancilla construction used in the previous section.
Unitary dilation theorem. A set of maps {C(r)} acting on a d-dimensional system
Hilbert space HS forms an instrument if and only if there exists a d2 -dimensional
‘ancilla’ Hilbert space HA , a unitary USA acting on HS ⊗ HA , a pure ancilla state
|ϕiA ∈ HS and a set of projectors {ΠA (r)} acting on the ancilla space such that
†
C(r)ρS = trA {ΠA (r)USA (ρS ⊗ |ϕihϕ|A )USA } (1.48)
for all r. In particular, a map C is CPTP if and only if it can be written as
†
CρS = trA {USA (ρS ⊗ |ϕihϕ|A )USA }. (1.49)
We are not going to prove the unitary dilation theorem here, but only discuss its
consequences. For instance, the next exercise shows that the number of operators Kα
appearing in the operator–sum representation is at most d2 with d = dim HS .
Exercise 1.10 Use the unitary dilation theorem to derive eqns (1.45) and (1.46) by giving
explicit expressions for the operators Kαr and Kα in terms of USA , |ϕiA and ΠA (r). Confirm
that the maximum number of operators Kα is d2 . Hint: Remember eqn (1.33).
We remark that it is also possible to use the operator–sum representation to derive

the unitary dilation theorem. Hence, both are equivalent statements. The physical
significance of the unitary dilation theorem is that it teaches us that every allowed
state transformation in quantum mechanics satisfying desiderata (i′ ), (ii) and (iii) can
be constructed by more primitive transformations, namely unitary evolution and pro-
jective measurements. We therefore do not need to add any axiom to the standard
textbook framework of quantum mechanics in order to describe generalized measure-
ments or other general state transformations.
The unitary dilation theorem also fits well into our system–bath paradigm from
Section 1.2 if we regard the ancilla as the bath coupled to the system. From that
perspective, it tells us that every state change of a quantum system can be explained
through the interaction with an external environment, which we might call an ancilla
or a bath depending on the context. Care is, however, required here because the unitary
dilation theorem is an abstract statement and does not tell us for a given transformation
of the system what is the actual environment causing this transformation. This is
related to the non-uniqueness of the operator–sum representation: there are infinitely
many Hilbert spaces HA , unitary matrices USA , states |ϕiA (possibly also mixed states)
and projectors ΠA (r) satisfying eqns (1.48) and (1.49).
Before concluding this section, we point out a subtle but important observation.
What the unitary dilation theorem also teaches us is that satisfying desiderata (i′ ),
(ii) and (iii) is equivalent to representing the interaction of the system with some en-
vironment that is initially decorrelated from the system, i.e. of the form ρS ⊗ |ϕihϕ|A .
However, the evolution of an open quantum system (1.11) could also result from a
system initially correlated with the environment. In this case, the operator–sum repre-
sentation and the unitary dilation theorem break down. The reason is that desideratum
(iii) (linearity) is then no longer satisfied. This is exemplified in the next exercise and
a way out of this ‘dilemma’ is presented in Section 1.7.
Exercise 1.11 Consider the interaction of two qubits. The first qubit is called the system
and the second√ the ancilla. We further introduce the maximally entangled states |±i =
(|00i+|11i)/ 2, where by convention we set |iji ≡ |iiS ⊗|jiA for i, j ∈ {0, 1}. Let the system–
ancilla interaction be modelled by the unitary USA = |+ih+| + |−ih10| + |10ih−| + |01ih01|
(check that this is a unitary matrix). The evolution of the composite system is therefore
†
modelled via ρ′SA = USA ρSA USA , which is clearly linear with respect to the joint input state
ρSA and CPTP.
Next, consider the reduced dynamics of S assuming that the initial system–ancilla state
ρSA = |+ih+| is entangled. Verify the following results: (1) the initial reduced system state is
ρS = (|0ih0| + |1ih1|)/2; (2) the final system state is identical to the initial system state, ρ′S =
†
trA {USA |+ih+|USA } = ρS ; (3) if we first perform a measurement of the initial system in its
eigenbasis, the initial system state does not change on average: |0ih0|ρS |0ih0|+|1ih1|ρS |1ih1| =
ρS ; but (4) the final system state after such an initial measurement
† 1 3
trA {USA (|0ih0|ρSA |0ih0| + |1ih1|ρSA |1ih1|)USA }= |0ih0| + |1ih1| (1.50)
4 4
is different from ρS . Hence, we have a ‘paradox’ because the same reduced initial system state
gives rise to two different final states. Thus, we cannot associate any map C, which only acts
on the system part, with this input–output relation.
How can this be resolved? Clearly, from a global point of view there is no paradox: the
two initial states ρSA = |+ih+| and |0ih0|S ρSA |0ih0|S + |1ih1|S ρSA |1ih1|S are different (albeit
they give rise to the same reduced system state), and hence there can be two different output
states. With respect to our construction above, the key insight is to realize that it is no longer
meaningful to speak about different initial system states if the system is initially entangled with
its environment (here, the ancilla). In particular, we assumed that it is possible to mix (or
convex combine) different system states without influencing the dynamics, i.e. the definition
of the map. This assumption is incompatible with having an initial system state entangled
with its environment.
We summarize the content of the last two sections, which play an important role
in the following. First, quantum systems can undergo more general state transfor-
mations than unitary evolutions and projective measurements. However, even general
state transformations can be constructed using only unitary evolutions and projective
measurements on a bigger system–ancilla (or system–bath) space. If the desiderata
(i′ ), (ii) and (iii) are satisfied, where point (iii) assumes that the preparation of the
initial system state can be disentangled from the effect of the state transformation,
then we can use either the operator–sum representation or the unitary dilation the-
orem. If there is no post selection (i.e. conditioning on a measurement result r), the
state transformation is described by a CPTP map represented by either eqn (1.45) or
eqn (1.49). If there is conditioning, the state transformation is described by a CP map
represented by either eqn (1.46) or eqn (1.48). On average, all CP maps have to add
up to a CPTP map. Vice versa, every CPTP map can be decomposed into a set of
CP maps. Such a set is called an instrument.
Finally and for completeness, we mention one extension of the framework intro-
duced here. So far, we have assumed that the dimension d of the system space does
not change during the control operation, but for some applications it makes sense to
relax this requirement. These applications play a minor role in quantum stochastic
thermodynamics and the mathematical modifications we have to add only amount
to correct bookkeeping of the different input and output Hilbert spaces in the nota-
tion. Nevertheless, the final exercise shows a couple of neat examples of such control
operations, which are useful to keep in mind.
Exercise 1.12 Verify that the following maps are CPTP by finding an operator–sum repre-
sentation for them. Example 1: The ‘trace map’ is defined for any input state ρ by Cρ ≡ tr{ρ}.
This map is also often associated with ‘discarding a system’ as it destroys all the information
contained in a system and replaces it by a ‘trivial’ system living in the Hilbert space H = C
with the only possible density matrix ρC = 1. Example 2: Consider a bipartite system with
Hilbert space H1 ⊗H2 . Show that the ‘partial trace map’ is CPTP: Cρ12 ≡ tr2 {ρ12 } = ρ1 . Ex-
ample 3: In contrast to Example 1, one can also ‘create’ a system using a CPTP map, which
takes input states from the Hilbert space H = C, by defining Cρ c ≡ ρ for any c ∈ C and some
fixed density matrix ρ. Note that, according to this example, ‘states’ in quantum mechanics
are CPTP maps. Example 4: An extension of Example 3 and in some sense the opposite of
Example 2 is the map which ‘adds’ a fixed state ρ2 to an input state ρ1 : Cρ2 ρ1 ≡ ρ1 ⊗ ρ2 .
1.6 Classical Stochastic Processes

In the previous sections we have learned about some fundamental aspects of open
quantum systems and quantum measurement theory. Before we put these two ingre-
dients together to define a quantum stochastic process, it is worthwhile recapitulating
the definition of a classical stochastic process. Classical stochastic processes have found
widespread application in the natural and social sciences. They are commonly used to
describe a process in time that is characterized by a random variable whose observation
can be safely assumed not to change the process. It is obvious that this assumption
is no longer satisfied for quantum systems. Even classically, this assumption can be
violated. This leads to the much richer framework of classical causal models—a topic
which we briefly touch on at the end of this section.
Let R denote some random variable characterizing the state of a system evolving
in time. Simple examples for R with relevance for (quantum) stochastic thermody-
namics include the position of a colloidal particle suspended in water (which yields to
Brownian motion), the number of electrons in a nanostructure such as a quantum dot
or the number of photons in a cavity and conformational states of a macromolecule
(e.g. folded or unfolded), among many others. In order to infer something about the
Classical Stochastic Processes 23
time evolution of the system, the experimenter measures the value of R at an ar-
bitrary set of times {tℓ }nℓ=0 , where here and in the following we assume the order
0 = t0 < t1 < · · · < tn . The results of the measurement at time tℓ are denoted by rℓ . By
repeating the experiment many times, the experimenter can gather enough statistics to
construct (approximately) the joint probability distribution p(rn , tn ; . . . ; r1 , t1 ; r0 , t0 )
to observe r0 at time t0 , r1 at time t1 , and so on and so forth up to the last measure-
ment giving result rn at time tn . Since the subscript at the measurement result rℓ is in
one-to-one correspondence with the time tℓ of the measurement, we write for brevity
p(rn , . . . , r1 , r0 ) ≡ p(rn , tn ; . . . ; r1 , t1 ; r0 , t0 ). (1.51)
To be a valid probability distribution, p(rn , . . . , r1 , r0 ) must satisfy

X XX
p(rn , . . . , r1 , r0 ) ≥ 0, ··· p(rn , . . . , r1 , r0 ) = 1. (1.52)
rn r1 r0
To save further space in the notation, we write the sequence of measurement results
in boldface as rn P ≡ (rn , . . . , r1 , r0 ). In this notation the content of eqn (1.52) reduces
to p(rn ) ≥ 0 and rn p(rn ) = 1. Note that the sequence rn has n + 1 entries (and not
n) since our first measurement happens by convention at time t0 (and not at t1 ).
The conditions (1.52) of positivity and normalization have to be satisfied by any
probability distribution. Therefore, p(rn ) does not yet describe a stochastic process
that has one important additional structure. To uncover this additional structure, we
need to look at the joint probability distribution p(rn , . . . , rℓ+1 , rℓ−1 , . . . , r0 ) to obtain
the results rn , . . . , rℓ+1 , rℓ−1 , . . . , r0 by measuring the system at times tn , . . . , tℓ+1 ,
tℓ−1 , . . . , t0 , i.e. at all times except for time tℓ , where we perform no measurement.
To emphasize this fact, we denote this probability by p(rn , . . . , rℓ , . . . , r0 ). Then, we
define a classical stochastic process by the requirement that
X
p(rn , . . . ,
rℓ , . . . , r0 ) = p(rn , . . . , rℓ , . . . , r0 ) (1.53)
rℓ
for all ℓ ∈ {0, 1, . . . , n}. This is known as the Kolmogorov consistency condition
and it tells us that not measuring the observable R at some point in time is equivalent
to measuring it followed by marginalizing about the measurement result. Therefore,
the joint probabilities of a classical stochastic process form a hierarchy, where the
(n + 1)-time joint probability p(rn , . . . , r1 , r0 ) contains all the information about the
k-time probabilities (with k < n+1) obtained from measuring the system at any subset
of the times {tn , . . . , t1 , t0 }. Furthermore, an important mathematical result known as
the Daniell–Kolmogorov extension theorem says that we can also go the reverse way:
whenever we have a hierarchy of n-time joint probabilities satisfying the Kolmogorov
consistency condition, then they can be constructed as marginals of N -time joint
probabilities with N > n, which also satisfy the consistency condition. The important
point here is that this holds even in the limit of n → ∞. The Daniell–Kolmogorov
extension theorem therefore provides a bridge between experimental reality (where any
measurement statistics are always finite and described by n < ∞) and its theoretical
description (which often uses continuous-time dynamics in the form of, for example,
stochastic differential equations, which assume n → ∞).
It is clear that the Kolmogorov consistency condition is in general not satisfied for
quantum systems. It is, however, noteworthy that even for classical systems eqn (1.53)
can be broken. This can happen simply for the reason that the measurement of a
classical system can be disturbing as well. Another reason is an experimenter choosing
to actively break the consistency condition. Examples of the latter kind include the use
of feedback control, where an external agent manipulates the dynamics of a system
based on the so far available information (say, based on rk up to some time tk ).
The future probabilities of observing rn (n > k) are then in general different from
the situation where the external agent had not observed the system previously (and,
consequently, had not applied any feedback control). Another example is a clinical trial
where the health of patients is monitored at regular intervals and where the process
can be actively changed by giving drugs. The health of the patients in the future then
depends on the question of whether their health was monitored previously or not (and,
consequently, whether they received drugs or not).
These examples show that the theory of classical stochastic processes breaks down if
an external agent intervenes in the process. The mathematical language appropriate for
such situations is the theory of classical causal models. Without going into its details,
we point out that causal models can distinguish between causation and correlation,
whereas classical stochastic processes cannot. Indeed, classical stochastic processes
allow correlations to be quantified between two random variables, say X and Y , but
the consistency condition forbids us to infer whether X is a cause of Y , Y is a cause
of X, or whether there is no causal relation between X and Y and their correlation
is the result of another common cause Z. In order to infer causal relationships, it is
important that we can change the process and thus violate the consistency condition,
for instance by looking at the behaviour of Y after forcing X to take on a certain
value. The following example makes this qualitative reasoning quantitative.
Exercise 1.13 We consider three binary random variables S, B and C with values s, b
and c. On a given day S describes whether the sun is shining (s = 1) or not (s = 0), B
describes whether the number of sunburns is high (b = 1) or low (b = 0) and C describes
whether the number of ice cream sales is high (c = 1) or low (c = 0). We assume the
notion of ‘high’/‘low’ to be chosen according to some reasonable threshold. We further set
the conditional probabilities to be p(b = 1|s = 1) = p(c = 1|s = 1) = p(b = 0|s = 0) = p(c =
0|s = 0) = λ, with λ ∈ [1/2, 1]; for example, λ = 1 implies that the number of sunburns is
always high if the sun is shining. By conservation of probability, p(b = 0|s = 1) = p(c = 0|s =
1) = p(b = 1|s = 0) = p(c = 1|s = 0) = 1 − λ. Furthermore, we assume the probability for
a sunny day to be p(s = 1) = 1/2, which implies that the sun is not shining with the same
probability p(s = 0) = 1/2. Now, first confirm that B and C are correlated unless λ = 1/2.
This can be done in several ways, for instance by computing the mutual information, which
we introduce in Appendix A in detail, between B and C:
X p(b, c)
IB:C ≡ p(b, c) ln = ln 2 − SSh (2λ − 2λ2 ). (1.54)
p(b)p(c)
b,c
Here, SSh (p) ≡ −p ln

Pp−(1−p) ln(1−p)
P denotes the binary Shannon entropy. Verify eqn (1.54)
by using p(b, c) = s p(b, c, s) = s p(b|s)p(c|s)p(s). Show that IB:C = 0 (no correlations)
implies λ = 1/2 and IB:C = ln 2 (maximal correlations) implies λ = 1. Thus, while B and
C are in general correlated, they are not causally related as demonstrated below. Does this
sound plausible to you?
Quantum Stochastic Processes 25
Next, we turn to the correlations between S and B. Confirm that IB:S = ln 2 − SSh (λ)
and find the values of λ for which S and B are maximally correlated/uncorrelated.
Finally, we strongly believe that S is the cause of B (and also of C), i.e. the shining sun
triggers sunburn (and ice cream sales). But how can we make this intuition rigorous?
Assume that we have an external mechanism that can change whether the sun is shining
or not (which seems unrealistic at first sight, but we come back to it below). We therefore
introduce an additional intervention variable IS (not to be mixed up with the mutual in-
formation), which labels the following three actions i: do nothing, i.e. leave the sun as it
is (i = idle), make the sun shining (i = 1) or block sun shine (i = 0). Now, consider the
conditional probability p(b|s, i) for sunburns given sunshine s and intervention i. For i = idle
we set p(b|s, idle) = p(b|s) as defined above. Furthermore, we assume p(b|s, i = 1) = λ and
p(b|s, i = 0) = 1 − λ independent of s because i ∈ {0, 1} overwrites the natural value of s to
be identical to i. Next, confirm that the mutual information between B and IS is
X p(b, i)
IB:IS = p(b, i) ln = SSh [p(b)] − [1 − p(i = idle)]SSh (λ) − p(i = idle) ln 2, (1.55)
p(b)p(i)
b,i
where the marginal probability p(i) that we perform a certain intervention is assumed to be
controllable in an experiment. We now define that S is a cause of B if there are correlations
between IS and B. Confirm that there are no correlations between B and IS , i.e. S is not
the cause of B, if one of the following two cases happen: either p(i = idle) = 1, which
corresponds to the case that we do not perform any intervention and, hence, cannot test for
causality, or λ = 1/2, which implies that there are no correlations
P between B and S in the first
place. Furthermore, confirm that in general p(b, s) 6= i p(b, s, i), where p(b, s) = p(b|s)p(s)
is the joint probability from the beginning obtained without interventions and p(b, s, i) =
p(b|s, i)p(s)p(i) is the joint probability with interventions. Thus, the Kolmogorov consistency
condition (1.53) is broken in general. Show that the consistency condition is obeyed if and
only if p(i = idle) = 1 or p(i = 0) = p(i = 1). Can you see why? Of course, we could also
replace B by C above and find that sunshine causes a high number of ice cream sales.
Finally, let us return to our assumption that we can change the sunlight by an external
intervention. Indeed, such a mechanism is not easy to construct for a human. But to distin-
guish causation and correlation, it is not necessary that humans perform the intervention: it
could be also done by nature, for instance, due to a solar eclipse. Important is only that we
can fix the intervention variable independent of the other variables in the model.
1.7 Quantum Stochastic Processes

In this section we introduce the notion of a quantum stochastic process describing the
response of an open quantum system to the most general interventions we can perform
on it. This generality comes at the price of a high level of abstractness, even though
it might help to keep in mind that much of what follows below ‘merely’ presents
convenient terminology and notation. We proceed by first applying the notion of a
classical stochastic process directly to quantum systems. Afterwards, we show that
the idea can be fruitfully generalized to define a quantum stochastic process. Then, we
show how to reconstruct a quantum stochastic process experimentally and we finish
by exploring further mathematical consequences of this formalism.
Quantum statistics of a projectively measured system

Even though quantum measurements are in general disturbing, the joint probability
p(rn ) in eqn (1.51) remains a well-defined object in quantum mechanics. Experimen-
tally, it describes the probability of obtaining the results rn = (rn , . . . , r1 , r0 ) after
Another random document with
no related content on Scribd:
the way, his ¹wisdom heart he walks, and says
faileth him, and he saith to all, What elaborate
to every one that he is a folly this is!
fool.
¹ Hebrew his
heart.
(3.) And moreover in the way (which word ‘way’ is so

constantly used in an ethical sense――Psalms cxix. 1――that we
cannot overlook it here) like that which is the wise fool’s (the
Masorets notice the article here, and pronounce it superfluous, but it
is not so; for the meaning is, that it is like the perversely wise fool’s
way generically, in this) that as he walks, his heart (the third time
‘heart’ has occurred in this passage, raising the word into great
emphasis and importance), fails (the Authorized Version considers
this to mean a failure in wisdom, but it is rather a failure of
confidence, which is the ethical meaning of the term ‘heart’) and
says (the nearest nominative is ‫לב‬, heart, and so the LXX.
understood, for they render ἃ λογιεῖται, κ.τ.λ. ‘that which he thinks of’
is folly; this makes good sense) to all, perverse folly it is (emphatic,
hence the meaning is, ‘he is out of heart altogether,’ or ‘his heart
misgives him;’ and it says, ‘what perverse folly it all really is.’
Conscience convicts those clever wicked plans, and they who devise
them know that they are only elaborate mistakes).
4 If the spirit of the If the spirit of the

ruler rise up against ruling one should go
thee, leave not thy forth against thee, thy
place; for yielding station do not quit,
pacifieth great offences. because a remedy may
cure wicked errors which
are great.
(4.) If a spirit of the ruling one (not, as usually rendered, the

ruler, which does not exactly convey the idea) goes up against thee
(the LXX. show that they so understood it by rendering πνεῦμα τοῦ
ἐξουσιάζοντος) thy place do not yield (the sense of the passage is,
‘If there be too strong a spirit against you, if you are sailing, as it
were, in the teeth of the wind, do not yield when you have good
grounds for remaining:’ this makes excellent sense, is cognate to the
accompanying passages, and follows the LXX.) for a healing (‫מרפא‬,
occurs Proverbs xiv. 30 and xv. 4 only, the LXX. read ἴαμα, ‘a
remedy’) pacifies mistakes (with the usual idea of culpability
attaching to this word) great ones (the idea is ‘do not yield to mere
adverse circumstances when even culpable mistakes admit of a
remedy.’)
5 There is an evil There exists an evil

which I have seen under which I have observed in
the sun, as an error this work-day world, like
which proceedeth ¹from an error which goes forth
the ruler: from before the face of
the Powerful,
¹ Hebrew
from
before.
(5.) There exists an evil (notice abstract with its shade of

meaning, which) I have seen under the sun, like that which is
erroneous (‫שגגה‬, see chapter v. 5 (6), ‘an inadvertence’), which
goes out (the verb has the contract-relative joined with it; the exact
idea is that it is like an inadvertence, such as might go out on the
part of the ruler’s command, the great Ruler being in the mind of the
writer, but the proposition is general) from the face of the caused
to have power (a ‘providential mistake,’ then).
6 Folly is set ¹in great viz., the setting of false

dignity, and the rich sit in wisdom in high places,
low place. and the rich sit in low
estate.
¹ Hebrew in
great
heights.
(6.) Set (that is, the ruler does this, but, as usual, this is not
expressed when the proposition is intended to have a general
bearing) the perverse fool (generic――‘perverse folly’ then will be a
good rendering) in high places many a one, and the rich (but the
hiphil form is worthy of remark, ‘persons that make rich’) in a low
place (‫ ֵש ֶפל‬occurs so punctuated at Psalms cxxxvi. 23 only, rendered
‘low estate’) sit.
7 I have seen I have seen serfs on
servants upon horses, horseback, and princes
and princes walking as walking like serfs afoot.
servants upon the earth.
(7.) I have observed servants (slaves, that is, who ought to

serve) upon horseback, and princes walking as servants (‘ought
to do’ is no doubt involved in this expression――‘servants’ repeated
being emphatic) upon the earth (i.e. afoot).
8 He that diggeth a One digs a pit, into that

pit shall fall into it; and he falls: or breaks a
whoso breaketh an hedge, gets bitten by a
hedge, a serpent shall serpent.
bite him.
(8.) Dig (not necessarily either a participle or an imperative) a

pitfall (‫ גומץ‬occurs here only, and is said to be a late word; it occurs
in Arabic and Syriac. That a ‘pitfall’ is meant is evident from the
context), in it (emphatic) he falls (a sinister intent in digging this pit
is not necessarily implied, but the context shows that such is
primarily aimed at: this is the more evident when we recollect that
‫ ָח ַפר‬is to ‘dig,’ and ♦‫‘ ָח ֵפר‬to bring to confusion’); and break a wall (i.e.
an enclosure, see Job xix. 8 for the precise meaning of the root,
hence also Numbers xxii. 24), bites him a serpent (as we say, ‘gets
bitten by a serpent,’ which would naturally lurk in loose stone walls).
♦ “‫ ”ָח ֵפד‬replaced with “‫”ָח ֵפר‬
9 Whoso removeth Moves stones, and finds

stones shall be hurt them in his way: chops
therewith; and he that wood, must be careful
cleaveth wood shall be with it.
endangered thereby.
(9.) Cause to move (♦hiphil participle of ‫נסע‬, ‘bring up’――see

Exodus xv. 22) stones, be troubled (see Genesis xlv. 5) with them
(emphatic); cleaving (poel participle, occurs Psalms cxli. 7; Isaiah
lxiii. 12 only) wood (plural ‘logs of wood’) be endangered (this is
called a future niphal by the Masorets, who so point, but the real
meaning of ‫ סכן‬is evidently to ‘take care,’ so that the reading of the
LXX. by ♠κινδυνεύσει, ‘he shall be endangered,’ is ad sensum――it is
literally ‘he shall take care,’) with them (emphatic, all these are
instances of either unexpected or unintentional results).
♦ “hiphal” replaced with “hiphil” for consistency
♠ “κινδευνεύσει” replaced with “κινδυνεύσει”
10 If the iron be If the axe be blunt, then

blunt, and he do not its edge had best be set:
whet the edge, then and then if one of the
must he put to more strong hits prevail, the
strength: but wisdom is skilful hit was it.
profitable to direct.
(10.) If blunt (‫――קהה‬occurs Jeremiah xxxi. 29, 30, and Ezekiel

xviii. 2――in the sense of ‘teeth set on edge:’ there the Masorets
point as Kal, here as piel) the iron, and he (emphatic, but there is
no nominative expressed to which this can refer) not the faces
(usually considered to refer to the edges of the axe-head) sharpen
(occurs Ezekiel xxi. 21 (26), as pilpel of ‫קלל‬, which has the meaning
of ‘lightness,’ ‘swiftness;’ the word occurs as an adjective, Numbers
xxi. 5, in the sense of ‘light,’――our soul loatheth this light food) and
strong ones will prevail (singular, If ‘strong ones’ be the
nominative, this is an instance of a distributive plural――one or more
of these will; the future piel has the meaning ‘strengthen,’ the Kal ‘to
prevail,’ but we can only consider this as a Masoretic conjecture)
and profit causing success (but the LXX. render by περισσεία,
‘advantage’――see below; but ‫ כשר‬occurs only Esther viii. 5, and
chapter xi. 6; see however ‫כשרון‬, which occurs chapter ii. 21, iv. 4, v.
10 (11), which we have seen occasion to render ‘success;’ hence the
meaning, ‘the made successful is’) wisdom (not generic, i.e. a single
instance of it). The general scope is quite clear; it is the superiority of
wisdom to brute force, and so all commentators and versions
understand it; but the exact rendering is very difficult;――all the
versions are perplexed and discordant, and the copies of the LXX.
have an important textual variation. We will give these at length,
beginning with the LXX. as the most ancient. This reads――Ἐὰν
ἐκπέσῃ τὸ σιδήριον καὶ αὐτὸς πρόσωπον ἐτάραξεν καὶ δυνάμεις δυναμώσει
καὶ περισσεία τοῦ ἀνδρείου (which B. reads τῷ ἀνδρὶ οὐ, and E. X. τοῦ
ἀνδρὸς) σοφία――‘If the axe-head should fall off, then the man
troubles his countenance, and he must put forth more strength; and
wisdom is the advantage of an energetic man.’ The Syriac version,
――‘If the axe be blunt, and it troubles the face and increases the
slain; and the advantage of the diligent is wisdom.’ The Vulgate
reads――‘Si retusam fuerit ferrum et hoc non ut prius sed
hebetatum fuerit, multo labore ♦ exacuetur et post industriam
♠sequetur sapientia’――‘If the iron should be blunt, and this not as
before, but should have lost its edge, it is sharpened with much
labour; and after industry will follow wisdom.’ Jerome renders the
former part in conformity with the Vulgate; but after ‘non ut prius,’
which he also has, runs on with――‘sed conturbatum fuerit,
virtutibus corroborabitur, et reliquum fortitudinis sapientia est
...’――‘but is troubled; it shall be strengthened by virtues, and the
remainder of strength is wisdom.’ It will be seen then that we have
reason to suspect a corruption of the text; and we think that the
suspicious ‘non ut prius’ of the Vulgate and Jerome shows what this
corruption was. We notice also that neither the LXX. nor the Syriac
take any notice of the negative. Guided by the clue thus given, we
will venture on the following conjectural emendation of the text. We
imagine that it was originally written thus, ‫והוא להפנים קלקל‬, the ‫ ה‬being
written full――like ♣ ‫ שתקיף‬in chapter vi. 10, compare also chapter
viii. 1, Nehemiah ix. 19――and having the meaning, ‘to the faces’ or
‘edges.’ Such an insertion of ‫ ה‬being unusual, would cause suspicion
to rest on the passage, and the transition to ‫ לא פנים‬would be easy.
This, however, was but one out of many possible conjectures, and
the Vulgate has preserved another, namely, that the reading was ‫לפני‬,
‘as before,’ and, as was common with the ancient versions, inserts
both the reading and its variant into the text. This conjectural change
in the text will make all quite clear; the passage will then read
thus――‘If the iron be blunt, and so it is as to its edges whetted, and
so too blows prevail, and so too an advantage is the success [due to
an instance] of wisdom,’ i.e. in this case a skilful hit. That is, if the
axe be blunt, grinding, force, and skill together, will produce the
required result. No doubt this can only be put forth as mere
conjecture, but, in the absence of any satisfactory interpretation, may
be admitted; for, in fact, arbitrary senses given to words, and the
insertions of explanatory glosses not immediately deducible from the
original, do amount to alterations of the text. None of the other
ancient Greek versions have been preserved in this place, except a
reading of Symmachus, which is very curious, showing still more
forcibly how early the difficulty must have arisen, since it is at best a
reading ad sensum only, προέχει δὲ ὁ γοργευσάμενος εἰς σοφίαν, ‘and
the nimble advances into wisdom.’
♦ “exacueter” replaced with “exacuetur”
♠ “sequeter” replaced with “sequetur”
♣ “‫ ”שהתקיף‬replaced with “‫”שתקיף‬
11 Surely the serpent If bites the snake

will bite without before the charm is
enchantment; and ¹a sung, then what is the
babbler is no better. profit of the skilful
tongue?
¹ Hebrew the
master of
the tongue.
(11.) If bites the serpent (with the article, and therefore

generic――serpents generally) without (‫ ;בלוא‬we may well suppose
that the full form is used not without meaning; it occurs Isaiah lv. 1, 2,
in the sense of ‘the absence of,’ which well suits the context here,)
whispering (occurs Isaiah ii. 3, 20, and xxvi. 6; Jeremiah viii. 17,
etc.), and there is nothing of profit to the master of the tongue
(with article, hence generic. The rendering of the Authorized Version
is derived from the Vulgate. The alliteration shows that the aphorism
is equivocal, it is the converse of the former: skill will help force, but
after the mischief is done skill is of no use. There is also here an
ironical depreciation of serpent-charming).
12 The words of a Each word of a wise

wise man’s mouth are man’s mouth is grace,
¹gracious; but the lips of but the lips of a fool will
a fool will swallow up swallow him apace.
himself.
¹ Hebrew
grace.
(12.) The words of (in the usual sense of reasonings) the

mouth of a wise man, a favour (i.e. are each one so), but the lips
of the foolish swallow him (future piel, occurs 2 Samuel xx. 19, 20;
Job viii. 18, in the sense of ‘destroy;’ hence the LXX. render
καταποντιοῦσιν; compare Matthew xiv. 30, xviii. 6. Here too we have a
singular verb with a plural noun――‘any one of a fool’s words may
be his destruction.’ Notice also the implied difference――‘a fool talks
with his lips, a wise man reasons’).
13 The beginning of The beginnings of his
the words of his mouth is reasonings are each a
foolishness: and the end wise error, and the result
of his ¹talk is of what he says are
mischievous madness. disappointed
expectations, every one
¹ Hebrew his
of which is mischievous.
mouth.
(13.) The beginning of words (or reasonings) of his mouth,

elaborate follies (‫ סכלות‬in its usual sense; and the whole being
without the article gives the meaning――‘Each beginning of the
reasonings of his mouth is one out of a number of elaborate follies;
his reasonings are themselves elaborate mistakes’), and an end
(‫ אחרית‬is used to signify the last end, Numbers xxiii. 10; see chapter
vii. 8) of his mouth (repeated, ‘that same mouth’) disappointed
expectations (‫הוללות‬, in its usual sense in this book) mischievous
(singular, each one of which is so).
14 A fool also ¹is full And the wise fool

of words: a man cannot multiplies his reasons,
tell what shall be; and though no man
what shall be after him, understands the present,
who can tell him? and the future results no
one can declare.
¹ Hebrew
multiplieth
words.
(14.) And the elaborate fool multiplies words, not knowing

(i.e. when there is no knowing by) the man (humanity generally)
what it is which will be (but the Alexandrine and Vatican read
apparently ‫שהיה‬, γενόμενον, which A². E. X. alter to γενησόμενον,
‘which shall be.’ The Syriac supports the LXX., but Symmachus
reads τὰ προγενόμενα ἀλλ’ οὐδὲ τὰ ἐσόμενα――‘the things which were
before, but not those which come after’――which the Vulgate
follows. Jerome, however, follows the LXX. against the Vulgate;
nevertheless we should not be inclined to alter the text, but would
rather regard the reading of the LXX. as ad sensum――the object
being to give the difference between the contracted and full relative
and the subjunctive meaning attaching to this form. Thus ‫ שיהיה‬is that
which is or exists, the τὸ ὄν――‘he does not know then the real state
of things’――is the meaning; for with this agrees what follows), and
which (full relative) is (or will be) from after him (but there is no
reason why ‫ מאחריו‬should not be considered as a participial noun, as
the LXX. make it, and then we must render the ‘future’ in the sense
of what occurs in the future) who tells to him (emphatic). The
meaning of the passage is――‘That the elaborate fool multiplies
reasonings, which are sure to have an evil tendency, as they are
intended to promote his elaborate folly, although man generally
neither understands the meaning of the present, nor can divine the
future.’ The difficulty of the sentence arises from the play between
‫ מה־שיהיה‬and ‫מאחריו‬.
15 The labour of the A toil of fools will
foolish wearieth every weary them each one,
one of them, because he who has altogether lost
knoweth not how to go his way.
to the city.
(15.) The toil (i.e. ‘anxious care,’ which is the meaning of this
word) of the foolish ones wearies him (another distributive plural;
the result of these various fools’ labour is weariness to each of them.
It is also to be noticed that the verb is feminine, and yet ‫ עמל‬is usually
masculine. Several nouns are, Stuart observes, masculine or
feminine ad libitum scriptoris. There is however, we suspect, a
perceptible difference in the meaning in these cases. The stricter
agreement denotes closer union between the verb and its
nominative; and if this be so, the idea of the passage may be
rendered by ‘the toil of the fools is self-weariness’), which (full
relative, equivalent therefore to ‘because’ he does) not know (or is
instructed) to (in order to) go towards (‫אל‬, LXX. εἰς) a city (not the
city, as is usually rendered.) The obvious meaning would surely be,
that the fool had lost his way, and hence as he is going wrong he has
simply his trouble for his pains.
16 ¶ Woe to thee, O Ah! woe to thee, O

land, when thy king is a country, whose king is a
child, and thy princes eat child, and thy princes eat
in the morning! in the morning.
(16.) Woe to thee, land, whose king is a lad, and thy princes
in the morning eat (i.e. ‘feast,’ the morning being the proper time for
work, and not for feasting. Compare Isaiah v. 11).
17 ¶ Blessed art Blessed art thou, O

thou, O land, when thy country, whose king is
king is the son of nobles, the son of nobles, and
and thy princes eat in thy princes eat in due
due season, for strength, season, for strength and
and not for drunkenness! not for drunkenness.
(17.) Blessings on thee, land, whose king is a son of nobles

(ἐλευθέρου, LXX.), and thy princes in season eat, and not in
drunkenness (but the LXX. render καὶ οὐκ αἰσχυνθήσονται――‘and
shall not be ashamed’――reading the ‫ בשתי‬as though the ‫ ב‬were
radical, and deriving the word from ‫בוש‬, ‘to be ashamed.’ Thus is
probably preserved an intentional equivoke.)
18 ¶ By much When they are idle,

slothfulness the building there is a slender
decayeth; and through support, and when both
idleness of the hands hands hang down, the
the house droppeth roof-tree will weep.
through.
(18.) By idlenesses (Proverbs xix. 15 only; but ‫עצל‬, ‘the
sluggard,’ occurs continually in Proverbs, and once as a verb,
Judges xviii. 9. The word is pointed as a dual, but the meaning
‘idlenesses’ suits the context) decayeth (‫מכך‬, occurs Kal, Psalms
cvi. 43, niphal here, and hiphil Job xxiv. 24, all) the beam (‫ ַה ְּמ ָק ֶר ה‬here
only, but the word differs only in pointing from ‫――ַה ִמ ְק ֶר ה‬the hap, and
the equivoke could hardly be unintentional), and in lowness of
hands drops (occurs Job xvi. 20, Psalms cxix. 28; but notice the
readings of the LXX., which are peculiar) the house.
19 ¶ A feast is made For pleasure they

for laughter, and wine make bread, and wine
¹maketh merry: but rejoices life, but silver
money answereth all subserves with respect
things. to everything.
¹ Hebrew
maketh
glad the
life.
(19.) To laughter are makings (which the LXX. renders by

ποιοῦσιν, ‘they make’) bread and wine rejoices (the Masorets
consider this a piel and transitive) lives, and the silver (with the
article, and therefore generic――money) answereth with respect
to all things (both senses of ‫ יענה‬are given in the versions of the
LXX. ἐπακούσεται, Alexandrine, ‘humbly obeys,’ and ταπεινώσει,
Vatican, ‘will humble.’ The Alexandrine also reads σὺν τὰ πάντα. The
Syriac reads also double, as do some copies of the LXX.――
――‘and money oppresses and
leads them astray in all.’ The Alexandrine reading, however, makes
quite consistent sense, and squares entirely with the rest of the
passage. Bread is prepared for pleasure rather than support, wine
rejoices hearts already merry――its real use is to cheer those who
are faint with toil or sorrow; and silver, which one can neither eat nor
drink, is preferred to bread and wine and everything else).
20 ¶ Curse not the Also, even in thy

king, no not in thy conscience a king do not
¹thought; and curse not revile, and in secret
the rich in thy bed- places of the bed-
chamber: for a bird of chamber neither do thou
the air shall carry the revile the rich: for a bird
voice, and that which of the heavens will carry
hath wings shall tell the out the rumour, and the
matter. swift one on wings shall
tell the matter.
¹ Or,
conscience.
(20.) Also in thy understanding (occurs Daniel i. 4, 17;

2 Chronicles i. 10, 11, 12 only, and always with this meaning: all the
ancient versions follow the idea contained in the LXX.’s συνείδησις,
which would seem to give the notion that this curse was a
reasonable, not a hasty one) a king (not the king, any king) do not
curse; and in the innermost of thy bed-chambers do not either
curse the rich person (the idea of cursing or reviling is of course
here prominent), for a bird of the heavens shall cause to convey
the voice (with ‫ את‬and the article, with ‘respect to that voice’ is the
meaning――the rumour will get abroad in a mysterious way) and a
lord of the winged ones (the Masorets wish to omit the article in
‫ )֯ה כנפים‬shall tell the matter (the LXX. note the emphasis given by ‫ה‬
and the articles by adding the pronoun σοῦ, which is simply a
rendering ad sensum――’Treason, like murder, will out’).
CHAPTER XI.
C AST thy bread

¹upon the waters: C AST thy bread on
the face of the
for thou shalt find it after waters: for in the
many days. multitude of the
days――thou wilt find it.
¹ Hebrew
upon the
face of the
waters.
XI. (1.) Cast thy bread upon the face of the waters, for in the
multitude of the days thou shalt find it. (This passage is usually
taken as an exhortation to liberality. Hengstenberg however
understands it to refer to ships and their cargo of grain. ♦ Zöckler
refers to Proverbs xi. 24 for a similar sentiment, and Luke xvi. 9; the
idea is clearly that of an unexpected return).
♦ “Zökler” replaced with “Zöckler” for consistency
2 Give a portion to Give a share all

seven, and also to eight; round, and to some one
for thou knowest not else beside, for thou
what evil shall be upon dost not know what sort
the earth. of mischief shall be in
the earth.
(2.) Give a portion to seven, and also to eight (see Job v. 19,
Micah v. 4 (5), for similar idioms; it is equivalent to our ‘everybody,
and some one else’), for not dost thou know what shall be
mischief upon the earth.
3 If the clouds be full If the clouds are

of rain, they empty full of rain, they empty
themselves upon the themselves upon the
earth: and if the tree fall earth; and if falls the tree
toward the south, or by the south [wind] or by
toward the north, in the the north――the place
place where the tree where the tree falls is
falleth, there it shall be. just where it will be.
(3.) If they are full the clouds (‫ עב‬is the thick vapour that
appears and disappears) rain (‫ גשם‬is the storm rain which does
mischief or good according to circumstances, see chapter xii. 2) they
cause to empty (clouds do not always prognosticate rain; and even
if they should, a storm may do mischief rather than good); and if is
falling a tree in the south, or if either in the north (‘if’ is hence
emphatic) the place where may fall (contracted relative) the tree
(now with the article, for it is the falling tree spoken of above) there it
will be (the unusual form ‫ יהוא‬has troubled the commentators much:
Moses Stuart pronounces the ‫ א‬to be otiose, which is not explaining
the form at all. But may not the following be a sufficient explanation?
――‫ הוא‬in this book is used in the sense of the existence of an
object: might not Koheleth coin a verb by adding the ‫ י‬of the present
tense, with the idea, ‘makes itself be’?――compare also Joshua
x. 24, Isaiah xxiii. 12, where this otiose ‫ א‬occurs; the rendering of the
LXX. by ἔσται shows how they understood it, and so also the Syriac
and Vulgate. The whole sentence is ironical, when the tree has really
fallen, then we know which way it fell. The Masoretic accentuation of
this passage is peculiar――we should naturally have expected them
to have divided the verse into two clauses, at ‫יריקו‬, ‘they empty,’
instead of which the greatest pause occurs at ‘north’ ‫ַּב ָּצ ֑פ ֹון‬, but this
method of reading renders the irony of the passage; the verse will
then stand thus:――‘If the clouds are full of rain they will empty
themselves upon the earth, and so if the tree should incline to the
south, or if it should incline to the north――the place where it falls is
where it really will be.’ The accentuation is rhetorical rather than
logical, and the Masorets have shown great taste in their pointing).
4 He that observeth Looking at the wind

the wind shall not sow; one does not sow, and
and he that regardeth gazing into the clouds
the clouds shall not one does not reap.
reap.
(4.) Regarding wind! not does one sow (impersonal), and

looking into clouds neither is one reaping (we must attend to the
precise form of the words in this sentence in order to gather the true

Instant Download Quantum Stochastic Thermodynamics: Foundations and Selected Applications (Oxford Graduate Texts) Philipp Strasberg PDF All Chapter

Uploaded by

Copyright:

Available Formats

Instant Download Quantum Stochastic Thermodynamics: Foundations and Selected Applications (Oxford Graduate Texts) Philipp Strasberg PDF All Chapter

Uploaded by

Document Information

Original Description:

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

Instant Download Quantum Stochastic Thermodynamics: Foundations and Selected Applications (Oxford Graduate Texts) Philipp Strasberg PDF All Chapter

Uploaded by

Copyright:

Available Formats

Full download test bank at ebook ebookmass.

Quantum Stochastic Thermodynamics:

Discrete Communication Systems (Oxford Graduate Texts)

Thermodynamics and Applications in Hydrocarbon Energy

Quantum Mechanics 3rd Edition Nouredine Zettili

Introduction to Quantum Field Theory with Applications

Optimizations and Programming: Linear, Non-linear,

Quantum Materials, Devices, and Applications Mohamed

ChatGPT for Beginners: Features, Foundations, and

Quantum Dots: Fundamentals, Synthesis and Applications

strategies, or energy-efficient computers, among other more exotic applications. To

How to Read this Book

References: In contrast to the exercises, all references are relegated to a special

General mathematics: Matrices are denoted by capital letters (such as M ). Bold-

Statistical mechanics: Equilibrium concepts are denoted by calligraphic letters

1 Quantum Stochastic Processes 1

4.3 Exchange and Further Fluctuation Theorems 183

1.1 Isolated Quantum Systems

ρ(t) = U (t)ρ(0)U (t)† , (1.2)

where ρ(0) denotes the initial state of the quantum system.

Quantum Stochastic Thermodynamics. Philipp Strasberg, Oxford University Press.

Here, λk is time independent and |ψk (t)i belongs to an orthonormal basis of

det{ρ(t) − λI} = det{U (t)[ρ(0) − λI]U (t)† } = det{ρ(0) − λI}. (1.4)

Here, det{. . . } denotes the determinant of a matrix.

SvN (ρ) ≡ −tr{ρ ln ρ}, (1.6)

which becomes exact in the limit n → ∞.

1.2 System–Bath Theories and the Origin of Noise

ρS (t) = trB {ρSB (t)}. (1.11)

ρSB (0) = ρS (0) ⊗ ρB (0). (1.14)

√ simulations, we take ρS (0) = |+ih+|S to be a pure state. Here, |+i ≡ (|0i +

D(ρ|σ) ≡ tr{ρ(ln ρ − ln σ)} ≥ 0 (1.16)

1.3 Equilibrium States of Open Quantum Systems

i.e. it is a sum of periodic functions with frequencies ωn ≡ En /h̄. If we wait long

ZSB trSB {e−βHSB }

However, similar to the case of equilibration, thermalization should also be re-

1.4 Quantum Measurement Theory

The average or unconditional post-measurement state ρ′S of the system follows as

where we interpret r + x − 1 modulo n whenever r + x − 1 > n. It is easy to verify

ρ̃′S (r) ≡ trA {|rihr|A V ρS ⊗ ρA V † } = ΠS (r)ρS ΠS (r). (1.31)

Next, we introduce the operators

This equation suggests introducing the following probability operators:

p we relabel j ≡ x in eqn (1.33) and model the measurement with operators

For {M (r)} to be a POVM,

To simplify the notation, we introduce the multi-index α = (r, j) and write

1.5 Operations, Interventions and Instruments

We remark that the distinction between positivity and complete positivity is a

To conclude, physically legitimate operations are mathematically described by an

for all r. In particular, a map C is CPTP if and only if it can be written as

We remark that it is also possible to use the operator–sum representation to derive

1.6 Classical Stochastic Processes

p(rn , . . . , r1 , r0 ) ≡ p(rn , tn ; . . . ; r1 , t1 ; r0 , t0 ). (1.51)

To be a valid probability distribution, p(rn , . . . , r1 , r0 ) must satisfy

Here, SSh (p) ≡ −p ln

1.7 Quantum Stochastic Processes

Quantum statistics of a projectively measured system

(3.) And moreover in the way (which word ‘way’ is so

4 If the spirit of the If the spirit of the

(4.) If a spirit of the ruling one (not, as usually rendered, the