Beruflich Dokumente
Kultur Dokumente
John Goold,1 Marcus Huber,2, 3, 4 Arnau Riera,4 Ldia del Rio,5 and Paul Skrzypczyk4, 5
1
The Abdus Salam International Centre for Theoretical Physics (ICTP), Trieste, Italy
2
Group of Applied Physics, University of Geneva, 1211 Geneva 4, Switzerland
3
Universitat Autonoma de Barcelona, 08193 Bellaterra, Barcelona, Spain
4
ICFO-Institut de Ciencies Fotoniques, The Barcelona Institute of
Science and Technology, 08860 Castelldefels (Barcelona), Spain
5
H. H. Wills Physics Laboratory, University of Bristol, Bristol BS8 1TL, United Kingdom
(Dated: August 29, 2016)
This topical review article gives an overview of the interplay between quantum information the-
ory and thermodynamics of quantum systems. We focus on several trending topics including the
foundations of statistical mechanics, resource theories, entanglement in thermodynamic settings,
fluctuation theorems and thermal machines. This is not a comprehensive review of the diverse field
of quantum thermodynamics; rather, it is a convenient entry point for the thermo-curious infor-
mation theorist. Furthermore this review should facilitate the unification and understanding of
different interdisciplinary approaches emerging in research groups around the world.
arXiv:1505.07835v3 [quant-ph] 26 Aug 2016
An early puzzle in thermodynamics: imagine a box filled with a gas, with a partition in the middle. An
agent (the demon) who can observe the microscopic details of the gas particles, controls a small gate in the
partition, selectively opening it to let slow particles flow to the left and fast ones to the right. This creates
a temperature differential between the two sides. The demon can exploit this difference to extract work, by
letting the hot gas on the right expand. This apparent contradiction with the second law of thermodynamics
can be easily explained from an information-theoretical viewpoint. The demon had access to much more
information than the standard observer assumed in the derivation of traditional thermodynamics, who can
only read a few macroscopic parameters of a gas and assumes a uniform distribution over all compatible
micro-states. Therefore, it seems natural that the demon may extract more work than predicted by standard
thermodynamicsand this insight motivates the need for thinking of thermodynamics as a subjective resource
theory, and extending it to the quantum regime. In the larger picture, Bennett showed that the amount of
work needed to erase the demons memory at the end of the procedure (or equivalently, to prepare the memory
to store the necessary information on the particles in the beginning) precisely makes up for the work extracted
[1]. For reviews, see [26].
tasks, thermodynamic witnesses of non-classicality, and non-equilibrium ones, and led to a revision on how we un-
entanglement witnesses in phase transitions. derstand the thermodynamics of systems far from equi-
Information theory also shed light on fundamental is- librium [2024]. Although this approach is relatively
sues in statistical mechanics - the mathematical back- recent from a statistical physics perspective, a cross-
bone of thermodynamics. Perhaps one of the earliest sig- fertilisation with concepts ubiquitous in quantum infor-
nificant contributions is the maximal entropy principle mation theory has already started, including phase esti-
introduced by Jaynes [17, 18]. In these seminal works mation techniques for extraction of work and heat statis-
Jaynes addresses the issue of justifying the methods of tics and feedback fluctuation theorems for Maxwells
statistical mechanics from microscopic mechanical laws demons. In Section V we identify these existing rela-
(classical or quantum) using tools from information the- tionships and review areas where more overlap could be
ory. In fact, deriving statistical mechanics, and hence developed.
thermodynamics from quantum mechanics is almost as As ideas and concepts emerge and develop it is not sur-
old as quantum mechanics itself starting with the work prising that quantum information theorists have started
of von Neumann [11, 19]. This is still very much and to turn towards the pragmatic goal of describing the ad-
ongoing and active research area and in recent years has vantages and disadvantages of machines which operate
received significant attention from the quantum informa- at and below the quantum threshold. Although ideas re-
tion community. The most significant contributions are lating quantum engines have been around now for a sur-
reviewed in Section II. prising long time [2527] - questions pertaining to the in-
In the past twenty years, the field of non-equilibrium trinsic quantumness in the functioning of such machines
statistical mechanics has seen a rapid development in have been raised using the tools of quantum information
the treatments of driven classical and quantum systems theory only relatively recently. We review progress along
beyond the linear response regime. This has culmi- these lines in Section VI. In summary, we will review
nated in the discovery of various fluctuation theorems landmark and recent articles in quantum thermodynam-
which relate equilibrium thermodynamic quantities to ics, discuss different approaches and models, and peek
4
In 1932, von Neumann designed this thought experiment to determine the entropy of a density operator [11].
The experiment Paccounts for the work cost of erasing the state of a gas of n atoms, initially in an ensemble
n
n , with = k pk |k i hk |, by transforming it into a pure state |1 i by means of a reversible process.
It consists of 3 steps: 1. Separation of the species: the atoms in different states |1 i,. . . ,|m i inside a box
of volume V are separated in different boxes of the same volume V by means of semi-permeable walls (from
a to b and finally c). Note that no work has been done and no heat has been exchanged. 2. Compression:
every gas |k i is isothermally compressed to a volume Vk = pk V (from c to d). The mechanical work done
in thatPprocess is Wk = npk ln(Vk /V ) = pk ln pk . The total entropy increase per particle of that process is
S = k pk ln pk . 3. Unitary transformation: every gas is put in the |1 i state by applying different unitary
transformations |k i |1 i, which are taken for free (from d to e). As the entropy of the final state is zero,
the entropy of the initial ensemble reads S() = Tr( ln ).
Historically, it is remarkable that the Shannon entropy, which can be seen as a particular case of the von
Neumann entropy for classical ensembles, was not introduced until 1948 [7], and Landauers principle was
proposed only in 1961 [12].
In Ref. [39] it is shown that, if we look only at the sub- be equivalent to the set of local random quantum circuits,
system S, most of the states in HR are indistinguishable that is, quantum circuits of qubits composed of polyno-
from the equal a priori probability state, i. e. for most mially many nearest neighbour two-qubit gates [42]. In
|i HR , TrB |i h| S . More explicitly, if |i is Ref. [48], it was shown that the local random quantum
randomly chosen in HR according to the uniform distri- circuits form an approximate unitary 2-design, i. e. that
bution given by the Haar measure, then the probability random circuits of only polynomial length will approx-
that TrB |i h| can be distinguished from S decreases imate the first and second moments of the Haar distri-
exponentially with the dimension of HR , dR bution. In Ref. [44] the previous work was extended to
poly(n)-designs. Finally, let us mention that the entan-
1/3 1/3 glement properties of typical physical states were studied
Prob kTrB (|i h|) S k1 dR 2 exp CdR ,
in Ref. [45].
(3)
Let us mention that k-designs also appear naturally in
where C is a constant and kk1 is the trace norm. The the context of decoupling theorems in which a the sub-
trace norm k k1 measures the physical distinguisha- system S undergoes a physical evolution separated from
bility between the states and in the sense that a the environment B, and one wonders under what con-
k k1 = supO1 | Tr(O) Tr(O)|, where the maxi- ditions this evolution destroys all initial correlations be-
mization is made over all the observables O with operator tween S and B. In particular, in Ref. [49] it is shown
norm bounded by 1. The proof of Eq. (3) relies upon con- that almost-2-designs decouple the subsystem S from B
centration of measure and in particular on Levys Lemma independently of Bs size.
(see Ref. [39] for details). Let us mention that ideas in Another objection against typicality is that there are
this spirit can be already found in S. Lloyds Ph.D. thesis many physically interesting systems, e. g. integrable mod-
[41] published in 1991. In particular, he presents bounds els, which, although their initial state belongs to a cer-
on how the expectation values of a fixed operator taken tain restricted subspace HR , their expectation values dif-
over random pure states of a restricted subspace fluctu- fer from the completely mixed state in R, R , as ex-
ate. pected from typicality arguments. This is a consequence
The weakness of the previous result lies in that the use of the fact that their trajectories in the Hilbert subspace
of typicality is made in the whole subspace HR and, as HR dont lie for the overwhelming majority of times on
we will justify next, this is not a physical assumption. In generic states (see Fig. 3). Hence, in practice, state-
nature, Hamiltonians have local interactions and systems ments on equilibration and thermalization will depend
evolve for times that are much smaller than the age of on the dynamical properties of every system, that is, on
the universe. Most states in the Hilbert space simply their Hamiltonian. This leads us to the notion of dy-
cannot be generated by evolving an initial product state namical typicality. In contrast to the kinematic typical-
under an arbitrary time-dependent local Hamiltonian in ity presented in this section, where an ensemble has been
a time that scales polynomially in the system size [42]. defined by all the states that belong to a certain sub-
Therefore, sampling uniformly from the whole Hilbert space, in dynamical typicality the ensemble is defined by
space is not physically meaningful. There has been a all states that share the same constants of motion given
strong effort to generalize the concept of typicality for a Hamiltonian H and an initial state |(0)i. Studying
different sets of states [4345]. whether typicality also holds in such a set will be pre-
The first realistic set of states in which typicality cisely the problem addressed in the next section.
was studied was the set of matrix product states (MPS)
[46, 47]. These type of states have been proven to de-
scribe ground states of one-dimensional gapped Hamil- B. Equilibration. Maximum entropy principle from
tonians. They are characterized by the rank of a bipar- quantum dynamics
tition of the state. This parameter quantifies the maxi-
mum entanglement between partitions of an MPS. The In this context of deriving thermodynamics from quan-
MPSs with fixed rank form a set of states with an efficient tum mechanics the first problem that needs to be ad-
classical representation (they only require polynomial re- dressed is equilibration, that is, understand how the re-
sources in the number of particles). In Ref. [43], it is versible unitary dynamics of quantum mechanics make
proven that typicality occurs for the expectation value of systems equilibrate and evolve towards a certain state
subsystems observables when the rank of the MPS scales where they remain thereafter.
polynomially with the size of the system with a power Because of the unitary dynamics, equilibration is only
greater than 2. possible if the set of observables is restricted. In this
Another set recently considered in the literature has spirit, a set of sufficient conditions for equilibration to-
been the so called set of physical states which consists wards the time averaged state has been presented for lo-
of all states that can be produced by evolving an ini- cal observables [50, 51] and observables of finite precision
tial product state with a local Hamiltonian for a time [52, 53]. The two approaches are proven to be equivalent
polynomial in the number of particles n. By Trotter de- in Ref. [54] and it is remarkable that the conditions given
composing the Hamiltonian, such a set can be proven to are weak and naturally fulfilled in realistic situations.
7
Scheme of the restricted subspace HR with its untypical states forming little islands coloured in yellow. The
left trajectory (dashed line) passes mostly on typical states while the right trajectory (solid line) has a non-
negligible support on states that are not typical.
For simplicity, let us here focus on equilibration of on below), then the average distance from equilibrium is
subsystems and, as above, identify in the total system bounded by
a subsystem S and its environment B. The dynam- s s
ics of the total system are governed by the Hamilto- dS d2S
nian H with eigenvalues {Ek }k and eigenvectors {|Ek i}k . hkS (t) S k1 it , (6)
deff ( B ) deff ()
This leads to the time evolution |(t)i = eiHt |(0)i
and the reduced state of S is S (t) = TrB (t) with
(t) = |(t)ih(t)|. where deff () := 1/ Tr(2 ) is the effective dimension of
and B = TrS . Roughly speaking, the effective di-
If equilibration happens, then it happens towards the
mension of a state tells us how many eigenstates of the
time averaged state i. e. S := TrB with
Hamiltonian support such state. It can also be related to
Z T the 2-Renyi entanglement entropy by S2 () = log deff ().
1 X
= lim (t)dt = Pk (0)Pk (4) Hence, equation (6) guarantees equilibration for Hamil-
T T 0 k tonians with non-degenerate energy gaps as long as the
initial state is spread over many different energies.
with Pk the projectors onto the Hamiltonian eigenspaces. Although the condition of having non-degenerate gaps
The time averaged state is the initial state dephased in may look very restrictive at first sight, note that Hamilto-
the Hamiltonian eigenbasis. For this reason it is also nians that do not fulfil it form a set of zero measure in the
called diagonal ensemble. set of Hamiltonians, since any arbitrarily weak perturba-
In Ref. [50], a notion of equilibration is introduced by tion breaks the degeneracy of the gaps. In Ref. [51], the
means of the average distance (in time) of the subsys- non-degenerate gaps condition was weakened by showing
tem S (t) from equilibrium. A subsystem S is said to that equilibration occurs provided that no energy gap is
equilibrate if hugely degenerate. This condition can be understood as
a way of preventing the situation where there is a sub-
Z T
1 system which does not interact with the rest.
hkS (t) S k1 it := lim dt kS (t) S k1 1 ,
T T 0 Let us finally point out that the equilibrium state intro-
(5) duced in Eq. (4) is precisely the state that maximizes the
where kS (t) S k1 is the trace distance. If this average von Neumann entropy given all the conserved quantities
trace distance can be proven to be small, then the sub- [55]. This observation turns the principle of maximum
system S is indistinguishable from being at equilibrium entropy into a consequence of the quantum dynamics.
for almost all times. The principle of maximum entropy was introduced by
Equilibration as a genuine property of quantum me- Jaynes in Ref. [17] and states that the probability distri-
chanics is shown in Ref. [50] by precisely proving that this bution which best represents the current state of knowl-
average distance is typically small. More concretely, if edge of the system is the one with largest entropy given
the Hamiltonian that dictates the evolution of the system the conserved quantities of the system. We will come
has non-degenerate gaps i. e. all the gaps of the Hamilto- back in more detail to the Jaynes principle in the next
nian are different (an assumption which we will comment section when the thermalization for integrable systems is
8
specifying only a polynomial number of conserved quan- equilibration times have been proven for generic observ-
tities? If so, what are these relevant conserved quantities ables [67], Hamiltonians [6872], and initial states [73].
Q that allow for an accurate and efficient representation In nature, systems seem to equilibrate in a time that is
of the ensemble? This question is tackled in Ref. [65]. neither microscopic nor exponential in the system size. A
There, it is argued that the relevant conserved quanti- relevant open question is what properties of the Hamilto-
ties are the ones that make the GGE as close as possible nians and operators lead to reasonable equilibration time.
to the diagonal ensemble in the relative entropy distance As a first step, in Ref. [74], a link between the complexity
D(||GGE ), which in this particular case can be written of the Hamiltonians eigenvectors and equilibration time
as is presented. The result does not completely solve the
question, since the given bounds are not fulfilled by all
D(||GGE ) = S(GGE ) S() , (8) Hamiltonians but only by a fraction of them, and further
research in this direction is needed.
where we have used that the diagonal ensemble and
the GGE have by construction the same expectation
values for the set of selected conserved quantities, E. Outlook
i. e. Tr(Q GGE ) = Tr(Q ). Equation (8) tells us that
the relevant conserved quantities are the ones the min- The aim of this section has been to justify that thermal
imize the entropy S(GGE ). Note that in contrast to states emerge in Nature for generic Hamiltonians. To
Jaynes approach, where entropy is maximized for a set of complete the picture presented here we recommend the
observables defined beforehand, here the notion of phys- article [28] where an extensive review of the literature on
ically relevant is provided by how much an observable is foundations of statistical mechanics is provided.
able to reduce the entropy by being added into the set of The main ideas presented here have also been widely
observables that defines the GGE. studied in the context of condensed matter physics, in
If instead of calculating the relative entropy between which systems are typically brought out of equilibrium
the diagonal ensemble and the GGEs we do it with re- by sudden (and slow) quantum quenches: the Hamilto-
spect to the set of product states, i. e. nian of a system (that is initially in the ground state)
is suddenly (or smoothly) changed in time. We recom-
T () := min D(||1 2 . . . n ), (9) mend the review article [29] on non-equilibrium dynamics
1 ,2 ,...,n
of closed interacting quantum systems.
then we obtain a measure of the total (multipartite) corre- Let us finish the section with a list of some of the open
lations of the diagonal ensemble. In Ref. [66] the scaling problems that we consider most relevant in the field:
with system size of the total correlations of the diagonal
Typicality for symmetric states. Hamiltonians in
ensemble has been shown to be connected to ergodic-
nature are not generic but have symmetries. Hence,
ity breaking and used to investigate the phenomenon of
the notion of typicality should be extended to phys-
many-body localization.
ical states that are produced by symmetric Hamil-
tonians.
D. Equilibration times Quantum notion of integrability. One of the reasons
why it is so difficult to extract strong statements
Maybe the major challenge that is still open in the on the equilibration and thermalization of many
equilibration problem is to determine the equilibration body quantum systems is the absence of a satis-
timescale. It turns out that even if we know that a system factory quantum notion of integrability [75]. This
equilibrates, there are no relevant bounds on how long leads first to some widespread confusion, since in-
the equilibration process takes. There could be quan- tegrability is mentioned very often in the field of
tum systems that are going to equilibrate, but whose non-equilibrium dynamics, and second it does not
equilibration times are of the order of magnitude of the allow us to classify quantum systems into classes
age of the universe, or alternatively, some systems, like with drastically different physical behaviour, like
glasses, which do not relax to equilibrium at all, but have what occurs in classical mechanics.
metastable states with long lifetimes. The problem of es- Equilibration times. Without bounds on the equi-
timating equilibration timescales is thus essential in order libration time scales, statements on equilibration
to have a full understanding of thermalization. become useless. As we have seen, the equilibration
So far, progress on this issue has taken place from two times are model dependent. We need then to un-
different approaches. On the one hand, rigorous and derstand how the equilibration times depend on the
completely general bounds on equilibration times have features of the Hamiltonian and the set of observ-
been presented in Ref. [51]. Due to their generality, ables considered.
these bounds scale exponentially with the system size,
leading to equilibration times of the age of the universe Relative thermalization. It was highlighted in
for macroscopic systems. On the other hand, very short Ref. [76] that local thermalization of a subsystem S,
10
as described here, is not enough to guarantee that ; this results from composing operations one after the
S will act as thermal bath towards another physical other).
system R. In other words, imagine that we want to The task now is to find general properties of this struc-
perform quantum thermodynamics on a reference ture. A paradigmatic example is looking for simple nec-
system R, using S as a thermal bath. To model a essary and sufficient conditions for state transformations.
thermodynamic resource theory that recovers the The most general case are functions such that
laws of thermodynamics, it is not sufficient to de-
mand that S be in a local Gibbs state S (). Indeed f (, ) 0 (that is, f (, ) 0 is a
we need S to be thermalized relative to R, that necessary condition for state transformations), or
is the the two systems should be uncorrelated, in
f (, ) 0 (that is, f (, ) 0 is a
global state, S () R . If this does not hold, then
sufficient condition for state transformations).
we cannot recover the usual thermodynamic mono-
tones (for instance, there could be anomalous heat Often, we try to find necessary and sufficient conditions
flows against the temperature gradient). Therefore, as functions that can be written like f (, ) = g()
the relevant question for resource theories of ther- h(). In the special case where g = h for a necessary
modynamics is not only does S thermalize locally condition ( g() g()), we call g a monotone
after evolving together with an environment?, but of the resource theory. For example, in classical, large-
rather does S thermalize relative to R after evolv- scale thermodynamics, the free energy is a monotone.
ing together with an environment?, and the results In order to quantify the cost of state transformations,
discussed in this section should be generalized to we often fix a minimal unit in terms of a standard resource
that setting. First steps in this direction can be that can be composed. For example, in entanglement
found in Ref. [76], where the authors use decou- theory the standard resource could be a pair of maximally
plinga tool developed in quantum information entangled qubits, and in quantum thermodynamics we
theoryto find initial conditions on the entropies could take a single qubit (with a fixed Hamiltonian) in a
of the initial state that lead to relative thermaliza- pure state. The question then is how many pure qubits
tion. do I need to append to in order to transform it into
? or, more generally, what is the cost or gain, in terms
of this standard resource, of the transformation ?
III. RESOURCE THEORIES [7779].
One may also try to identify special sets of states. The
In the previous section we saw the progress that has most immediate one would be the set of free states: those
been made in understanding how systems come to equi- that are always reachable, independently of the initial
librium, in particular thermal equilibrium, and as such state. In standard thermodynamics, these tend to be
a justification for the thermal state. In the rest of this what we call equilibrium states, like Gibbs states. An-
review we will now take the thermal state as a given, other interesting set is that of catalysts, states that can
and see what is the thermodynamics of quantum systems be repeatedly used to aid in transformations. We will
which start thermal or interact with thermal states. We revisit them shortly.
will start from an operational point of view, treating the
thermal state as a free resource, a view inspired by
other resource theories from quantum information. A. Models for thermodynamics
In this section we discuss the approach of thermo-
dynamics as a resource theory in more detail. Let us Now that we have established the basic premise and
start by introducing the basic ideas behind resource the- structure of resource theories, we may look at different
ories that can be found in the literature, entanglement models for resource theories of thermodynamics, which
theory being the paradigmatic example. The first step vary mostly on the set of allowed operations. In the good
is to fix the state space S, which is usually compati- spherical cow tradition of physics, the trend has been to
ble with a composition operationfor instance, quan- start from a very simple model that we can understand,
tum states together with the tensor product, in systems and slowly expand it to reflect more realistic scenarios.
with fixed Hamiltonians. The next step is to define the In general there are two types of operations allowed: con-
set of allowed state transformations. For thermodynam- tact with a thermal bath and reversible operations that
ics, these try to model adiabatic operationslike energy- preserve some thermodynamic quantities. Each of those
preserving reversible operations, and contact with a heat may come in different flavours.
bath.
The set of allowed operations induces a structure on
the state space: we say that if there is an allowed 1. Noisy and unital operations
transformation from to . The relation is a pre-
order, that is, a binary relation that is both reflexive In the simplest case, all Hamiltonians are fully degen-
( ) and transitive ( and implies erate, so thermal states of any temperature are just fully
11
mixed states, and there are no special conserved quan- results are only known for classical states (states that
tities. In this setting, thermodynamics inherits directly are block-diagonal in the energy eigenbasis) [83] and any
from the theory of noisy operations [80]. We may model state of a single qubit [90, 91]. In the limit of a fully
contact with a thermal bath as composition with any sys- degenerate Hamiltonian, we recover the resource theory
tem in a fully mixed state, and reversible operations as of noisy operations.
any unitary operation. Furthermore, we assume that we
can ignore, or trace out, any subsystem. Summing up,
noisy operations have the form 3. Gibbs-preserving maps
1B
T (A ) = TrA0 UAB A UAB , Following the example of the theory of noisy opera-
|B| tions, we could try to replace these thermal operations
where A0 is any subsystem of AB and U is a unitary with so-called Gibbs-preserving maps, that is, maps such
matrix. Alternatively, we may allow only for maps that that TAB (A ()) = B (). This constraint is easier
preserve the fully mixed state, TAB : TAB ( |A| 1A ) = 1B , to tackle mathematically, and the two resource theo-
|B|
called unital maps (an example would be applying one ries induce the same pre-order on classical states, lead-
of two isometries and then forgetting which one). The ing to a condition for state transformation called Gibbs-
two setsnoisy operations and unital mapsinduce the majorization (which is majorization after a rescaling of
same pre-order structure in the state space. In this set- the eigenvalues) [88]. However, Gibbs-preserving maps
ting, majorization is a necessary and sufficient condition are less restrictive than thermal operations for general
for state transformations [80]. Roughly speaking, ma- quantum states [93]. For example, suppose that you
jorization tells us which state is the most mixed. Let have a qubit with the Hamiltonian H = E |1i h1|, and
r = (r1 , r2 , . . . , rN ) and s = (s1 , s2 , . . . , sN ) be the eigen- you want to perform the transformation |1i |+i =
values of two states and respectively, in decreasing (|0i + |1i)/ 2. This is impossible through thermal op-
Pk Pk erations, which cannot create coherence; yet there exists
order. We say that r majorizes s if i=1 ri i=1 si ,
a Gibbs-preserving map that achieves the task. We may
for any k N . In that case ; monotones
still use Gibbs-preserving maps to find lower bounds on
for this setting are called Schur monotone functions, of
performance, but at the moment we cannot rely on them
which information-theoretical entropy measures are ex-
for achievability results, as they are not operationally de-
amples [79, 8184]. For example, if majorizes , then
fined.
the von Neumann entropy of , S() = Tr( log2 ), is
smaller than S(). For a review, see [84].
4. Coherence
2. Thermal operations
The difference between thermal operations and Gibbs-
preserving maps is not the only surprise that quantum
The next step in complexity is to let systems have non-
coherence had in store for thermodynamics enthusiasts.
degenerate Hamiltonians. The conserved quantity is en-
The question of how to create coherence in the first place
ergy, and equilibrium states are Gibbs states of a fixed
led to an intriguing discovery. In order to achieve the
temperature T . For instance for a system A with Hamil-
above transformation |1i |+i through thermal oper-
tonian HA , the equilibrium state is A () = eHA /Z.
ations, we need to draw coherence from a reservoir. A
We can model contact with a heat bath as adding any
simple example of a coherence reservoir Pwould be a dou-
system in a Gibbs statethis corresponds to the ideal-
bly infinite harmonic oscillator, H = n= n |ni hn|,
ization of letting an ancilla equilibrate for a long time. Pa+N
A first approach to model physical reversible transfor- in a coherent state like |i = N 1 n=a |ni. Lasers ap-
mations is to allow for unitary operations U that pre- proximate such reservoirs, which explains why we can use
serve energyeither absolutely ([U, H] = 0) or on aver- them to apply arbitrary transformations on quantum sys-
age (Tr(H) = Tr(H (U U )) for specific states). Fi- tems like ion traps. One may ask what happens to the
nally, we are again allowed to forget, or trace out, any reservoir after the transformation: how much coherence
subsystem. Together, these transformations are called is used up? Can we use the same reservoir to perform
thermal operations, a similar operation in a new system? The unexpected
answer is that coherence is, in a sense, catalytic: while
T (A ) = TrA0 UAB [A B ()] UAB , the state of the reservoir is affected, its ability to imple-
ment coherent operations is not [94]. What happens is
where A0 is any subsystem of AB and U is an energy- that the state of the reservoir spreads out a little with
conserving unitary [85]. The monotones found so far are each use, but the property that determines the efficacy of
different versions of the free energy, depending on the the reservoir to implement operations stays invariant. In
exact regime [83, 8689] (see Example 1). It is worth more realistic models for coherence reservoirs, where the
mentioning we can build necessary conditions for state Hamiltonian of the reservoir has a ground state, the cat-
transformations from these monotones, but sufficiency alytic properties hold for some iterations, until the state
12
This is an example of finding monotones for the resource theory of thermal operations [86]. We are interested
in finding the optimal rates of conversion between two states and , in the limit of many independent copies,
R( ) := sup lim n Rn .
R n
If both R( ), R( ) > 0, and these quantities represent optimal conversion rates, then the process
must be reversible, that is, R( ) = 1/R( ); otherwise we could build a perpetual motion engine,
and the resource theory would be trivial. The idea is to use a minimal, scalable resource as an intermediate
step. We can think of as a currency: we will sell n copies of for a number of coins, and use them to buy
some copies of . To formalize this idea, we define the selling and buying cost of a state , or more precisely
the distillation and formation rates,
RD () := R( ), RF () := R( ) = 1
RD ()
.
spreads all the way down to the ground state. At that 5. Catalysts
stage, the reservoir needs to be recharged with energy
to pump up the state again. Crucially, we do not need The catalytic nature of coherence raises more general
to supply additional coherence. In the converse direc- questions about catalysts in thermodynamics. Imagine
tion, we know that coherence reservoirs only are critical that we want to perform a transformation in a
in the single-shot regime of small systems. Indeed, in the system S, and we have access to an arbitrary ancilla in
limit of processing many copies of a state simultaneously, any desired state . Now suppose that our constraint is
the work yields of doing it with and without access to a that we should return the ancilla in a state that is -close
coherence reservoir converge [95] to :
S A SA : kA A k1 .
The question is whether we can overcome the usual limits
found in thermal operations by use of this catalyst. In
13
other words, can we perform the above transformation approximately be implemented for free (if we neglect the
in cases where would not be allowed? It turns informational cost of designing the global Hamiltonian).
out that if no other restrictions are imposed on the cata- Note that this is still an idealized scenario, in which the
lyst, then for any finite and any two states and , we clock is infinite-dimensional and moves like a relativistic
can always find a (very large) catalyst that does the job particle (the Hamiltonian is proportional to the parti-
[83]. These catalysts are the thermodynamic equivalent cles momentum). A relevant open question is whether
of embezzling states in LOCC [96]. However, if we impose there exist realistic systems with the properties assigned
reasonable energy and dimension restrictions on the cat- to this clock, or alternatively how to adapt the protocol
alyst, we recover familiar monotones for state transfor- to the behaviour of known, realistic clocks. That direc-
mations [83, 97]. These restrictions and optimal catalysts tion of research can be related to the resource theory of
result from adapting the concept of trumping relations on quantum reference frames [90, 100102]. An alternative
embezzling states [98, 99] to the thermodynamic setting. direction would be to ask what happens if we do not have
In particular, if we demand that n1 , where n is a clock at allcan we extract all the work from a quan-
the number of qubits in the catalyst, we recover the free tum state if we are only allowed weak thermal contact?
energy constraint for state transformations [97]. A rele- This question is studied (and answered in the negative,
vant open question, motivated by the findings of catalytic for general states) in Ref. [103].
coherence, is what happens if we impose operational con-
straints on the final state of the catalyst. That is, instead
of asking that it be returned -close to , according to the 7. Free states and passivity
trace distance, we may instead impose that its catalytic
properties stay unaffected. It would be interesting to see It is now time to question the other assumption behind
if we recover similar conditions for allowed transforma- the framework of thermal operations: that Gibbs states
tions under these constraints. come for free. There are two main arguments to support
it: firstly, Gibbs states occur naturally under standard
conditions, and therefore are easy to come by; secondly,
6. Clocks they are useless on their own. The first point, typicality
of Gibbs states, is essentially the fundamental postulate
All of resource theories mentioned allow for energy- of statistical mechanics: systems equilibrate to thermal
preserving unitary operations to be applied for free. That states of Gibbs form. This assumption is discussed and
is only the first order approach towards an accurate ultimately justified from first principles in Section II The
theory of thermodynamics, though. Actually, in order second point is more subtle. Pusz and Woronowicz first
to implement a unitary operation, we need to apply a introduced the notion of passive states, now adapted to
time-dependent Hamiltonian to the systems involved. To the following setting [108110]. Let S be a system with a
control that Hamiltonian, we require very precise time- fixed Hamiltonian H, in initial state . We ask whether
keepingin other words, precise clocks, and we should there is a unitary U that decreases the energy of S, that
account for the work cost of using such clocks. Fur- is
thermore, clocks are clearly out of equilibrium, and us- Tr(H) > Tr(U U H).
ing them adds a source of free energy to our systems.
Including them explicitly in a framework for work ex- If we can find such a unitary, then we could extract work
traction forces us to account for changes in their state, from S by applying U and storing the energy difference
and ensures that we do not cheat by degrading a clock in a weight system. If there is no U that achieves the
and drawing its free energy. First steps in this direc- condition above, then we cannot extract energy from ,
tion can be found in [67]. There, the goal is to im- and we say that the state is passive. The latter applies to
plement a unitary transformation in a system S, using classical states (i.e., diagonal in the energy basis) whose
a time-independent Hamiltonian. For this, the authors eigenvalues do not increase with energy. However, sup-
introduce an explicit clock system C hat runs continu- pose that now we allow for an arbitrary number n many
ously, as well as a weight W that acts as energy and copies of and a global unitary Ugl . The question be-
coherence reservoir. The global system evolves under a comes whether
time-independent Hamiltonian, designed such that the 1
Hamiltonian applied on S depends on the position of Tr(H) > Tr(Ugl n Ugl Hgl ),
the clockwhich effectively measures time. The authors n
show that such a construction allows us to approximately where Hgl is the global Hamiltonian, which is the sum of
implement any unitary operation on S, while still obeying the independent local Hamiltonians of every system. If
the first and second laws of thermodynamics. Further- this is not possible for any n, we say that is completely
more, the clock and the weight are not degraded by the passive, and it turns out that only states of Gibbs form,
procedure (just like for catalytic coherence). In particu- = () are completely passive. Moreover, Gibbs states
lar, this result supports the idea behind the framework of are still completely passive if we allow each of the n sub-
thermal operations: that energy-conserving unitaries can systems to have a different Hamiltonian, as long as all
14
The extreme case where one of our resources is in itself a second heat bath is of particular interest. This is
a very natural scenario in traditional thermodynamics: steam engines used a furnace to heat a chamber, and
exploit the temperature difference to the cooler environment. The study of this limit led to landmark findings
like trains, fridges and general heat engines, and to theoretical results on the efficiency of such engines. One
might wonder whether these findings can also be applied at the quantum scale, and especially to very small
systems composed only of a couple of qubits [25, 104]. The answer is yes: not only is it possible to build
two-qubit heat engines, but they achieve Carnot efficiency [105, 106]. It is possible to build heat engines that
do not require a precise control of interactions, in other words, that do not require a clock [105, 107].
the states correspond to the same inverse temperature iments, and quantify things like the average work cost of
. This justifies the assumption that we may bring in transformations or conversion rates [86, 95]. The mono-
any number and shape of subsystems in thermal states tones found (like the von Neumann entropy and the usual
for free, because we could never extract work from them free energy) are familiar from traditional thermodynam-
aloneanother resource is necessary, precisely a state out ics, because this regime approximates the behaviour of
of equilibrium. More formally, it was shown that if a re- large uncorrelated systems. As we move towards a ther-
source theory allows only for energy-conserving unitaries modynamic theory of individual quantum systems, it be-
and composition with some choice of free states, Gibbs comes increasingly relevant to work in the single-shot
states are the only choice that does not trivialize the the- regime. Some studies consider exact state transforma-
ory [83, 111]. tions [77, 78, 114], while others allow for a small error
tolerance [79, 81, 82, 87, 88, 111, 115, 117, 118]. The
monotones recovered correspond to operational entropy
8. Different baths measures, like the smooth max-entropy (see Example 3),
and variations of a single-shot free energy that depend on
The results outlined above suggest that thermodynam- the conservation laws of the setting; in general, they can
ics can be treated as information processing under conser- be derived from quantum Renyi relative entropies [92] be-
vation laws, and so researchers began to experiment with tween the initial state and an equilibrium state [83, 119].
other conserved quantities, like angular momentum [112 Single-shot results converge asymptotically to the tradi-
114], using the principle of maximum entropy to model tional ones in the limit of many independent copies. The
thermal momentum baths. The state of those baths has relation between single-shot and average regimes is stud-
again an exponential Gibbs form, with operators like L ied via fluctuation theorems in [120].
replacing H. The same type of monotones emerged, and
similar behaviour was found for more general conserved
quantities [111, 115]. 11. Definitions of work
SB the initial state [116]. Another example, inheriting 1. Starting from the pre-order
more directly from classical thermodynamics, assumes
that we can change the Hamiltonian of S and bring it As mentioned before, the set of allowed transforma-
in contact with an implicit heat bath [121]; work at a tions imposes a pre-order structure (S, ) on the state
time t is then defined as space S. One direction towards exploring the concept
of resource theories could be to start precisely from
t
dHS (t0 )
Z
0 0 such a pre-order structure. That was the approach of
W (t) := dt Tr S (t ) . Caratheodory, then Giles and later Lieb and Yngvason,
0 dt0
who pioneered the idea of resource theories for thermody-
namics [77, 78, 126, 127]. In their work, the set of allowed
To study fluctuations around this average value, we con-
transformations is implicitly assumed, but we work di-
sider work to be a random variable in the single-shot
rectly with an abstract state space equipped with a pre-
settingthis is explored in Section V. Note that in these
order relation. They were largely inspired by classical,
examples work is not operationally motivated; rather it
macroscopic thermodynamics, as one may infer from the
is defined as the change of energy that heat cannot ac-
conditions imposed on the state space, but their results
count for. Resource theories of thermodynamics, with
can be applied to thermodynamics of small quantum sys-
their conservation laws, force us to consider an explicit
tems [114]. Assuming that there exist minimal resources
system W for work storage. We act globally on S W ,
that can be scaled arbitrarily and act as currency, the
and we can define work in terms of properties of the re-
authors obtain monotones for exact, single-shot state
duced state of W . One proposal for the quantum equiva-
transformations. When applied to the pre-order relation
lent of a weight that can be lifted, for the resource theory
on classical states that emerges from thermal operations,
of thermal operations, is a P harmonic oscillator, with a
these monotones become single-shot versions of the free
regular Hamiltonian HW = n n |ni hn|. The energy
energy [114].
gaps need to be sufficiently small to be compatible with
the Hamiltonian Rof S; in the limit 0 the Hamiltonian
becomes HW = dx x |xi hx| [87, 106]. Average work is
2. Starting from the set of free resources
defined as Tr(HW final
W ) Tr(HW W
initial
), and fluctua-
tions can be studied directly in the final state of the work
storage system, W . This approach also allows us to ob- In Ref. [119] general quantum resource theories are
serve other effects, such as the build up of coherences characterized based on the set of free resources of each
in W , and of correlations between W and S. Another theory. Assuming that the set of free states is well-
advantage is that we can adapt the storage system to behaved (for instance, that it is convex, and that the
other resource theories: for instance, we can have an an- composition of two free states is still a free state), they
gular momentum reservoir composed of many spins, and show that the relative entropy between a resource and
count work in terms of polarization of the reservoir [113]. the set of free states is a monotone. This is because
These approaches are critically analysed in Ref. [122]; in the relative entropy is contractive (non-increasing under
particular, it is highlighted that they do not distinguish quantum operations); the same result applies to any con-
work from heat. For instance, thermalizing the work stor- tractive metric. Finally, they find an expression for the
age system may result in an increase of average energy, asymptotic value of a resource in terms of this monotone:
which is indiscriminately labelled as average work. In the conversion rate between two resources is given by the
the same paper, an axiomatic approach to define work ratio between their asymptotic value.
is proposed, based on concepts from resource theories
and interactive proofs. There, work is seen as a figure of
3. In category theory
merit: a real function assigned to state transformations,
W( ). Starting from a couple of assumptions, the
authors derive properties of acceptable work functions Ref. [16], and more recently Ref. [37] have generalized
W: for instance, that they can be written as the dif- the framework of resource theories to objects known as
ference between a monotone for initial and final state, symmetric monoidal categories. These can represent es-
W( ) = g() g(). The free energy is an example sentially any resource that can be composed (in the sense
of such a valid work function. of combining copies of different resources, like tensoring
states in quantum theory). The authors consider both
physical states and processes as possible resources. After
obtaining the pre-order structure from a set of allowed
operations, resource theories can be classified according
B. Generalizing resource theories
to several parameters. For instance, the authors iden-
tify quantitative theories (where having more of a re-
Let us now abstract from particular resource theories, source helps, like for thermal operations) and qualitative
and think about their common features, and how we may ones (where it helps to have many different resources).
generalize them. They find expressions for asymptotic conversion rates in
16
How much energy is needed to perform logical operations? What are the ultimate limits for heat dissipation of
computers? These questions lie at the interface between thermodynamics and information theory, are of both
foundational and practical interest. As Bennett realized, all computations can be decomposed into reversible
operations followed by the erasure of a subsystem [123]. If we assume that the physical support of our computer
is degenerate in energy, we recover the setting of noisy operations, in which unitaries are applied for free. That
way, the thermodynamic cost of computation is simply the cost of erasure, which is defined as taking a system
from its initial state to a standard, predefined pure state |0i (like when we format a hard drive). Rolf
Landauer first proposed that the work cost of erasing a completely unknown bit of information (think of a
fully mixed qubit) in an environment of temperature T is kB T ln 2 [12]. That very same limit was also found
for quantum systems, in the setting of thermal operations [116, 124], for the ideal case of an infinitely large
heat bath and many operations; finite-size effects are analysed in Ref. [116].
Using Landauers principle as a building block, we can approach the more general question of erasing of a
system that is not in a completely unknown state, but rather about which we have partial information. For
example, imagine that we want to perform an algorithm in our quantum computer, and then erase a subsystem
S (which could be a register or ancilla). The rest of our computer may be correlated with S, and therefore
we can use it as a memory M , and use those correlations to optimize the erasure of S. In short, we want
to take the initial state SM to |0i h0|S M , erasing S but not disturbing M . It was shown [79, 82] that
the optimal work cost of that transformation is approximately Hmax (S|M ) kB T ln 2, where parametrizes our
error tolerance and Hmax (S|M ) is the smooth max entropy, a conditional entropy measure that measures our
uncertainty about the state of S, given access to the memory M . It converges to the von Neumann entropy
in the limit of many independent copies. In the special case where S and M are entangled, it may become
negativemeaning that we may gain work in erasure, at the cost of correlations. Not incidentally, these results
use quantum information processing techniques to compress the correlations between S and M before erasure;
after all, information is physical [125].
different regimes and, crucially, give varied examples of of Ref. [97]). This would give us a better under-
resource theories, within and beyond quantum theory, standing of the thermodynamic power and limita-
showing just how general this concept is. tions of coherent quantum states. It would also
allow us to account for all costs involved in state
transformations.
4. Resource theories of knowledge
IV. ENTANGLEMENT THEORY IN given entropy. Some of the first considerations in this di-
THERMODYNAMIC SETTINGS rection were motivated by the prospect of using nuclear
magnetic resonance (NMR) for quantum computation.
In the previous sections we have established how quan- Due to non-zero temperature, i.e. non-trivial restrictions
tum information can be used to understand the very on the entropy of the state, such systems would always
foundation of thermodynamics, from the emergence of be fairly close to the maximally mixed state.
thermal states to the resource theory of manipulating In this context the most natural question to ask, is
these with energy conserving unitaries. We have seen whether a unitary transformation is capable of entan-
that phrasing thermodynamics as a resource theory can gling a given input state. As a precursor to studying the
elucidate the meaning of thermodynamic quantities at possibility of entangling multipartite states, the complete
the quantum scale, and how techniques originally devel- solution for two qubits was found in Ref. [131] and later
oped for a resource theory of communication can facil- decent bounds on bipartite systems of arbitrary dimen-
itate this endeavour. The motivation behind this ap- sion were presented in Ref. [132].
proach is a very practical one: finding the ultimate lim- Another pathway was pursued by Refs. [133135],
itations of achievable transformations under restrictions where with NMR quantum computation in mind, vol-
that follow from the nature of the investigated system umes of separable states around the maximally mixed
that naturally limits the set of operations we can perform. state were identified. These volumes imply that if any
As quantum information processing is becoming increas- initial state is in close proximity of the maximally mixed
ingly applied, we also need to think about fundamen- state, there can be no chance of ever creating entangle-
tal restrictions to quantum information itself, emerging ment in such states, as the distance from the maximally
from unavoidable thermodynamic considerations. There mixed state is invariant under unitary transformations.
has thus been an increased interest in investigating sce- Further improvements in terms of limiting temperatures
narios of quantum information processing where thermo- were obtained in Ref. [136].
dynamic considerations cannot be ignored. From funda- The question of whether a given state can be entan-
mental limitations to the creation of QIP resources to gled under certain entropy restrictions clearly relies only
their inherent work cost. In this section we try to give a on the eigenvalue spectrum of the considered state, as
brief overview over recent developments in this intersec- the best conceivable operation creating entanglement is a
tion with a focus on the paradigmatic resource of QIP: unitary one (which leaves eigenvalues unchanged). These
entanglement. questions were further pursued under the name of sep-
arability from spectrum in Refs. [137139]. One of the
main results important in the context of quantum ther-
A. Correlations and entanglement under entropic modynamics is the following: A state with eigenvalues
restrictions i , ordered by size, i.e. {i i+1 } can be entangled by
an appropriate unitary if
Entanglement theory is in itself one of the most promi- p
nent examples of resource theories. Entanglement, a re- 1 > d1 + 2 d2 d . (10)
source behind almost all tasks in quantum information
processing, is hard to create and once distributed can More importantly, for 2 m dimensional states, this con-
only decrease. Thus in entanglement theory classically dition is not only sufficient, but also necessary [139].
correlated states come for free and local operations are Moving beyond the mere presence of entanglement in
considered cheap, which singles out entanglement as the the unitary orbit of input states, one encounters an in-
resource to overcome such limitations. These limitations trinsic difficulty of properly quantifying the entanglement
and resources are of course very different to the resources created. There is a whole zoo of entanglement mea-
and tasks explored in the previous sections. A compre- sures [143, 144] and only in the bipartite case there is a
hensive comparison between the principles behind these unique currency known, i.e. a paradigmatic resource
and more general resource theories is made in Ref. [129] state from which all other states can be created via LOCC
and as examples of a more abstract treatment in Ref. [16]. (although recent progress has been made in the four qubit
Such resource theories are always designed to reflect case, where it has been shown that after exclusion of a
specific physical settings, such as local operations and measure zero set, such a set of resource states can indeed
classical communication (LOCC) [130] as a natural con- be identified [145]).
straint for communication. It is therefore unavoid- In any case one can at least study general correla-
able that when describing various physical circumstances tions with a clear operational interpretation, such as
these resource theories can be combined yielding hybrid the mutual information, which has been performed in
theories. One natural example is the desire to process Refs. [140, 146, 147]. In these papers the authors have,
quantum information in a thermodynamic background. among other things, identified minimally and maximally
Ignoring limitations coming from available energies in a correlated states in the unitary orbit of bipartite systems.
first step this leads to the task of producing resources for It turns out that at least here the entropy poses only a
computation (such as entanglement or correlation) at a rather trivial restriction and for any d-dimensional state
18
Two systems, initially in local, product thermal states of temperature T , are correlated using a) any unitary
[140], b) a unitary changing the average energy by at most E [141] or c) a unitary on the systems and another
bath at the same temperature T [142]. The local states of the two systems may heat up in the process: T
is infinite in a), a function of E in b) and forced to be zero in c). In general, for setting b), the optimal
conversion rate of average energy into mutual information is still unknown.
a mutual information of I (A : B) = 2 log2 (d) S() thus can also reduce the entropy of the target system.
can be achieved via global unitary rotations. Disregarding energy costs in this context of course yields
Exploiting these results Ref. [141] continued to study the rather trivial result that any quantum information
the generation of correlations and entanglement under processing resource can be produced, simply by cooling
entropic restrictions for multipartite systems. Inspired the system (close) to the ground state and then perform-
by the idea of thermal states as a free resource, the au- ing the adequate global unitary operation on it. Taking
thors consider a multipartite system initially in a thermal into account the free energy costs of correlating trans-
state. They ask what is the highest temperature Tent at formations, Ref. [142] has shown that every bit of corre-
which entanglement can still be created, it scales with the lation embodies an intrinsic work value proportional to
dimension of the partitions and quantify the inherent cost the temperature of the system. For mutual information
in terms average energy change (see example (4) for an this yields a relation akin to Landauers principle for the
exemplary two qubit energy cost). By introducing con- work cost of creating correlations Wcor ,
crete protocols, i.e. unitary operations, the authors show
that bipartite entanglement generation across all parti- Wcor kB T I (A : B) , (11)
tions of n-qubits is possible iff kB T /E < n/(2 ln(1+ 2)) and it implies a general free energy cost of entanglement
and genuine multipartite entanglement across all parties that is bounded from above and below for the bipartite
can be created if kB T /E < n/(2 ln(n)) + O(n/ ln(n)2 ). case in Ref. [142]. All previous considerations are illus-
trated in Fig. 4.
That extractable work can be stored in correlations is
B. Correlations and entanglement in a by no means a purely quantum phenomenon. Even clas-
thermodynamic background sical correlations can store work in situations where local
work extraction is impossible. In Ref. [148] the quantum
In the context of thermodynamics the previous sub- vs. classical capacity for storing extractable work purely
section can be viewed as a very special case of operating in correlations was compared. For two qubits twice as
on closed systems with an unlimited external energy sup- much work can be stored in entangled correlations as the
ply or a fully degenerate Hamiltonian. As elaborated in best possible separable (or even classical, which turns
Section III of the review this does not encompass the out to be the same) correlations admit (a fact that is
whole potential of thermodynamic transformations. If also mentioned in Ref. [76] in a different setting). How-
the necessary correlating unitary does not conserve the ever the difference between separably encoded work from
total energy, we should account for the difference in av- correlations Wsep and the maximally possible work in
erage energy between initial and final states. Taking into correlations Wmax scales as
account also the average energy cost reveals an intrinsic Wsep
work value of correlations and entanglement in general. = 1 O(n1 ) , (12)
This fundamental fact was first quantified in Ref. [141]). Wmax
Accounting for the average energy change in the uni- i.e. the quantum advantage vanishes in the thermody-
tary orbit of initial quantum states however still does namic limit of large systems.
not encompass the whole potential of thermodynamic re- Concerning the extractable work from correlations one
source theories. Thermal operations on target states can can also find seemingly contrary results if the figure of
also make use of a thermal bath at temperature T and merit changes. The above considerations apply only if
19
Creating entanglement from thermal states will always cost some energy. For the simplest case of entangling
two qubits with energy gap E at zero temperature one can find a closed expression, e.g. for the concurrence,
in terms of the invested average energy E = W :
s
W W
C(W ) = 2 ,
E E
the target is an extraction of average energy or standard sources in a thermodynamic background. The other ob-
free energy, partially neglecting the details of the work vious connection between the resource theories of entan-
distribution arising in the receiving system (detailed con- glement and thermodynamics is taking the converse ap-
siderations of such work distribution fluctuations will be proach. Here one is interested in thermodynamic opera-
discussed in section (V)). One can just as well be inter- tions under additional locality restrictions.
ested in a guaranteed amount of work. If that is the In Ref. [154] the difference between the extractable
case one can arrive at more restrictions concerning work work from bipartite quantum states in thermodynamics
extraction as also recently demonstrated in [118]. Cu- both with and without locality restrictions was studied.
riously in Ref. [149] it was shown, however, that these The resulting difference, called the work deficit, can be
restrictions can be overcome by considering k initially bounded via
uncorrelated catalysts that build up correlations in the
process. In that context one can extract more determinis- = max[S(A ), S(B )] S(AB ) , (14)
tic work and can thus regard the stochastic independence which for pure states coincides with entanglement of for-
of the input catalysts as a resource for work extraction, mation (or any other sensible choice of entanglement
which is quite contrary to the case considered before and measure that all reduce to the marginal entropy in case
the thermodynamic limit. of pure states). In the above equation it is assumed that
A different, but very related, setting exploring work bits which are sent down the communication channel are
gain from correlations is studied in the context of quan- treated as classical in the sense that they are only de-
tum feedback control. Here the task is rather to quantify phased once, and not again in a second basis. This inter-
the inevitable work cost arising from information gain in play led to subsequent investigations into the thermody-
the process of a measurement. As in order to measure a namic nature of entanglement in Ref. [154], where analo-
system one needs to correlate with the system in question gies between irreversible operations in thermodynamics
it follows intuitively that this scenario will also always in- and bound entanglement were drawn, and to concrete
duce work cost related to bipartite correlations between physical scenarios satisfying this bound in Ref. [121].
the system and the memory storing the information gain
about the system. Here the work cost coming purely from
correlations was quantified in Ref. [147], building upon D. Entanglement resources in thermodynamic
older results on the inevitable work cost of quantum mea- tasks
surements [150153] and Landauers principle. To model
the necessary feedback control, the authors included a
Apart from resource theory inspired questions, one
general model of a quantum memory upon which projec-
might study the role of informational quantities through
tive measurements can be performed. The authors also
their inevitable appearance in thermodynamic operations
studied the possible work gain from bipartite quantum
at the quantum level. For instance the role of entan-
states in this context. Denoting the state of the memory
gling operations and entanglement generation in extract-
as M the authors find an upper bound on the work gain
ing work from multiple copies of passive states , i.e. states
(defined as the work extracted from both subsystems mi-
where no local work extraction is possible [108, 109], has
nus the work cost of the measurements and subsequent
attracted some attention recently. The implied fact that
erasure of the quantum memory) as
global unitary operations are required to extract work
indicates some form of non-local resource being involved
Wnet kB T (I (A : B) I (A : B|M )) F () .
in the process.
(13)
In general passive states are always diagonal in the en-
ergy eigenbasis [108, 109], which implies that one starts
and ends the protocols with diagonal states. In these
C. Thermodynamics under locality restrictions scenarios the individual batteries from which work is to
be extracted are considered non-interacting, directly im-
In the previous subsections we have reviewed the plying the separability of initial and final states in these
prospect of creating quantum information processing re- protocols. Nonetheless the fact that local unitaries can
20
never extract any work from copies of passive states di- lined in this section emphasise the fact that thermody-
rectly implies that entangling unitaries enable work ex- namic constraints have drastic consequences for process-
traction from such states [155]. In that sense entangling ing quantum information and that locality constraints
power of unitaries can be seen as a resource for work ex- will change thermodynamic considerations at the quan-
traction purposes (which in conventional thermodynamic tum scale. One path to explore could now be a consis-
resource is of course considered a free operation). tent resource theory that adaptively quantifies possible
In Ref. [156] the role of quantum resources in this con- resources from different restrictions. This would further-
text was further explored. While it is true that the abil- more elucidate the exact role played by genuine quantum
ity to perform entangling unitaries is required for this effects, such as entanglement, in thermodynamics.
particular work extraction problem, this does not imply
that any entanglement is ever generated in the process.
In fact the whole procedure can dynamically be imple-
mented without ever generating the slightest bit of en- V. QUANTUM FLUCTUATION RELATIONS
tanglement [156], however the most direct transformation AND QUANTUM INFORMATION
can considerably entangle the systems in the process. In
Ref. [157] it was demonstrated that if the work per unit
A. Introduction
time (power) is considered with cyclic operations in mind
then a quantum advantage for charging power can be
achieved. The phenomenological theory of thermodynamics suc-
cessfully describes the equilibrium properties of macro-
scopic systems ranging from refrigerators to black holes
E. Using thermodynamics to reveal quantumness that is the domain of the large and many. By extrapo-
lating backwards, from the domain of the many to the
That entanglement plays a special role in quantum few, we venture further from equilibrium into a regime
many-body physics is a well established fact that has re- where both thermal and quantum fluctuations begin to
ceived adequate attention in numerous publications (see dominate and correlations proliferate. One may then ask
e.g. Ref. [158] and the extensive list of citations therein). the question - what is an appropriate way to describe this
In this topical review we want to at least mention a re- blurry world which is dominated by deviations from the
lated question that connects quantum thermodynamics average behaviour?
directly with entanglement theory: The possibility to One way to describe the thermodynamics of small sys-
use thermodynamic observables to reveal an underlying tems where fluctuations cannot be ignored is by using
entanglement present in the system. At zero tempera- the framework of stochastic thermodynamics [170]. In
ture it is already known that many natural interaction this approach the basic objects of traditional statistical
Hamiltonians have entangled ground states (in fact often mechanics such as work and heat are treated as stochastic
many low energy eigenstates even of local Hamiltonians random variables and hence characterised by probability
feature entanglement). This fact can be exploited to di- distributions. Over the last 20 years various approaches
rectly use the energy of a system as an entanglement have lead to sets of theorems and laws, beyond the lin-
witness, even at non-zero temperatures [159]. Intuitively ear response regime, which have revitalised the already
this can be understood through the fact that a low av- mature study of non-equilibrium statistical mechanics.
erage energy directly implies that the density matrix is Central to these efforts are the fluctuation relations that
close to the entangled ground state. If this distance is connect the non-equilibrium response of a system to its
sufficiently small that can directly imply entanglement equilibrium properties. A wealth of results have been
of the density matrix itself. The known results and open uncovered in both the classical and the quantum regimes
questions of this interplay including Refs. [160167] are and the interested reader is directed to the excellent re-
also discussed in the review Ref. [36]. Furthermore, other views on the topics [2123]. Here we focus on aspects of
macroscopic thermodynamic quantities can also serve as this approach that have been specifically influenced by
entanglement witnesses through a similar intuition, such concepts in quantum information, or show promise for
as e.g. the magnetic susceptibility [168] or the entropy symbiosis. We hope that by reviewing the existing con-
[169]. tributions as well as suggesting possible research avenues,
further cross fertilisation of the fields will occur.
To begin with, it is useful to illustrate how the proba-
F. Outlook bility distributions of a thermodynamic variable like work
is defined. Consider a quantum system with a time-
Resource theories always have their foundation in what dependent Hamiltonian H((t)), parametrized by the ex-
we believe to be hard/impossible to implement and what ternally controlled work parameter (t). The system is
resources allow us to overcome such limitations. As prepared in a thermal state by allowing it to equilibrate
such they always only capture one specific aspect of the with a heat bath at inverse temperature for a fixed
physical systems under investigation. The results out- value of the work parameter (t < ti ) = i . The initial
21
state of the system is therefore the Gibbs state, In order to understand what is meant by a fluctua-
tion theorem, we introduce a backward process which
eH(i ) is the time reversed protocol of the forward one previ-
(, ) :=
Z (i ) ously defined. Now PB (W ) is the work distribution cor-
responding to the backward process, in which the system
At t = ti the system-reservoir coupling is removed and is prepared in the Gibbs state of the final Hamiltonian
a fixed, reversible protocol is performed on the system H(f ) at t = 0 and subjected to the time-reversed proto-
taking the work parameter from its initial value i to col that generates the evolution U (tf , ti ) , where is
the final value f at a later time t = tf . The initial and the anti-unitary time-reversal operator. It turns out that
final Hamiltonians are defined by their spectral decom- the following theorem holds, the Tasaki-Crooks relation
positions [172, 173],
P
H(i ) = n En (i ) |n i hn | PF (W )
P = e(W F ) , (18)
and H(f ) = m Em (f ) |m i hm | PB (W )
respectively, where |n i (|m i) is the nth (mth) eigen- which shows that, for any closed quantum system under-
state of the initial (final) Hamiltonian with eigenvalue going an arbitrary non-equilibrium transformation, the
En (i ),Em (f ). The protocol connecting the initial and fluctuations in work are related to the equilibrium free
final Hamiltonians generates the unitary evolution oper- energy difference for the corresponding isothermal pro-
ator U (tf , ti ), which in general has the form cess between the equilibrium states (i ) and (f ),
Z t 1 Z (i )
0 0 F = ln . (19)
U (t, ti ) = T exp i dt H((t )) , (15) Z (f )
ti
This relationship is further emphasized by a corollary
where T denotes the time-ordering operation. We to Eq. 18 known as the Jarzynski equality [174],
stress here that, in this framework, one typically assumes Z
that the system is initially prepared in a thermal state dW PF (W )eW = heW i = eF (20)
but after the unitary protocol the system is generally in
a non-equilibrium state. which states that F (of the corresponding isother-
The work performed (or extracted) on (or from) the mal process) can be extracted from by measuring the
system as a consequence of the protocol is defined by the exponentiated work. A straightforward application of
outcomes of two projective energy measurements [171]. Jensens inequality for convex functions allows one the
The first, at t = ti , projects onto the eigenbasis of the ini- retrieve the expected expression hW i F . The aver-
tial Hamiltonian H(i ), with the system in thermal equi- age energetic deviation of a non-equilibrium process from
librium. The system then evolves under the unitary op- the equivalent reversible isothermal process is known as
erator U (tf , ti ) before a second projective measurement is dissipated work
made onto the eigenbasis of the final Hamiltonian H(f ) hW idiss = hW i F. (21)
at t = tf . The joint probability of obtaining the outcome
En (i ) for the initial measurement followed by Em (f ) Due to the Jarzynski equality this quantity is positive,
for the final one is easily shown to be hW idiss 0. This can be also directly seen from the
Crooks relation, taking the logarithm of both sides of
eEn (i ) the equality in Eq. 18 and integrating over the forward
p(n, m) = | hm | U (tf , ti ) |n i |2 . (16) distribution we find
Z(i )
hi = (hW i F ) = K(PF (W )||PB (W )) (22)
Accordingly, the quantum work distribution is defined as
X where K is the classical Kullback Leibler divergence
PF (W ) = p(n, m) (W [ Em (f ) En (i ) ] ) . and we have introduced the average irreversible entropy
n,m change hi corresponding to the dissipated work. Phys-
(17) ically the irreversible entropy change, in this context,
would be the internal entropy generated due to the non-
where is the Dirac delta function. For reasons which equilibrium process which would manifest itself as an ad-
will become clear shortly we use the subscript F to de- ditional source of heat if an ideal thermal bath would
note forward protocol. Physically, Eq. 17 states that be reconnected to the system at the end of the protocol.
the work distribution consists of the discrete number of In Ref. [175] it was shown that the irreversible entropy
allowed values for the work (Em (f ) En (i )) weighted change can also be expressed in terms of a quantum rel-
by the probability p(n, m) of measuring that value in a ative entropy
given realisation of the experiment. The quantum work
distribution therefore encodes fluctuations in the mea- hi = D(|| (f , )) (23)
sured work arising from thermal statistics (first measure- where = U (tf , ti ) (i , )U (tf , ti ) is the out of equi-
ment) and from quantum measurement statistics (second librium state at the end of the protocol. This is fully
measurement). consistent with the open system treatment in [176].
22
B. Phase estimation schemes for extraction of sampled with just a single measurement in time. In this
quantum work and heat statistics work it was suggested that the algorithm proposed could
be used, in combination with the fluctuation theorems,
Surprisingly, perhaps one of the most important contri- to estimate the free energy of quantum states on a quan-
butions that ideas from quantum information have made tum computer. The scheme was recently extended and
to this field in statistical mechanics is the experimental developed in Ref. [189] along with a promising implemen-
acquisition of statistics of work. In the classical setting tation using ultra-cold atoms. This would be a promising
considerable progress has been made in the experimen- avenue to explore work statistics in a many-body physics
tal extraction of the relevant stochastic thermodynamic setting where the statistics of work can be shown to have
distributions to explore and verify the fluctuation theo- universal behaviour at critical points [190].
rems [23]. Up until very recently, no such experimental
progress had been made for quantum systems. A central
issue is the problem of building the quantum work distri- C. Fluctuation relations with feedback,
measurement and CPTP maps
bution as it requires to make reliable projective energy
measurements on to the instantaneous energy eigenbasis
of an evolving quantum system [22, 171]. It was proposed The relationship between thermodynamics and the in-
in Ref. [177] that these measurements could be reliably formation processing is almost as old as thermodynamics
performed on a single trapped ion, an experiment that itself and is no where more dramatically manifested than
was recently performed [178]. by Maxwells demon [26]. One way of understanding the
Alternatives to the projective method have been pro- demon paradox is by viewing the demon as performing
posed [179, 180], based on phase estimation schemes, well feedback control on the thermodynamic system. In this
known in quantum information and quantum optics [181]. case the framework for stochastic thermodynamics and
In these schemes, we couple our system to an ancillary the fluctuation theorems needs to be expanded. Build-
system, and perform tomography on that system. The ing upon previous work [152, 153], Sagawa and Ueda
spirit is very similar to the DQC1 algorithm put forward have generalised the Jarzynski equality to incorporate
in Ref. [182]. The characteristic function of the work the feedback mechanism [191, 192] for classical systems.
probability distribution (Eq. 24) can be obtained from This theoretical breakthrough allowed for an experimen-
the ancilla, and the work statistics are then extracted by tal demonstration of information to energy conversion in
Fourier transform. The characteristic function is defined a system by means of of non-equilibrium feedback of a
as Brownian particle [193]. These feedback based fluctua-
Z tion theorems were further modified to incorporate both
F (u) = dW eiuW PF (W ). (24) initial and final correlations [194]. These works, in par-
ticular, highlight the pivotal role played by mutual infor-
mation in non-equilibrium thermodynamics [6].
The proposals to measure the characteristic function were
The Sagawa-Ueda relations were generalized to quan-
first tested in the laboratory only quite recently in a Liq-
tum systems in Ref. [195]. For reasons of pedagogy we
uid state NMR setup [183]. This experiment is the first
will follow this approach here. In the work of Morikuni
demonstration of the work fluctuation theorems and ex-
and Tasaki an isolated quantum system is considered
traction of work quantum statistics, and is expected to
where an external agent has control of the Hamiltonian
inspire a new generation of experiments at the quantum
parameters. The system is initialised in a canonical state,
level. Another interesting extension of these schemes is to
(), and an initial projective measurement of the energy
go beyond the closed system paradigm and to study open
is made whose outcome is Ei0 . The Hamiltonian is then
system dynamics at and beyond the weak coupling limit.
changed via a fixed protocol and evolves according to the
The first extensions have been proposed in Refs. [184
unitary operator U . In the next stage a projective mea-
186]. In Ref. [187] the proposal outlined in Ref. [186]
surement is performed with outcomes j = 1, . . . , n and
to measure the statistics of dissipated heat was imple-
described by a set of projection operators 1 , . . . , n .
mented in order to perform a study of the information
Now the time evolution is conditioned on the outcome
to energy conversion in basic quantum logic gates at the
j so the Hamiltonian is changed according to these out-
fundamental Landauer Limit.
comes. This is the feedback control stage. Finally, one
Another interesting suggestion made to access the
makes a projective measurement of the energy of the fi-
quantum work statistics is the use the concept of a pos-
nal Hamiltonian with outcome Ekj . In this setting it is
itive operator valued measure, or POVM [188], a well-
shown that
known concept within quantum information and quan-
tum optics. A POVM is the most general way to describe he(W F ) i = , (25)
a measurement in a quantum system, with the advantage
that it can always be seen as a projective measurement on where W = Wi,j,k = Ei0 Ekj is the work and F is the
an enlarged system. In this work the authors show that free energy difference between the initial state and the
by introducing an appropriate ancilla that the POVM canonical state corresponding to the final value of the
description allows the work distribution to be efficiently Hamiltonian H j . We see that in this feedback controlled
23
scenario a new term enters on the right hand side. A tems, the relative entropy plays a central role [192]. As
straightforward calculation shows that this term evalu- first pointed out in Ref. [209] this is due to its close re-
ates as = j Tr[j Uj ()Uj j ]. This quantity is lationship with the free energy of a quantum state. The
P
shown in Refs. [191, 195] to be related to the efficiency of relative entropy also plays a central role in quantum in-
the demon in making use of the information it acquires formation theory, in particular, in the geometric picture
during the feedback process. When it becomes less than of entanglement and general quantum and classical corre-
one it provides an example of a failed demon who did lations [210, 211]. In the non-equilibrium formulation of
not make a good use of the information acquired. On thermodynamics [22] it is omnipresent for the description
the other hand it can become larger than one indicating of irreversible entropy production in both closed [175]
that the feedback is working efficiently. Another rela- and open driven quantum systems [212] (see also [213]).
tion discovered by Sagawa and Ueda and quantized by One may then wonder if there exists a relationship be-
Morkikuni and Taskaki concerns almost the same proto- tween the entropy produced by operations that generate
col as just explained only now classical errors are made in or delete correlations in a quantum state and the mea-
the intermediate measurement stage. Again let the inter- sures for correlations in that state? Given the youthful
mediate measurement be described by 1 , ..., n which nature of the field the question is largely unanswered but
yield the result j but the controller misinterprets the re- some progress in this direction has been made.
sult as j 0 with a certain probability. In this framework The relationship between the relative entropy of en-
another generalised fluctuation theorem can be derived, tanglement and the dissipated work was first proposed
as an entanglement witness in Ref. [214]. Going beyond
he(W F I) i = 1 (26) the geometric approach a functional relationship between
the entanglement generated in a chain of oscillators and
where I is the mutual information between the set of
the work dissipated was explored in Ref. [215] and also
measurement outcomes the demon actually records and
later for more general quantum correlations [216]. In
what is the true result of the projection. These feedback
an open systems framework it was shown that the ir-
fluctuation theorems for quantum systems were further
reversible entropy production maybe attributed to the
generalised to the situation when a memory system is
total correlations between the system and the reservoir
explicitly accounted for in Ref. [147] and shed light on
[217] (we note that this derivation is entirely analogous
the amount of thermodynamic work which can be gained
to the formulation of the Landauer principle put forward
from entanglement. In addition to feedback, fluctuation
by in Ref. [116]). The exchange fluctuation relation and
theorems were investigated under continuous monitoring
the consequences for correlated quantum systems were
[196, 197] and analysed for general measurements [198,
studied in Ref. [218].
199].
A recent series of papers have analysed fluctuation-
like relations from the operational viewpoint employing
the full machinery of trace-preserving completely posi- E. Outlook
tive maps. In Ref. [200] the formalism is used to give
an alternative derivation of the Holevo bound [201]. In As fluctuation theorems are exact results, valid for ar-
Ref. [202] an information-theoretical Jarzynski equality bitrary non-equilibrium dynamics, they are currently be-
was derived. It was found that fluctuation relations can ing used to understand the non-linear transport of en-
be derived if the map generated by the open dynamics ergy, heat and even information in quantum technologies.
obeys the unital condition. This has been connected to This is a relatively new research avenue and the appli-
the breakdown of micro-reversibility for non-unital quan- cations of quantum fluctuation theorems in other fields
tum channels [203206]. In Ref. [207] the authors anal- such as condensed matter physics, quantum optics and
ysed the statistics of heat dissipated in a general proto- quantum information theory are in their infancy. Ulti-
col and found that the approach can be used to derive a mately, the hope would be that they provide a unifying
lower bound on the heat dissipated for non-unital chan- framework to understand the relationship between infor-
nels. Recently this bound has been used to investigate mation and energy in non-equilibrium quantum systems.
the connection with the build up of multipartite correla- Ultimately one would like to form a picture of informa-
tions in collisional models [208]. tion thermodynamics of quantum systems under general
non-equilibrium conditions.
As we have seen above, quantum phase estima-
D. Entropy production, relative entropy and tion, a central protocol in quantum information theory,
correlations has been applied successfully to extract work statistics
from a small non-equilibrium quantum system and per-
With the surge of interest in the thermodynamics of haps other such unexpected interdisciplinary links will
quantum systems and the development of quantum fluc- emerge. For example one wonders if existing experimen-
tuation relations, research has been directed to micro- tal schemes could be modified to deal with situations
scopic expressions for entropy production. In formulat- dealing with non-passive initial states so as to study max-
ing thermodynamics for non-equilibrium quantum sys- imal work extraction problems and also to extend to more
24
complicated many-body and open system scenarios. the role of coherence in the transient behaviour when the
In Refs. [120, 219, 220] the first steps towards unifica- refrigerator is first switched on. We review a recent pro-
tion of the work statistics and fluctuation theorems ap- posal for a witness that quantum machines are provably
proach to thermodynamics and the single shot statistical outperforming their classical counterparts. Finally, we
mechanics approaches mentioned have been taken (see look at the idea of using thermal machines as a means of
Sec. III). We are confident that other links will emerge entanglement generation (switching the focus away from
between various approaches in the not so distant future. the traditions resources of work or heat currents).
A related idea is that of algorithmic cooling, which
we summarise in Example 5, and which was recently re-
VI. QUANTUM THERMAL MACHINES viewed in [222].
Consider a collection of n qubits, all at inverse temperature , with corresponding populations in the ground
and excited states p and (1 p) respectively. The goal of algorithmic cooling is to bring m qubits to the ground
state by an arbitrary unitary transformation. A fundamental upper bound can be placed on m, purely by
entropic considerations. The initial entropy is nS( () = nH(p), where H(p) = p log2 p (1 p) log2 (1 p)
is the binary Shannon entropy. Since unitary transformations do not change the entropy, this easily leads to
the upper bound on m,
which would be achieved if the remaining n m qubits are all left at infinite temperature (maximally mixed
state) with entropy S( (0)) = 1. In [226] it was shown that as n tends to infinity this fundamental limit can
be approached using an algorithm which uses O(n log2 n) unitary gate operations. It was later realised that
given access to an external bath this limit can be surpassed: the qubits which end this protocol at infinite
temperature can be refreshed to temperature and the protocol can be run again on the remaining (n m)
qubits, for example [227]. This is referred to as heat-bath algorithmic cooling.
In order to understand the basic principle, one can focus instead on 3 qubits and assume that the first is the
one which is to be cooled down (now not to zero temperature, but any colder temperature). Let us consider
the populations of the two states |100i and |011i, which are p2 (1p) and p(1p)2 respectively. The state|100i,
in which qubit one is excited (and therefore hot) has more population than the state |011i, where qubit one
is in the ground state (and therefore cold). Thus, by swapping the population of these two states the first
qubit is cooled down. Indeed, after the application of such a unitary, the final population p0 in the ground
state of the first qubit is
which is greater than p whenever (2p 1) > 0, i.e. whenever the first qubit was at a positive temperature.
Finally, a unitary which implements |011i |100i whilst leaving all other energy eigenstates the same can
easily be constructed from the CNOT and Toffoli gates as
A recent review giving many more details about algorithmic cooling can be found in Ref. [222]
Schematic diagram of a three qubit autonomous refrigerator (inside circle), coupled to three thermal reservoirs.
The interaction Hamiltonian is represented by the green and orange arrows.
if the couplings are chosen appropriately, (in particular thermal state (of a given mode, i.e. a harmonic oscil-
such that the weakest coupling is to the hot reservoir), lator) is Usq ()Usq . Whereas normally the variances of
then the system can quickly remains for a long time in the quadratures (x = (a + a )/2 and p = (a a )/2i)
a temperature below the stationary temperature, in par- are symmetric, the squeezed modes become asymmet-
ticular without oscillating above it. This demonstrates a ric, with one the former amplified by the factor er , and
particular stable regime for the preparation of the system the latter shrunk by er . The important point is that a
at temperatures below its stationary temperature. system placed in thermal contact with such a squeezed
In order to explore more the advantage offered by co- reservoir will not thermalize towards a thermal state at
herence, Ref. [230] also considered varying the initial , but rather to a squeezed thermal state, which has the
state, by altering the coherence in the subspace where the same average number of photons as a thermal state at
Hamiltonian operates. Interestingly, with only a small temperature (r) < . That is, in terms of average num-
amount of initial coherence even when considering case ber of photons, a squeezed thermal state appears hotter
(a) of weak-interaction dynamics, oscillations in the tem- than a thermal bath.
perature are seen, again allowing for cooling below the Starting first with Ref. [233], a model of an absorption
stationary temperature. In the other two cases, the mag- refrigerators is considered, identical to the one outlined
nitude of the oscillations is also seen to increase (i.e. the in the previous section. Here, in accordance with the
system achieves lower temperatures transiently), demon- above, in the weak coupling regime the effect of the reser-
strating an advantage in all situations. voir engineering amounts to modifying the Linbladian L,
Finally, in Ref. [231] the amount of entanglement that such that the term corresponding to the hot reservoir
is generated in the transient regime was also studied. Fo- LH transforms to LH (r), where this now generates dis-
cusing on either genuine multipartite entanglement, or sipation towards the squeezed thermal state at H (r) .
entanglement across the partition R|CH, i.e. the one They show that maximal COP that the refrigerator can
corresponding to energy-in vs. energy out (as studied approach becomes
in Ref. [224]), considerably more entanglement can be
generated in the transient regime. R H (r) R H
(r) = > c = . (29)
C R C R
That is, the COP overcomes the Carnot limit that
B. Reservoir engineering
bounds the COP of any absorption refrigerator operating
between baths at C , R and H , if reservoir engineer-
As we have seen in previous sections of the review, ing is not carried out. Thus if reservoir engineering is
thermals states are naturally considered as a free resource more readily available than a hotter hot bath, then this
which can be utilised and manipulated. Likewise, the approach clearly provides an advantage in terms of COP.
ubiquity of thermal machines is that having access to In Ref. [232] a different model was considered, this time
two large thermal reservoirs can also be considered as a quantum heat engine operating a quantum Otto cycle,
something essentially free, and thermal machines con- a time dependent cycle, comprising two expansion stages
sider ways of utilising these resources. (changing the Hamiltonian of the system) and two ther-
One interesting avenue is to consider that any trans- malization stages. This system considered comprised of a
formation of a thermal reservoir which can easily be single harmonic oscillator, with initial spacing E1 . While
carried out can also be considered to be free, as an ideal- uncoupled to any environment, the first stage is an ex-
isation, and this motivates the idea of considering ther- pansion, whereby E1 E2 > E1 , i.e. the Hamiltonian is
mal machines which run between engineered reservoirs, changed in time. In the second stage the system is then
assuming that the engineering was an easy to perform placed in contact with a squeezed hot reservoir (this is the
transformation. In the present context, when one has stage which differs from a standard Otto cycle, where an
sufficient control over (part of) the reservoir, then the unsqueezed hot reservoir is used). After disconnection,
engineering can be at the quantum level. Here again we the third stage is a compression stage, bringing the spac-
are interested specifically in the role that quantum cor- ing back to from E2 to E1 . Finally, the system is placed
relations engineered in the bath have on the functioning in contact with a cold (unsqueezed) reservoir, in order to
of quantum thermal machines. thermalize at the cold temperature. This cycle is sum-
In Refs. [232, 233] reservoir engineering in the form of marised in Fig. 6. The authors perform an analysis of
squeezing is considered, since squeezing is relatively easy the system and similarly show that the maximum effi-
to carry out, and is furthermore known to offer quan- ciency of the engine exceeds the Carnot efficiency (of the
tum advantages in other contexts in quantum informa- Otto cycle, = H /C ). Moreover, if one considers the
tion. That is the reservoir, instead of consisting of a large efficiency at maximum power, then this can also be sur-
collection of modes in thermal states at inverse tempera- passed, and as the squeezing parameter becomes large,
ture H , are in fact squeezed thermal states (at the same the efficiency at maximum power approaches unity.
temperature). More precisely, the squeezing operator is Finally, we stress that these results do not constitute
2
Usq = exp((ra2 r a )/2) with a and a the annihilation a violation of the second law, since they consider a sce-
and creation operators respectively, and the squeezed nario outside the regime of applicability of the Carnot
28
Schematic diagram of a quantum Otto engine, depicting only a single pair of levels, which can represent either
a qubit, or the spacing of a harmonic oscillator. The four stages are (I) expansion from E1 to E2 . (II) contact
with a hot thermal reservoir. (III) compression from E2 to E1 . (IV) contact with a cold thermal reservoir.
limit (much in the same way that a regular car engine, (i.e. as a rate equation among the populations in the
consuming fuel, does not violate the second law, since it energy eigenbasis); (ii) that the energy level structure
is also outside the regime of applicability). Conversely, it and coupling strengths are unaltered compared to quan-
is interesting that the net effect of squeezing appears to tum model under comparison; (iii) that no new sources of
be as if the hot reservoir has been heated to a tempera- heat or work are introduced. A way to satisfy the above
ture H (r), and that the performance of the machines is three constraints is to add pure de-phasing noise in the
bounded exactly by the Carnot limit with respect to this energy eigenbasis on top of the dissipative dynamics of
new temperature. the quantum model (arising from the interaction with
the thermal reservoirs). One can then compare models
with and without de-phasing noise, and ask whether the
additional noise places an upper bound on the power of
C. Quantum thermodynamic signatures
the machine.
One way to differentiate between a system which is For simplicity in presentation, in what follows we will
genuinely using quantum effects and one which is only focus here on the results obtained for the four-stage qubit
using the formal structure of quantum mechanics (the Otto heat engine, similar to the one described in the pre-
discreteness of energy levels, for example) is to devise vious subsection (except now with a qubit in place of a
signatures, or witnesses, for quantum behaviour. This is harmonic oscillator). We note that the authors show that
similar to what is done in entanglement theory, or in Bell the same results hold for a two-stage engine [235] and for
nonlocality, where one finds witnesses which certify that continuous time engines [25], as well as for refrigerators
entanglement was present, since no separable quantum and heat pumps. As an aside, the reason why the re-
state could pass a certain test. An interesting question is sult holds for all three models is because Ref. [234] also
whether one can find analogous witnesses in a quantum proves that in the regime of weak-coupling to the bath,
thermodynamics setting. This is what was proposed in and weak driving, all three types of engine can be shown
Ref. [234] in the form of Quantum thermo signatures. to be formally equivalent, producing the same transient
and steady state behaviour at the level of individual cy-
In more detail, the main idea of Ref. [234] is to find a
cles.
threshold on the power of a thermal machine which would
be impossible to achieve for a machine which is classi- It is shown that a state independent bound can be
cal. The authors take as the minimal set of requirements placed on the power of a classical machine which is pro-
for a machine to be considered classical (i) that its oper- portional to the duration of a single cycle of the engine
ation can be fully described using population dynamics cyc , as long as the so-called engine-action s is small,
29
ing information as a central concept, just as Maxwell did and EPSRC grant DIQIP. MH acknowledges funding
when his demon was born, will lead to a deeper under- from the Juan de la Cierva fellowship (JCI 2012-14155),
standing of many active areas of physics research beyond the European Commission (STREP RAQUEL) and the
quantum thermodynamics. Spanish MINECO Project No. FIS2013-40627-P, the
Generalitat de Catalunya CIRIT Project No. 2014 SGR
966. MH furthermore acknowledges funding through the
AUTHOR CONTRIBUTIONS AMBIZIONE grant PZ00P2 161351 from the Swiss Na-
tional Science Foundation (SNF). PS Acknowledges sup-
All authors contributed equally to this review. Sec- port from the European Union (Projects FP7-PEOPLE-
tions I and III were adapted from LdRs PhD thesis [128]. 2010-COFUND No. 267229, ERC CoG QITBOX and
ERC AdG NLST). AR thanks support from the Beat-
riu de Pinos fellowship (BP-DGR 2013), the EU (SIQS),
ACKNOWLEDGEMENTS the Spanish Ministry Project FOQUS (FIS2013-46768-
P), the Generalitat de Catalunya (SGR 874 and 875)
We thank Fernando Brandao, Aharon Brodutch, Nico- and the Spanish MINECO (Severo Ochoa grant SEV-
lai Friis, Marti Perarnau-Llobet, Joe Renes, Raam Uzdin 2015-0522). All authors acknowledge the COST Action
and Nicole Yunger-Halpern for helpful feedback on the MP1209.
manuscript. LdR thanks support from ERC AdG NLST
[1] C. H. Bennett, Int. J. Theor. Phys. 21, 905 (1982). [20] C. Jarzynski, Annu. Rev. Cond. Mat. Phys. 2, 329
[2] H. S. Leff and A. F. Rex, Maxwells demon: Entropy, (2011).
information, computing (Taylor and Francis, 1990). [21] M. Esposito, Rev. Mod. Phys. 81, 1665 (2009).
[3] M. B. Plenio and V. Vitelli, Contemp. Phys. 42, 25 [22] M. Campisi, P. Hanggi, and P. Talkner, Rev. Mod.
(2001). Phys. 83, 771 (2011).
[4] H. S. Leff and A. F. Rex, Maxwells demon 2: Entropy, [23] U. Seifert, Rep. Prog. Phys. 75, 126001 (2012).
classical and quantum information, computing (Taylor [24] P. Hanggi and P. Talkner, Nature Phys. 11, 108 (2015).
and Francis, 2002). [25] H. Scovil and E. Schulz-DuBois, Phys. Rev. Lett. 2, 262
[5] K. Maruyama, F. Nori, and V. Vedral, Rev. Mod. Phys. (1959).
81, 1 (2009). [26] J. E. Geusic, E. O. Schulz-DuBios, and H. E. D. Scovil,
[6] J. M. R. Parrondo, J. M. Horowitz, and T. Sagawa, Phys. Rev. 156, 343 (1967).
Nature Phys. 11, 131 (2015). [27] R. Alicki, J. Phys. A: Math. Gen. 12, L103 (1979).
[7] Bell Syst. Tech. J. 27, 379 (1948). [28] C. Gogolin and J. Eisert, Rep. Prog. Phys. 79, 056001
[8] T. M. Cover and J. A. Thomas, Elements of information (2016).
theory (John Wiley and sons, 2006). [29] A. Polkovnikov, K. Sengupta, A. Silva, and M. Ven-
[9] M. L. Nielsen and I. L. Chuang, Quantum Computa- galattore, Rev. Mod. Phys. 83, 863 (2011).
tion and Quantum Information (Cambridge University [30] A. J. Daley, M. Rigol, and D. S. Weiss, New J. Phys.
Press, 2000). 16, 095006 (2014).
[10] M. M. Wilde, Quantum Information Theory (Cam- [31] R. Kosloff, Entropy 15, 2100 (2013).
bridge University Press, 2013). [32] R. Kosloff and A. Levy, Annu. Rev. Phys. Chem. 65,
[11] J. von Neumann, Mathematical Foundations of Quan- 365 (2014).
tum Mechanics, Princeton landmarks in mathematics [33] D. Gelbwaser-Klimovsky, W. Niedenzu, and G. Kurizki,
and physics series (1955) pp. xii + 445, translated from Advances In Atomic, Molecular, and Optical Physics
the German edition by Robert T. Beyer. Original first 64, 329 (2015).
edition published in German in 1932. [34] L. Szilard, Zeitschrift fur Physik 53, 840 (1929).
[12] R. Landauer, IBM J. Res. Dev. 5, 183 (1961). [35] J. Gemmer, M. Michel, and G. Mahler, Quantum ther-
[13] R. Horodecki, H. Pawel, M. Horodecki, and modynamics (Lecture Notes in Physics vol 784) (Berlin,
K. Horodecki, Rev. Mod. Phys. 81, 865 (2009). Heidelberg:Springer, 2009).
[14] S. Bauml, M. Christandl, K. Horodecki, and A. Winter, [36] L. Amico, A. Osterloh, and V. Vedral, Rev. Mod. Phys.
Nat. Commun. 6, 6908 (2015). 80, 517 (2008).
[15] C. H. Bennett, G. Brassard, S. Popescu, B. Schumacher, [37] T. Fritz, Mathematical Structures in Computer Science
J. A. Smolin, and W. K. Wootters, Phys. Rev. Lett. 76, FirstView, 1 (2015).
722 (1996). [38] J. Millen and A. Xuereb, New J. Phys. 18, 011002
[16] B. Coecke, T. Fritz, and R. W. Spekkens, Inform. Com- (2016).
put. , (2016). [39] S. Popescu, A. J. Short, and A. Winter, Nature Phys.
[17] E. Jaynes, Phys. Rev. 106, 620 (1957). 2, 754 (2006).
[18] E. Jaynes, Phys. Rev. 108, 171 (1957). [40] S. Goldstein, J. L. Lebowitz, R. Tumulka, and
[19] J. von Neumann, Eur. Phys. J. H 35, 201 (2010). N. Zangh, Phys. Rev. Lett. 96, 050403 (2006).
31
[41] S. Lloyd, Black Holes, Demons and the Loss of Co- [74] L. Masanes, A. J. Roncaglia, and A. Acn, Phys. Rev.
herence: How complex systems get information, and E 87, 032137 (2013).
what they do with it, Ph.D. thesis, Rockefeller University [75] J.-S. Caux and J. Mossel, J. Stat. Mech. Theor. Exp.
(1991). 2011, P02023 (2011).
[42] D. Poulin, A. Qarry, R. Somma, and F. Verstraete, [76] L. del Rio, A. Hutter, R. Renner, and S. Wehner,
Phys. Rev. Lett. 106, 170501 (2011). (2014), arXiv:1401.7997.
[43] S. Garnerone, T. R. de Oliveira, and P. Zanardi, Phys. [77] E. H. Lieb and J. Yngvason, Phys. Rep. 310, 1 (1999).
Rev. A 81, 032336 (2010). [78] E. H. Lieb and J. Yngvason, Current Developments in
[44] F. G. S. L. Brandao, A. W. Harrow, and M. Horodecki, Mathematics, 2001 , 89 (2002).
(2012), arXiv:1208.0692. [79] P. Faist, F. Dupuis, J. Oppenheim, and R. Renner, Nat.
[45] A. Hamma, S. Santra, and P. Zanardi, Phys. Rev. Lett. Commun. 6, 7669 (2015).
109, 040502 (2012). [80] M. Horodecki, P. Horodecki, and J. Oppenheim, Phys.
[46] I. Affleck, T. Kennedy, E. H. Lieb, and H. Tasaki, Com- Rev. A 67, 062104 (2003).
mun. Math. Phys. 115, 477 (1988). [81] O. C. O. Dahlsten, R. Renner, E. Rieper, and V. Vedral,
[47] G. Vidal, Phys. Rev. Lett. 93, 040502 (2004). New J. Phys. 13, 053015 (2011).
[48] A. W. Harrow and R. A. Low, Commun. Math. Phys. [82] L. del Rio, J. Aberg, R. Renner, O. Dahlsten, and
291, 257 (2009). V. Vedral, Nature 474, 61 (2011).
[49] O. Szehr, F. Dupuis, M. Tomamichel, and R. Renner, [83] F. G. S. L. Brandao, M. Horodecki, N. H. Y. Ng, J. Op-
New J. Phys. 15, 053022 (2013). penheim, and S. Wehner, PNAS 112, 3275 ((2015)).
[50] N. Linden, S. Popescu, A. J. Short, and A. Winter, [84] G. Gour, M. P. Muller, V. Narasimhachar, R. W.
Phys. Rev. E 79, 061103 (2009). Spekkens, and N. Yunger Halpern, Phys. Rep. 583,
[51] A. J. Short and T. C. Farrelly, New J. Phys. 14, 013063 1 (2015).
(2012). [85] D. Janzing, P. Wocjan, R. Zeier, R. Geiss, and T. Beth,
[52] P. Reimann, Phys. Rev. Lett. 101, 190403 (2008). Int. J. Theor. Phys. 39, 2717 (2000).
[53] P. Reimann, Phys. Scripta 86, 058512 (2012). [86] F. G. S. L. Brandao, M. Horodecki, J. Oppenheim, J. M.
[54] A. J. Short, New J. Phys. 13, 053009 (2011). Renes, and R. W. Spekkens, Phys. Rev. Lett. 111,
[55] C. Gogolin, M. P. Muller, and J. Eisert, Phys. Rev. 250404 (2013).
Lett. 106, 040401 (2011). [87] J. Aberg, Nat. Commun. 4, 1925 (2013).
[56] A. Riera, C. Gogolin, and J. Eisert, Phys. Rev. Lett. [88] M. Horodecki and J. Oppenheim, Nat. Commun. 4, 2059
108, 08040 (2012). (2013).
[57] The density of states of the bath %B (E) is the number [89] J. M. Renes, The European Physical Journal Plus 129,
of eigenstates of the bath with energy close to E. 153 (2014), 10.1140/epjp/i2014-14153-8.
[58] A. Ferraro, A. Garca-Saez, and A. Acn, EPL (Euro- [90] M. Lostaglio, K. Korzekwa, D. Jennings, and
physics Letters) 98, 10009 (2012). T. Rudolph, Phys. Rev. X 5, 021001 (2015).
[59] M. Kliesch, C. Gogolin, M. J. Kastoryano, A. Riera, [91] P. Cwiklinski, M. Studzinski, M. Horodecki, and J. Op-
and J. Eisert, Phys. Rev. X 4, 031019 (2014). penheim, Phys. Rev. Lett. 115, 210403 (2015).
[60] M. P. Muller, E. Adlam, L. Masanes, and N. Wiebe, [92] M. Muller-Lennert, F. Dupuis, O. Szehr, S. Fehr, and
Commun. Math. Phys. 340, 499 (2015). M. Tomamichel, J. Math. Phys. 54, 122203 (2013).
[61] M. Srednicki, Phys. Rev. E 50, 888 (1994). [93] P. Faist, J. Oppenheim, and R. Renner, New J. Phys.
[62] M. Rigol, V. Dunjko, and M. Olshanii, Nature 452, 854 17, 043003 (2015).
(2008). [94] J. Aberg, Phys. Rev. Lett. 113, 150402 (2014).
[63] F. G. S. L. Brandao and M. Cramer, (2015), [95] P. Skrzypczyk, A. J. Short, and S. Popescu, Nat. Com-
arXiv:1502.03263. mun. 5, 4185 (2014).
[64] J. P. Keating, N. Linden, and H. J. Wells, Commun. [96] D. Jonathan and M. Plenio, Phys. Rev. Lett. 83, 3566
Math. Phys. 338, 81 (2015). (1999).
[65] D. Sels and M. Wouters, Phys. Rev. E 92, 022123 [97] N. H. Y. Ng, L. Maninska, C. Cirstoiu, J. Eisert, and
(2015). S. Wehner, New J. Phys. 17, 085004 (2015).
[66] J. Goold, C. Gogolin, S. R. Clark, J. Eisert, A. Scardic- [98] M. Klimesh, (2007), arXiv:0709.3680.
chio, and A. Silva, Phys. Rev. B 92, 180202 (2015). [99] S. Turgut, J. Phys. A: Math. Theor. 40, 12185 (2007).
[67] A. S. L. Malabarba, L. P. Garca-Pintos, N. Linden, [100] G. Gour and R. W. Spekkens, New J. Phys. 10, 033023
T. C. Farrelly, and A. J. Short, Phys. Rev. E 90, 012121 (2008).
(2014). [101] I. Marvian and R. W. Spekkens, Phys. Rev. A 90,
[68] S. Goldstein, J. L. Lebowitz, C. Mastrodonato, R. Tu- 062110 (2014).
mulka, and N. Zanghi, Phys. Rev. E 81, 011109 (2010). [102] M. F. Frenzel, D. Jennings, and T. Rudolph, Phys. Rev.
[69] F. G. S. L. Brandao, P. Awikliski, M. Horodecki, E 90, 052136 (2014).
P. Horodecki, J. K. Korbicz, M. Mozrzymas, and [103] H. Wilming, R. Gallego, and J. Eisert, Phys. Rev. E
P. Cwikliski, Phys. Rev. E 86, 031101 (2012). 93, 042126 (2016).
[70] Vinayak and M. Znidaric, J Phys. A: Math. Theor. 45, [104] N. Ramsey, Phys. Rev. 103, 20 (1956).
125204 (2012). [105] N. Linden, S. Popescu, and P. Skrzypczyk, Phys. Rev.
[71] M. Cramer, New J. Phys. 14, 053051 (2012). Lett. 105, 130401 (2010).
[72] S. Goldstein, T. Hara, and H. Tasaki, Phys. Rev. Lett. [106] N. Brunner, N. Linden, S. Popescu, and P. Skrzypczyk,
111, 140401 (2013). Phys. Rev. E , 051117 (2012).
[73] A. Hutter and S. Wehner, Phys. Rev. A 87, 012121 [107] S. Popescu, (2010), arXiv:1009.2536.
(2013).
32
[108] W. Pusz and S. L. Woronowicz, Commun. Math. Phys. [141] M. Huber, M. Perarnau-Llobet, K. V. Hovhannisyan,
58, 273 (1978). P. Skrzypczyk, C. Klockl, N. Brunner, and A. Acn,
[109] A. Lenard, J. Stat. Phys. 19, 575 (1978). New J. Phys. 17, 065008 (2015).
[110] D. Janzing, J. Stat. Phys. 122, 531 (2006). [142] D. E. Bruschi, M. Perarnau-Llobet, N. Friis, K. V. Hov-
[111] N. Yunger Halpern and J. M. Renes, Phys. Rev. E 93, hannisyan, and M. Huber, Phys. Rev. E 91, 032118
022126 (2016). (2015).
[112] S. M. Barnett and J. A. Vaccaro, Entropy 15, 4956 [143] R. Horodecki, M. Horodecki, and K. Horodecki, Rev.
(2013). Mod. Phys. 81, 865 (2009).
[113] J. A. Vaccaro and S. M. Barnett, Proc. R. Soc. A 467, [144] C. Eltschka and J. Siewert, J. Phys. A: Math. Theor.
1770 (2011). 47, 424005 (2014).
[114] M. Weilenmann, L. Kramer, P. Faist, and R. Renner, [145] J. I. de Vicente, C. Spee, and B. Kraus, Phys. Rev.
(2015), arXiv:1501.06920. Lett. 111, 110502 (2013).
[115] N. Yunger Halpern, (2014), arXiv:1409.7845. [146] S. Jevtic, D. Jennings, and T. Rudolph, Phys. Rev. A
[116] D. Reeb and M. M. Wolf, New J. Phys. 16, 103011 85, 052121 (2012).
(2014). [147] K. Funo, Y. Watanabe, and M. Ueda, Phys. Rev. A 88,
[117] D. Egloff, O. C. O. Dahlsten, R. Renner, and V. Vedral, 052319 (2013).
New J. Phys. 17, 073001 (2015). [148] M. Perarnau-Llobet, K. V. Hovhannisyan, M. Huber,
[118] M. P. Woods, N. Ng, and S. Wehner, (2015), P. Skrzypczyk, N. Brunner, and A. Acn, Phys. Rev. X
arXiv:1506.02322. 5, 041011 (2015).
[119] F. G. S. L. Brandao and G. Gour, Phys. Rev. Lett. 115, [149] M. Lostaglio, M. P. Muller, and M. Pastena, Phys. Rev.
070503 (2015). Lett. 115, 150402 (2015).
[120] N. Y. Halpern, A. J. P. Garner, O. C. O. Dahlsten, and [150] H. J. Groenewold, Int. J. Theor. Phys. 4, 327 (1971).
V. Vedral, New J. Phys. 17, 095003 (2015). [151] M. Ozawa, J. Math. Phys. 25, 79 (1984).
[121] R. Alicki, M. Horodecki, P. Horodecki, and [152] T. Sagawa and M. Ueda, Phys. Rev. Lett. 100, 080403
R. Horodecki, Open Systems & Information Dynamics (2008).
(OSID) 11, 205 (2004). [153] T. Sagawa and M. Ueda, Phys. Rev. Lett. 102, 250602
[122] R. Gallego, J. Eisert, and H. Wilming, (2015), (2009).
arXiv:1504.05056. [154] J. Oppenheim, M. Horodecki, P. Horodecki, and
[123] C. H. Bennett, IBM J. Res. Dev. 17, 525 (1973). R. Horodecki, Phys. Rev. Lett. 89, 180402 (2002).
[124] P. Skrzypczyk, A. J. Short, and S. Popescu, (2013), [155] R. Alicki and M. Fannes, Phys. Rev. E 87, 042123
arXiv:1302.2811. (2013).
[125] R. Landauer, Physics Today 44, 23 (1991). [156] K. V. Hovhannisyan, M. Perarnau-Llobet, M. Huber,
[126] R. Giles, Mathematical foundations of thermodynamics, and A. Acn, Phys. Rev. Lett. 111, 240401 (2013).
International series of monographs in pure and applied [157] F. C. Binder, S. Vinjanampathy, K. Modi, and
mathematics (Pergamon Press; [distributed in the West- J. Goold, New J. Phys. 17, 075015 (2015).
ern Hemisphere by Macmillan, New York], 1964). [158] M. Lewenstein, A. Sanpera, V. Ahufinger, B. Damski,
[127] C. Caratheodory, Math. Ann. 67, 355 (1909). A. Sen(De), and U. Sen, Adv. Phys. 56, 243 (2007).
[128] L. del Rio, Resource theories of knowledge, Ph.D. thesis, [159] V. Vedral, New J. Phys. 6, 102 (2004).
ETH Zurich (2015). [160] M. Dowling, A. Doherty, and S. Bartlett, Phys. Rev. A
[129] M. Horodecki and J. Oppenheim, Int. J. Mod. Phys. 27, 70, 062113 (2004).
1345019 (2012). [161] C. Brukner and V. Vedral, (2004), arXiv:quant-
[130] C. H. Bennett, H. J. Bernstein, S. Popescu, and ph/0406040.
B. Schumacher, Phys. Rev. A 53, 2046 (1996). [162] G. Toth, Phys. Rev. A 71, 010301 (2005).
[131] F. Verstraete, K. Audenaert, and B. De Moor, Phys. [163] O. Guhne, G. Toth, and H. J. Briegel, New J. Phys. 7,
Rev. A 64, 012316 (2001). 229 (2005).
[132] T.-C. Wei, K. Nemoto, P. Goldbart, P. Kwiat, [164] O. Guhne and G. Toth, Phys. Rev. A 73, 052319 (2006).
W. Munro, and F. Verstraete, Phys. Rev. A 67, 022110 [165] J. Anders, D. Kaszlikowski, C. Lunkes, T. Ohshima,
(2003). and V. Vedral, New J. Phys. 8, 140 (2006).
[133] L. Gurvits and H. Barnum, Phys. Rev. A 66, 062311 [166] V. Vedral, (2009), arXiv:0905.3057.
(2002). [167] M. Wiesniak, V. Vedral, and v. Brukner, Phys. Rev. B
[134] L. Gurvits and H. Barnum, Phys. Rev. A 68, 042312 78, 064108 (2008).
(2003). [168] M. Wiesniak, V. Vedral, and v. Brukner, New J. Phys.
[135] L. Gurvits and H. Barnum, Phys. Rev. A 72, 032322 7, 258 (2005).
(2005). [169] S. Bauml, D. Bru, M. Huber, H. Kampermann, and
[136] T. Yu, K. Brown, and I. Chuang, Phys. Rev. A 71, A. Winter, New J. Physics 18, 015002 (2016).
032341 (2005). [170] K. Sekimoto, Stochastic Energetics (Springer, 2010) p.
[137] M. Kus and K. Zyczkowski, Phys. Rev. A 63, 032307 322.
(2001). [171] P. Talkner, E. Lutz, and P. Hanggi, Phys. Rev. E 75,
[138] N. Johnston, Phys. Rev. A 88, 062330 (2013). 050102 (2007).
[139] S. Arunachalam, N. Johnston, and V. Russo, Quant. [172] G. E. Crooks, Phys. Rev. E 60, 2721 (1999).
Inf. Comput. 15, 0694 (2015). [173] H. Tasaki, , 11 (2000), arXiv:cond-mat/0009244.
[140] S. Jevtic, D. Jennings, and T. Rudolph, Phys. Rev. [174] C. Jarzynski, Phys. Rev. Lett. 78, 2690 (1997).
Lett. 108, 110403 (2012). [175] S. Deffner and E. Lutz, Phys. Rev. Lett. 105, 170402
(2010).
33
[176] H. Spohn, J. Math. Phys. , 1227 (1978). [208] S. Lorenzo, R. McCloskey, F. Ciccarello, M. Paternos-
[177] G. Huber, F. Schmidt-Kaler, S. Deffner, and E. Lutz, tro, and G. M. Palma, Phys. Rev. Lett. 115, 120403
Phys. Rev. Lett. 101, 070403 (2008). (2015).
[178] S. An, J. J.-N. Zhang, M. Um, D. Lv, Y. Lu, J. J.-N. [209] M. J. Donald, J. Stat. Phys. 49, 81 (1987).
Zhang, Z.-Q. Yin, H. T. Quan, and K. Kim, Nature [210] V. Vedral, Rev. Mod. Phys. 74, 197 (2002).
Phys. 11, 193 (2014). [211] K. Modi, T. Paterek, W. Son, V. Vedral, and
[179] R. Dorner, S. R. Clark, L. Heaney, R. Fazio, J. Goold, M. Williamson, Phys. Rev. Lett. 104, 080501 (2010).
and V. Vedral, Phys. Rev. Lett. 110, 230601 (2013). [212] S. Deffner and E. Lutz, Phys. Rev. Lett. 107, 140404
[180] L. Mazzola, G. De Chiara, and M. Paternostro, Phys. (2011).
Rev. Lett. 110, 230602 (2013). [213] F. Plastina, A. Alecce, T. J. G. Apollaro, G. Falcone,
[181] V. Giovannetti, S. Lloyd, and L. Maccone, Nature Pho- G. Francica, F. Galve, N. Lo Gullo, and R. Zambrini,
ton. 5, 222 (2011). Phys. Rev. Lett. 113, 260601 (2014).
[182] E. Knill and R. Laflamme, Phys. Rev. Lett. 81, 5672 [214] J. Hide and V. Vedral, Phys. Rev. A 81, 062303 (2010).
(1998). [215] F. Galve and E. Lutz, Phys. Rev. A 79, 032327 (2009).
[183] T. B. Batalhao, A. M. Souza, L. Mazzola, R. Auccaise, [216] A. Carlisle, L. Mazzola, M. Campisi, J. Goold,
R. S. Sarthour, I. S. Oliveira, J. Goold, G. De Chiara, F. Semiao, A. Ferraro, F. Plastina, V. Vedral, G. D.
M. Paternostro, and R. M. Serra, Phys. Rev. Lett. 113, Chiara, and M. Paternostro, (2014), arXiv:1403.0629.
140601 (2014). [217] M. Esposito, K. Lindenberg, and C. Van den Broeck,
[184] M. Campisi, R. Blattmann, S. Kohler, D. Zueco, and New J. Phys. 12, 013013 (2010).
P. Hanggi, New J. Phys. 15, 105028 (2013). [218] S. Jevtic, T. Rudolph, D. Jennings, Y. Hirono,
[185] L. Mazzola, G. De Chiara, and M. Paternostro, Int. J. S. Nakayama, and M. Murao, Phys. Rev. E 92, 042113
Quantum Inf. 12, 1461007 (2014). (2015).
[186] J. Goold, U. Poschinger, and K. Modi, Phys. Rev. E [219] O. Dahlsten, M. Choi, A. Garner, N. Yunger Halpern,
90, 020101 (2014). and V. Vedral, (2015), arXiv:1504.05152.
[187] J. P. S. Peterson, R. S. Sarthour, A. M. Souza, I. S. [220] S. Salek and K. Wiesner, (2015), 1504.05111.
Oliveira, J. Goold, K. Modi, D. O. Soares-Pinto, and [221] J. E. Geusic, E. O. Schulz-DuBois, and H. E. D. Scovil,
L. C. Celeri, Proc. R. Soc. Lond. A 472 (2016). Phys. Rev. 156, 343 (1967).
[188] A. J. Roncaglia, F. Cerisola, and J. P. Paz, Phys. Rev. [222] D. K. Park, N. A. Rodriguez-Briones, G. Feng, R. R.
Lett. 113, 250601 (2014). Darabad, J. Baugh, and R. Laflamme, (2015),
[189] G. De Chiara, A. J. Roncaglia, and J. P. Paz, New J. arXiv:1501.00952.
Phys. 17, 035004 (2015). [223] S. Carnot, Reflections sur la Puissance Motrice du Feu
[190] E. Mascarenhas, H. Braganca, R. Dorner, M. Franca et sur les Machines propres a Developper cette Puis-
Santos, V. Vedral, K. Modi, and J. Goold, Phys. Rev. sance) (Paris: Bachelier, 1824).
E 89, 062103 (2014). [224] N. Brunner, M. Huber, N. Linden, S. Popescu, R. Silva,
[191] T. Sagawa and M. Ueda, Phys. Rev. Lett. 104, 090602 and P. Skrzypczyk, Phys. Rev. E 89, 032115 (2014).
(2010). [225] L. A. Correa, J. P. Palao, G. Adesso, and D. Alonso,
[192] T. Sagawa and M. Ueda, Phys. Rev. E 85, 021104 Phys. Rev. E 87, 042131 (2013).
(2012). [226] L. Schulman and U. Vazirani, in Proceedings of the
[193] S. Toyabe, T. Sagawa, M. Ueda, E. Muneyuki, and 31st STOC (ACM Symposium on Theory of Comput-
M. Sano, Nature Phys. 6, 988 (2010). ing) (ACM Press, New York, 1999).
[194] T. Sagawa and M. Ueda, Phys. Rev. Lett. 109, 180602 [227] P. O. Boykin, T. Mor, V. Roychowdhury, F. Vatan, and
(2012). R. Vrijen, Proc. Natl. Acad. Sci. U.S.A. 99, 3388 (2002).
[195] Y. Morikuni and H. Tasaki, J. Stat. Phys. 143, 1 (2011). [228] H. Ollivier and W. H. Zurek, Phys. Rev. Lett. 88,
[196] M. Campisi, P. Talkner, and P. Hanggi, Phys. Rev. 017901 (2001).
Lett. 105, 140601 (2010). [229] L. Henderson and V. Vedral, J. Phys. A: Math. Gen.
[197] M. Campisi, P. Talkner, and P. Hanggi, Phys. Rev. E 34, 6899 (2001).
83, 041114 (2011). [230] M. T. Mitchison, M. P. Woods, J. Prior, and M. Huber,
[198] G. Watanabe, B. P. Venkatesh, and P. Talkner, Phys. New J. Phys. 17, 115013 (2015).
Rev. E 89, 052116 (2014). [231] J. B. Brask and N. Brunner, Phys. Rev. E 92, 062101
[199] G. Watanabe, B. P. Venkatesh, P. Talkner, M. Campisi, (2015).
and P. Hanggi, Phys. Rev. E 89, 032114 (2014). [232] J. Ronagel, O. Abah, F. Schmidt-Kaler, K. Singer, and
[200] D. Kafri and S. Deffner, Phys. Rev. A 86, 044302 (2012). E. Lutz, Phys. Rev. Lett. 112, 030602 (2014).
[201] A. Holevo, IEEE Trans. Inf. Theory 44, 269 (1998). [233] L. A. Correa, J. P. Palao, D. Alonso, and G. Adesso,
[202] V. Vedral, J. Phys. A: Math. Theor. 45, 272001 (2012). Sci. rep. 4, 3949 (2014).
[203] A. E. Rastegin, J. Stat. Mech. , P06016 (2013). [234] R. Uzdin, A. Levy, and R. Kosloff, Phys. Rev. X 5,
[204] T. Albash, D. A. Lidar, M. Marvian, and P. Zanardi, 031044 (2015).
Phys. Rev. E 88, 032146 (2013). [235] A. E. Allahverdyan, K. Hovhannisyan, and G. Mahler,
[205] A. E. Rastegin and K. Zyczkowski, Phys. Rev. E 89, Phys. Rev. E 81, 051129 (2010).
012127 (2014). [236] L. Quiroga, F. J. Rodrguez, M. E. Ramrez, and
[206] J. Goold and K. Modi, (2014), arXiv:1407.4618. R. Pars, Phys. Rev. A 75, 032308 (2007).
[207] J. Goold, M. Paternostro, and K. Modi, Phys. Rev. [237] I. Sinaysky, F. Petruccione, and D. Burgarth, Phys.
Lett. 114, 060602 (2015). Rev. A 78, 062301 (2008).
[238] M. Ban, Phys. Rev. A 80, 032114 (2009).
34
[239] E. Ferraro, M. Scala, R. Migliore, and A. Napoli, Phys. [244] L.-A. Wu and D. Segal, Phys. Rev. A 84, 012319 (2011).
Scripta T140, 014042 (2010). [245] B. Bellomo and M. Antezza, EPL (Europhysics Letters)
[240] F. Kheirandish, S. J. Akhtarshenas, and H. Moham- 104, 10006 (2013).
madi, The European Physical Journal D 57, 129 (2010). [246] B. Bellomo and M. Antezza, New J. Phys. 15, 113052
[241] M. Scala, R. Migliore, A. Messina, and L. L. Sanchez- (2013).
Soto, Euro. Phys. J. D 61, 199 (2010). [247] J. B. Brask, G. Haack, N. Brunner, and M. Huber, New
[242] X. L. Huang, J. L. Guo, and X. X. Yi, Phys. Rev. A J. Phys. 17, 113029 (2015).
80, 054301 (2009).
[243] N. Pumulo, I. Sinayskiy, and F. Petruccione, Physics
Letters A 375, 3157 (2011).