Quench Good Review

Dynamics of Quantum Many Body Systems Far From Thermal Equilibrium – DRAFT
Marco Schiró
T
Institut de Physique Théorique, Université Paris Saclay, CNRS, CEA, F-91191 Gif-sur-Yvette, France
(Dated: February 16, 2016)
The concept of thermodynamic equilibrium is one of the building blocks of our modern under-
standing of interacting many-particle systems. Together with quantum mechanics it provides a
powerful framework to describe a large variety of natural phenomena. However, recent years have
seen enormous experimental progress in preparing, controlling and probing strongly interacting
quantum systems in different non-equilibrium regimes. These developments have brought fresh new
interest around a set of fundamental questions concerning dynamics, dissipation, transport and the
approach to thermal equilibrium in isolated and open quantum many-body systems. The aim of this
course is to present the motivations and the main theoretical questions behind this research, and
to discuss the recent progress in the general understanding of strongly interacting quantum systems
AF
far from equilibrium, with a special emphasis on their dynamics.
Contents
I. Introduction, Motivations and Physical Examples 1

A. Falling Gently Out of Equilibrium 2
B. Why Quantum Nonequilibrium? 4
C. Outline of the course 5
D. General Readings (to be completed) 5
II. Dynamics of Closed Quantum Systems: Intro 7

A. Quantum Dynamics, Generalities 7
1. The simplest nonequilibrium protocol: a "Quantum Quench" 8
B. Time-Independent Hamiltonian, Isolated Systems 10
C. Quantum Thermalization, Eigenstate Thermalization Hypothesis 12
1. More on ETH: Origin and Consequences 13
D. Dephasing after a Sudden Quench: The example of Quantum Ising Chain 17
1. Diagonalization by Jordan-Wigner Transformation 17
E. Equilibrium Transverse Magnetization and Total Energy 20
DR
1. Quantum Dynamics in the TFIC 20
F. Results for the TFIC Quench Dynamics 22
G. Generalized Gibbs Ensemble 23
III. Short-to-Intermediate Time Dynamics: Prethermalization and Dynamical Transitions 24

A. Prethermalization in the Fermi Hubbard Model 24
B. Dynamical Transitions in the Fully Connected Transverse Field Ising 27
1. Dynamical Restoration of the Symmetry 29
C. The Role of Dephasing in the Dynamical Transition: Quenches in the AFM Hubbard Model 30
IV. Periodically Driven Isolated Quantum Systems 31

A. Intro to Floquet Theory - First formulation 31
B. Floquet Theory, second formulation 32
C. Floquet Approach and the Transformation to Rotating Frame 33
References 34
I. INTRODUCTION, MOTIVATIONS AND PHYSICAL EXAMPLES
These lectures focus on the physics of quantum systems made of many interacting degrees of freedom, either
fermions, bosons or quantum spins, in regimes far away from thermal equilibrium. This last statement is
both crucial, since it will force us to look at things in a very different way from the traditional focus of condensed
2
matter physics, and intentionally vague, reflecting the intrinsic generality of out of equilibrium physics. This indeed
requires to specify not only information on the system but also on the protocol used to bring it away from the thermal
equilibrium state. The aim of this first lecture is to put this research in a broad general context, to give some
introduction to the subject and to provide few physical examples that serve as motivation.
A. Falling Gently Out of Equilibrium
We start the discussion from a very general and qualitative definition of what thermal equilibrium is. We could say
that a macroscopic system is said to be in thermal equilibrium when
• its state (physical properties) is defined in terms of a unique set of intensive and extensive variables which
do not change with time (stationarity)
• no currents of charges associated to conserved quantities (particles, energy, ...) flow through it
Needless to say, such a concept is more an exception than a rule in every day life. Indeed, non equilibrium effects
are extremely common in many different physical situations of the greatest simplicity, the flow of electric current
through a metallic conductor being just a trivial example. This is even more true if we think that the very basic
idea of performing experiments on materials and compounds amounts to act with some external field on an otherwise
equilibrium system and to monitor its response to the applied perturbation. As a result one could be tempted to
ask why, inspite of the restrictiveness of the above definition, the assumption of thermal equilibrium has been (and
actually still is) so powerful and useful to describe physical properties of macroscopic systems. A possible answer to
this question should be based on two observations.
1. From one side, as often happens in physics, what matters are the orders of magnitude of tipical time scales a
system need to reach a quasi-equilibrium state compared to the time scales on which observation takes place.
As a consequence, with a good degree of approximation, a macroscopic system can be considered as in thermal
equilibrium if all fast processes have taken place while the slowest one still have to occur. Clearly the distinction
between fast and slow depends on the observation time that is considered1–3 . Example: in solids, the electronic
relaxation times are of the order of picoseconds 10−12 s due to the coupling with lattice vibrations or other
excitations, so essentially in most conventional experiments one can take the electronic liquid as in thermal
equilibrium with its environment.
2. A second key observation comes from a basic result in the statistical theory of many particle systems which
goes under the apparently innocuous name of linear response theory. It simply states that a small external
perturbation can only probe small fluctuations around equilibrium. Hence, as long as applied fields are weak
enough, the system can be considered as in thermal equilibrium for all practically purposes. This major result is
of paramount importance in connecting theory to experiments, since it provides a way to compute experimental
relevant quantities such as susceptibilities in terms of equilibrium correlation functions, which are the natural
quantities in terms of which the many body problem is formulated.
Before moving into the realm of nonequilibrium quantum physics it might be therefore useful to recall some basic
textbook fact about equilibrium quantum statistical physics that will be useful throughout the course to better
appreciate the challenges associated with out of equilibrium quantum physics.
Let’s consider for the sake of simplicity a system with Hamiltonian H in canonical thermal equilibrium at temper-
ature T . Its properties can be described by a density matrix of Boltzmann-Gibbs form
e−βH
ρeq (H) = Z = Tre−βH (1)
Z
and from this quantity one can compute thermal averages,
hAieq = Tr (ρeq A) (2)
from which one can derive all thermodynamics, as well as dynamical correlation functions
CAB (t, t′ ) = hA(t)B(t′ )ieq = Tr (ρeq A(t)B(t′ )) (3)
where operators are evolved in the Heisenberg picture as A(t) = eiHt Ae−iHt . An important property of the equilibrium
state is that its density matrix its directly known once the Hamiltonian of the system is, i.e.
[ρeq , H] = 0 (4)
3
This has few important consequences: first, the system is time-translational invariant (a property that can equally be
taken as a definition of stationarity) i.e. all single-time observable are time independent while multipoint functions
only depend on time-differences, for example
CAB (t, t′ ) = Tr (ρeq A(t)B(t′ )) = Tr (ρeq A(t − t′ )B(0)) = CAB (t − t′ ) (5)
Secondly, one can take advantage of Eq.(4) to derive a set of very general results linking these correlation functions,
or to be more precise the so called "fluctuations"
1 1
FAB (t) = h{A(t), B(0)}i = (CAB (t) + CBA (−t)) (6)
2 2
to experimentally relevant quantities such as the so called linear response functions, encoding the response of the sys-
tem in thermal equilibrium to a small external perturbation, the most simple and gentle example of out of equilibrium
quantum systems.
To see this let us assume first to perturb the system with a time-dependent field fext (t) that couples to some
observable B, i.e. H ′ = H + Vext with Vext = fext (t)B, and then to compute an observable A in presence of such a
small perturbation. To linear order in fext we get
Z
δhA(t)i ≡ hA(t)i − hAi0 = dt′ χAB (t − t′ )f (t′ ) (7)
where the linear response or retarded susceptibility is defined as

i
χAB (t) = − θ(t)h[A(t), B(0)]i (8)
h̄
We can then define the dissipative response function as
′′ 1 1
χAB (t) = h[A(t), B(0)]ieq = (CAB (t) − CBA (−t)) (9)
2 2
and use the result
CBA (−ω) = e−βω CAB (ω) (10)
to obtain the so called Fluctuation-Dissipation theorem

βω ′′
FAB (ω) = coth χAB (ω) (11)
2
The above is the bosonic version, which is valid for physical observables. For fermionic fields, such as those entering
the single particle Green’s function, one defines for example the retarded and Keldysh components as

GR (t, t′ ) = −iθ(t − t′ )h ψ(t), ψ̄(t′ ) i (12)
K ′
′

G (t; t ) = ih ψ(t), ψ̄(t ) i (13)
and obtaines the fermionic version of FDT relation

K βω
G (ω) = tanh 2iImGR (ω) (14)
2
The basic fact encoded in Eq. (4), namely that the density matrix is an exponential function of the Hamiltonian,
suggests the idea of performing a Wick rotation from real to immaginary time evolution and consider correlation
functions of the form
CAB (τ ) = −hTτ A(τ )B(0)i A(τ ) = eHτ A e−Hτ
The resulting Matsubara technique, which forms the basic of equilibrium quantum many body physics, allows to
obtain real-frequency retarded correlation functions from imaginary time evolution through analytic continuation,
and therefore, as a consequence, crucially removes "time" from the picture from the very beginning. We will see that
an important consequence of going beyond linear response is that time evolution will come back at play and in a
pretty major role.
4
Another interesting example of basic equilibrium assumptions that will need to be revised in the more general
out of equilibrium context concerns the way quantum many body theory is usually built at zero temperature. Here
one assumes to start from a non-interacting Hamiltonian H0 in the infinite distant past and slowly switching on the
interaction
H(t) = H0 + e−|ε|t Hint (15)
Then one typically focus on multi-point correlations functions of the form
G(t1 , · · · , tn ) = hΨ|T (A(t1 ) · · · A(tn )) |Ψi (16)
where |Ψi is the interacting ground state in the infinite future. A crucial result is that this can be written as
G(t1 , · · · , tn ) = h+∞|S[+∞, −∞]T (A(t1 ) · · · A(tn )) | − ∞i (17)

R
t
where we have defined the S-matrix S[+∞, −∞] = T exp −i 0 dtV (t) and where |± ∞i are the ground-states in the
infinite past and infinite future, in the interaction representation. The crucial result4 that follows from adiabaticity
is that these two states are essentially the same, up to a phase, since the process of switching on and off slowly the
interaction does not affect the system, hence
| + ∞i = eiδ |∞i eiδ = h−∞| + ∞i (18)
so that we obtain
h−∞|S[+∞, −∞]T (A(t1 ) · · · A(tn )) | − ∞i
G(t1 , · · · , tn ) = (19)
h−∞| + ∞i
This is a crucial result to build the field theory approach to zero temperature quantum many body systems which
however will need to be revised when dealing with arbitrary time dependent switching protocols which do not guarantee
adiabaticity. The proper field theory approach goes under the name of Keldysh techniques (also due to Schwinger,
Kadanoff, Baym, see Refs. 5–7). We will come back to it later in more detail, here we just mention the main idea. In
a generic nonequilibrium condition the state of the system at infinite time is not known a-priori, but it is itself an
unknown of the problem together with the energy spectrum, which can be strongly modified by the nonequilibrium
perturbation. Then the basic is to design an approach where only the initial condition enters: this amounts to fold
the real-time countour around the time axis and give rise to the celebrated Keldysh contour.
Equilibrium Quantum Statistical Physics is well developed and very solid theoretical framework to describe the
properties of quantum many body systems (mostly inspired from solid state physics). From the basic assumptions
reviewed above many powerful concepts have been developed, mainly inspired by the concept of equilibrium univer-
sality and by the fact that at low temperature the physics is dominated by the groundstate plus low lying
excitations. Within this low energy approach one can attack the quantum many body problem using a broad array
of concepts and tools, such as field theory techniques, renormalization group approaches, various sorts of mean field
theories plus fluctuations, or powerful numerical methods (different flavours of Quantum Monte Carlo algorithms,
DMRG and Matrix Product States algorithms and their extensions,..). Obviously this does not mean that everything
that is equilibrium is simple and easy, for example strongly correlated quantum many body systems are a continu-
ous challenge to theory and experiments for their unusual and exotic properties. Still it is good to appreciate the
conceptual challenges associated with building a theory of quantum nonequilibrium systems.
B. Why Quantum Nonequilibrium?
A natural question that can arises at this point is therefore why should we care about quantum non-equilibrium
problems. The reasons are twofold
• Fundamental Interest: the condition of thermal equilibrium is not only an idealized situation but also a very
special one. Relaxing this constraint we open to questions about fundamental aspects of quantum mechanics
in unconventional regimes. At the same time it is clear this cannot be enough, since many of these questions
were around since the early days of quantum mechanics (Von Neumann 1929, for example) and the theoretical
framework was developed in the mid sixties (Keldysh,..).
5
• Experimental Relevance: Recent years have seen tremendous progress in experimentally realizing quantum
systems which can be brought out of equilibrium in a controlled way. Quantum Control is the key here and
in particular quantum control of light-matter interactions. These progresses are happening at a new interface,
between disciplines that did not use to cross each other that much in the past, such as Condensed Matter,
Atomic Physics and Quantum Optics.
Below (see associated slides of Lecture 1) we will give a quick introduction to the relevant experimental systems,
although unfortunately we won’t have the time to cover in details the physics behind these platforms. Rather will
present the main ideas that serve as a motivation for theoretical research. Luckily there are already few excellent
reviews that cover the latest experimental progresses in this rapidly evolving field. For example, a classic and
general review on Ultracold Atoms in Optical Lattices is Ref. 8, while Refs. 9–11 have more focus on nonequilibrium
aspects. For what concerns strongly interacting light-matter systems we can mention few references: a recent and
general review of collective quantum phenomena with light is Ref. 12, with Ref. 13 more focused on exciton-polariton
condensates; for cavity/circuit QED arrays we can mention few reviews, such as Refs 14–16, while for Ultracold Atoms
in Optical Cavities a classic reference is Ref. 17.
C. Outline of the course
These lectures will be divided in three main parts, covering the dynamics of quantum many body systems under
different nonequilibrium regimes. Clearly the subject is extremely broad both in terms of nonequilibrium settings and
model systems considered and we will have to make a selection.
In the first part we will focus on the dynamics of thermally isolated systems after rapid change in time of
some parameters, the so called quantum quenches, motivated by experiments with ultracold gases and pump-probe
spectroscopies. This will allow us to touch fundamental questions related to time evolution in quantum mechanics, the
approach to a stationary regime and to the state of thermal equilibrium, the possibility of non trivial phenomenon in
the quantum evolution, the existence of general mechanisms which can prevent the system from reaching equilibrium.
The second main subject of these lectures will be the dynamics of open driven and dissipative quantum
systems, i.e. systems which are under the continuous effect of an external driving field and of a coupling to an
environment. This part will be motivated by experiments with quantum optics platforms. We will discuss the
theoretical framework of open quantum systems (master equations, Linblad evolution,..) and connect it with the
more conventional (in condensed matter) Keldysh field theory approach. We will address questions on the nature
of the steady state in driven dissipative systems, the possibility of dissipative quantum phase transitions and their
relation with standard (quantum) critical phenomena, the emergence of universality out of equilibrium.
In between will briefly review the problem of closed quantum many body systems under periodic driving, the so
called Floquet quantum mechanics now in the context of a many body problem. This is emerging as a new exciting
direction in nonequilibrium quantum physics, with several theoretical open questions and experimental platforms and
it will be a natural crossover between the two worlds.
D. General Readings (to be completed)
The literature on systems far away from thermal equilibrium is very broad and there are a number of articles and
reviews of general interest on the topic, covering a variety of subjects both in classical and quantum physics. Among
these, I could mention in particular a recent Nature Physics Insight57 , together with the references already mentioned
for the experimental platforms for quantum nonequilibrium physics (see previous page).
In addition there are excellent reviews focused on quantum many body systems far from equilibrium with some (or
more than some) overlap with the material that will be coverd in the lectures. We list here a few general references
divided for topic and refer the reader to the extended bibliography at the end of the lecture notes, where we quote
various works relevant to our discussion.
• Dynamics of Closed Quantum Many Body Systems
1. Polkovnikov, Sengupta, Silva, Vengalattore, Review of Modern Physics, 83, 863 (2011)
2. Eisert, Friesdorf, Gogolin, Nature Physics, 11, 124 (2015) (see also arXiv:1503.07538 )
• Dynamics of Open Driven and Dissipative Quantum Many Body Systems
1. Sieberer, Buchhold, Diehl, arXiv:1512.00637
6
There are also few books that cover what is by now the more standard material, such as for example the Keldysh
formalism. Among these we could mention
• Rammer, Quantum Field Theory of Nonequilibrium States, Cambridge University Press, 2007
• Kamenev, Field Theory of Non-Equilibrium Systems, Cambridge University Press, 2011
• Altland and Simons, Condensed Matter Field Theory, Cambridge University Press, 2010
Rammer’s book has an introduction to Keldysh technique which is formal, abstract and field theory inspired (diagram-
matics, generating functionals, Effective Action, ...) but lacks a bit on the side of applications and examples, which
are few and pretty old fashioned. On the contrary Kamenev’s book has more a focus on applications than formalism,
although Keldysh technique is explained in detail (through several examples) mostly in the language of path integrals.
Altland and Simons book is a truly excellent condensed matter theory book, which covers nonequilibrium physics
(classical and quantum) among many other different subjects, but which is worth to be included in the list since the
presentation is generally clear and slightly complementary to the previous two books. Overall the three books are
good for what concerns general motivations and basic formalism, however they are all a bit off in terms of topics and
subjects with respect to what is currently in the focus of the research on nonequilibrium quantum many body systems
for which I would certainly recommend the reviews above and the articles mentioned in the lectures.
7
II. DYNAMICS OF CLOSED QUANTUM SYSTEMS: INTRO
Here we could discuss some general fact about dynamics of closed quantum systems evolving unitarily, motivated
by experiments with ultracold atoms and (to a certain extent) pump and probe spectroscopies.
A. Quantum Dynamics, Generalities
We consider a closed quantum many-body system initially prepared at time t0 = 0 in some initial state |Ψ(0)i and
evolving unitarily under the action of its Hamiltonian H(t), possibly time dependent. The state of the system at time
t satisfies the time-dependent Schroedinger equation
i∂t |Ψ(t)i = H(t)|Ψ(t)i (20)
whose general solution reads
Z t
′ ′
|Ψ(t)i = U (t)|Ψ(0)i = T exp −i dt H(t ) |Ψ(0)i (21)
0
Let us recall some basic and general fact about unitary evolution, before start specifying more in detail the problem.
• First, the evolution is linear, that is the superposition principle holds, i.e. if |Ψ1,2 (t)i are two solutions of
Eq. (20) with initial conditions |Ψ1,2 (0)i then one can construct the state
|Ψ(t)i = c1 |Ψ1 (t)i + c2 |Ψ2 (t)i (22)
which is again solution of Eq.(20) with initial condition |Ψ(0)i = c1 |Ψ1 (0)i + c2 |Ψ2 (0)i.
• If we construct the density matrix ρ(0) = |Ψ(0)ihΨ(0)| whose time evolution reads
ρ(t) = U (t)ρ(0)U † (t) ∂t ρ(t) = −i[H(t), ρ(t)] (23)
we can see that the purity is conserved
Trρ2 (t) = TrU (t)ρ(0)U † (t)U (t)ρ(0)U † (t) = Trρ2 (0) = 1 (24)
i.e. in other words pure states remain pure.
• We can see this also by introducing the Von Neuman entropy associated to the density matrix ρ(t), i.e.
S(t) = −Tr ρ(t) log ρ(t) (25)
It is then easy to show that the entropy remains constant for unitary evolution. Indeed using the definition of
the logarithm of an operator, we have
log ρ(t) = log(U (t)ρ0 U † (t)) = U (t) log ρ0 U † (t), (26)
from which we get
S(t) = −Tr ρ(t) log ρ(t) = Tr ρ(0) log ρ(0) = 0 (27)
the last step being valid only for pure states. Alternatively we can compute the time derivative
d d
S = − Trρ(t) log ρ(t) = −Tr (ρ̇ log ρ) − Tr ρρ−1 ρ̇ = iTr ([H, ρ] log ρ) = 0 (28)
dt dt
• The average energy reads
E(t) = hΨ(t)|H(t)|Ψ(t)i = hΨ(0)|U † (t) H(t) U (t)|Ψ(0)i (29)

is instead not constant in time and its time-derivative satisfies
d
E(t) = hΨ(t)|∂t H(t)|Ψ(t)i (30)
dt
which can be easily obtained using the fact that i∂t U = H(t)U .
These results emphasize the linearity of quantum evolution and the fact that for closed systems the quantum infor-
mation is never really lost.
8
Figure 1: A Quantum Quench is among the simplest nonequilibrium protocol to study the unitary dynamics of closed quantum
systems. The Hamiltonian of the system is assumed to depend on time through some classical parameter λ(t) which is varied
in time according to some protocol. In figure we show a linear ramp of λ(t) between λ0 and λ with a ramp time τ . Two limits
have been particularly studied, the slow/adiabatic limit (τ → ∞) and the sudden one (τ → 0+ ). In the main text we will focus
on the latter.
1. The simplest nonequilibrium protocol: a "Quantum Quench"
So far we haven’t said much about the explicit form of the time dependent Hamiltonian, and now we would like
to be more specific about this and introduce a very neat and popular way of studying quantum dynamics, through
a non-equilibrium protocol known as quantum quench 18 . This will give a bit more of physical understanding of the
resulting nonequilibrium process.
Let us imagine that the system Hamiltonian depends on time through some global (i.e. equal everywhere in the
system) control parameter λ(t), i.e. H[λ(t)]. At time t = 0, when λ(0) = λ0 the system is prepared in the ground-state
of the Hamiltonian H[λ0 ] and then the parameter is varied between two values λ0 at time t = 0 and λ at time t = τ
according to some protocol (see figure). This is called quantum quench, in analogy with classical thermal quenches
where the temperature of the bath is changed in time.
The limit τ → 0+ , a so called sudden quantum quench, corresponds to the case in which the Hamiltonian changes
abruptly but the state has no time to adapt so, effectively, this corresponds to prepare a certain initial state and let
it evolve. This is the opposite regime of the adiabatic switching that we have discussed in the introduction.
In the language of the quench it is useful to discuss the energetics once more. Let us assume to have an initial
Hamiltonian H0i corresponding to the value λ0 (where the superscript i reminds that this is the initial pre-quench)
whose spectrum reads
H0i |φin i = Eni |φin i (31)
and let assume to prepare the system in the groundstate |φi0 i before performing the quench λ0 → λ. We can think of
this quench process as a nonequilibrium transformation in the space of couplings λ0 → λf and we can characterize the
energetics by looking at the statistics of the work done after the quench19 . In thermodynamics, the work is not a state
function (such as the internal energy) and therefore, quantum-mechanically, it cannot be described in terms of an
observable20,21 . Rather one can define its probability distribution, obtained by summing the conditional probabilities
i
of starting from state |φ0 and ending in the state |ni, such that En − E0i = W , i.e.
X
P (W ) = δ(W − En + E0i ))|hn|φi0 i|2 (32)
n
We notice that P (W ) has a threshold, i.e.
P (W ) = 0, W < δE = E0 − E0i
and that the spectral weight of the threshold peak is given by h0|φi0 i|2 which is nothing but the overlap between the
initial and final ground state. The average work done during the quench reads
hW i = EQ − E0i = E0 − E0i + (EQ − E0 ) = δE + Wexc (33)

9
Figure 2: Energetics of a Quantum Quench: After the sudden change of parameters, the system evolves with a constant in
time and extensive (i.e. proportional to the volume) energy EQ which is in principle far above its groundstate. The average
work performed during this change can be written as the sum of a purely "equilibrium" part - δEgs the difference in ground
state energy between initial and final Hamiltonians, which would correspond to an infinitely slow quench or a quasi-static
transformation - and a non-equilibrium contribution that accounts for excitation due to the finite velocity of the quench, Wexc .
Notice that while δEgs has not a definite sign (could be positive or negative depending on the specifics), the excitation energy
is positive and vanishes in the limit of τ → ∞.
where the excitation energy Wexc = EQ − E0 is in general positive and vanishes in the limit of infinitely slow process.
Finally, in the general case of an initial state at finite temperature we would get for the work statistics
X
P (W ) = ρim δ(W − En + Emi
))|hn|φim i|2 (34)
n,m
i
with ρim = e−βEm /Z. Quite interestingly, one can show that the work statistics P (W ) satisfies very general relations,
known as fluctuation theorems out of equilibrium22,23 , valid for both classical and quantum systems (for a review
in the quantum case see Refs. 20,21). In particular, the Crooks relation relates the probability distribution of the
work for two protocols (F,forward and B, backward) which are related by time reversal symmetry (so in the case of
a quantum quench this would be λ0 → λ and λ → λ0 ) and reads
PF (W )
= eβ(W −∆F ) (35)
PB (−W )
where ∆F is the equilibrium free energy difference between states with λ0 , λ. From this relation it immediately follows
the Jarzinsky equality, by integrating both sides of the equation
Z
dW PF (W )e−βW = e−β∆F (36)
These are truly remarkable results since relate completely out of equilibrium quantities, the statistics of the work
done during an arbitrary nonequilibrium process between two states, to the free energy difference between those
equilibrium states. Their applications are far reaching, from biophysics to quantum transport, see for example the
Nature Physics Insight on Nonequilibrium Physics, which I mentioned among general readings, for a recent review.
Inspite their generality, such relations are extremely easy to prove in the case of quantum isolated systems. To this
extent it is useful to introduce the generating function G(s) for forward and backward protocol
1 1
GF (s) = Tr e−βH0 eiH0 s e−iHs GB (s) = Tr e−βH eiHs e−iH0 s (37)
Z0 Z
Then it is easy using cycling property of the trace to show
Z0 GF (−s − iβ) = ZGB (s) (38)
10
if we Fourier transform
Z Z Z
Z0 Z0 βW ′ Z0 βW
PB (W ) = dseiW s GB (s) = dseiW s GF (−s − iβ) = e ds′ GF (s′ )e−iW s = e PF (−W ) (39)
Z Z Z
B. Time-Independent Hamiltonian, Isolated Systems
Let us now focus on the case of isolated and closed quantum systems, namely systems whose Hamiltonian is
time-independent, the state of the system at time t simply reads
|Ψ(t)i = e−iHt |Ψ(0)i (40)
and the energy is conserved
E(t) = hΨ(0)|H|Ψ(0)i ≡ EQ (41)
In this case there is a natural basis in which the quantum evolution becomes simple which corresponds to the eigenstates
of the Hamiltonian H, i.e. the states such that
H|Φn i = En |Φn i (42)
Then it is useful to decompose the initial state in this basis to obtain

X
|Ψ(0)i = cn (0)|Φn i (43)
n
with cn (0) = hΦn |Ψ(0)i, such that the time evolution of the state simply becomes
X
|Ψ(t)i = cn (t)|Φn i cn (t) = cn (0) eiEn t (44)
n
Then we can write the (conserved) energy

X
EQ = hΨ(0)|H|Ψ(0)i = |cn (0)|2 En (45)
n
as well as the energy fluctuation

2
X
(δEQ ) = |cn (0)|2 (En − EQ )2 (46)
n
It is useful to introduce the distribution function of the ensemble generated by the weights |cn (0)|2 , so called diagonal
ensemble, which reads
X
Pde (E) = |cn (0)|2 δ(E − En ) (47)
n
We notice that quite generically one can show that the diagonal ensemble is thermodynamically well behaved, i.e. the
fluctuations of the energy go to zero in the L → ∞, namely the distribution of |Cn |2 narrows around the value EQ .
HERE!!!! LOCAL HAMILTONIAN
In terms of the eigenstates and eigenenergies of the Hamiltonian the dynamics of any observable or correlation
function becomes "trivial", a series of oscillations at different frequencies, for example
X X X
hOit = hΨ(t)|O|Ψ(t)i = e−i(En −Em )t c∗m (0)cn (0) Omn = |cn (0)|2 Onn + e−i(En −Em )t c∗m (0)cn (0) Omn (48)
nm n n6=m
What can we say in general about this evolution? First and trivially not all operators will have a non trivial dynamics.
Simple exceptions are, for example, the conserved quantities, i.e. any operator I that commutes with the Hamiltonian
[H, I] = 0 will remain constant in time
hIit = hΨ(t)|I|Ψ(t)i = I0 (49)

11
In the following, we will distinguish cases in which the system admits few global conserved quantities (total energy,
particle number, total spin, etc) associated to obvious symmetries from cases of systems with an extensive set of (non
trivial) conserved quantities, so called integrable systems.58 .
Beside conserved charges what can we say about generic operators? Again we can cook up operators which have
trivial evolution, i.e. operators which project on a given eigenstate of the Hamilltonian, Λnm = |nihm| cannot relax
but just oscillate in time
Λnm (t) = ei(En −Em )t Λnm (50)
These however are very special cases which do not correspond to "reasonable" operators, since a many body eigenstate
is a highly non-local object and so the projector built out of it. Instead we will focus our attention to generic (and
physically motivated) few-body, local operators, for which we can expect in the thermodynamic limit, the dynamics
to reach a stationary (time-independent) state. This can be seen going back to the expression for the time-dependent
average and write it as
Z
hOit = hOiDE + dωeiωt FO (ω) (51)
where we have introduced the diagonal ensemble average (we will come back to it very soon)
X
hOiDE = |cn (0)|2 Onn (52)
n
as well as the spectral density

X
FO (ω) = c∗m (0)cn (0) Omn δ(ω − (En − Em )) (53)
n6=m
From this result we can understand why the process of relaxation depends crucially on taking the thermodynamic limit
first (and of course on the nature of the operator considered). In this limit the many body spectrum becomes dense
and the right-hand side vanishes at long time as a result of destructive interference of many very close frequencies, a
process called dephasing.
For finite size systems, characterized by a discrete spectrum, the oscillations in Eq. (REF) will revive at some point
in time. It is interesting to compute the temporal average as well as the temporal fluctuations of the classical variable
hOit . To this extent we define the time average of a time-dependent quantity as
Z T
1
f (t) = limT →∞ dtf (t) (54)
T 0
Then we have
X e−i(En −Em )T − 1
hOit = hOiDE + limT →∞ i c∗m (0)cn (0) Omn = hOiDE (55)
T (En − Em )
n6=m
Similarly we can compute the the fluctuations around this time average, namely
2
2
σO = hOi2t − hOit (56)
We note that these are classical fluctuations, and should not be confused with the fluctuations of the quantum average,
2
(see later). We can compute σO by direct substitution to get
 2
Z T
2 1 X
σO = limT →∞ dt hOit + e−i(En −Em )t c∗m (0)cn (0) Omn  − (hOit )2 = (57)
T 0 n6=m
Z T
1 X
= 2hOit limT →∞ dt e−i(En −Em )t c∗m (0)cn (0) Omn + (58)
T 0 n6=m
Z T
1 X
+limT →∞ dt c∗m (0)cn (0) c∗m′ (0)cn′ (0) e−i(En −Em +En′ −Em′ )t (59)
T 0 n6=m;n′ 6=m′
12
The first term vanishes while for the second term we get a finite contribution from the pairing n = m′ , m = n′ , i.e.
X
2
σO = |cm (0)|2 |cn (0)|2 |Onm |2 ≤ max|Onm |2 (60)
n6=m
In other words, the off diagonal matrix elements of an operator control the amplitude of temporal fluctuations around
the long-time average. These (classical) temporal fluctuations should not be confused with the quantum fluctuations
of the dynamics.
C. Quantum Thermalization, Eigenstate Thermalization Hypothesis
A fundamental question in statistical mechanics is wheter and under which condition isolated systems approach
thermal equilibrium under the evolution generated by their own Hamiltonian. Loosely with this statement we mean
that the dynamics goes to a stationary state and that this stationary state is thermal. More precisely we want that
• In the long-time limit the system reaches a stationary state, which can be described in terms of few parameters,
corresponding to the global conserved quantities of the dynamics (average energy, particle number,..).
• This should happen independently of the details of the initial condition (i.e. initial condition with the same
value of the global conserved quantities should approach the same long time limit), so first of all thermalization
requires loosing memory of the initial data.
• This steady state state can be described by the appropriate statistical ensemble consistent with the values of the
global conserved quantitites. Since we are considering the thermodynamic limit where the ensemble equivalence
generally holds (we consider system with short-range interctions) we could equally require thermalization to the
canonical ensemble at finite temperature T .
We are now in the position to start appreciating the apparent paradox of quantum thermalization. First,
due to the intrinsic linearity of unitary evolution, at the level of quantum states in the Hilber space there is no loss
of memory, i.e. neither the wave function or the density matrix could ever thermalize. Therefore to begin with, a
meaningful sense for thermalization can be only given for the average value of local operators (or for the density matrix
of subsystems of finite size embedded in our thermodynamically large system). A different way of seeing why we need
to focus on averages of operators is that in the Heisenberg picture they satisfy a dynamics, ∂t hOit = ih[H, O]it which
is in general highly non-linear (and possibly chaotic in the sense of classical mechanics). The reason for "locality" is
instead the intuition that as long as we observe a closed system locally we are essentially tracing out all the degrees
of freedom outside the region of interest, thus in a certain sense as if the rest of the system was playing of the role of
a bath for a smaller part of itself.
Still, for an operator approaching a stationary value, this would require according to Eq. (48)
X
hOiDE = |Cn (0)|2 Onn = hOiMC (EQ ) (61)
n
where we have introduced the equilibrium micro-canonical (MC) ensemble average

1 X
hOiMC (E) = Onn (62)
NE,∆E
n,|En −E|<∆E
with the energy set by the initial condition EQ = hΨ0 |H|Ψ0 i. At first sight the equality seems impossible, since on
the right hand side the information on the initial condition is only encoded in the total energy (plus the fluctuation
∆E) while on the left hand side clearly we have that the initial condition enters the distribution of Cn (0) = hn(0)|Ψ0 i.
Clearly, thermalization is hidden in the structure of the eigenstates and their projection on the initial state.
The Eigenstate Thermalization Hypothesis24–26 (ETH) sets a way out of this thermalization paradox by postulating,
in its essence, that thermalization is encoded in each eigenstate of a generic (non-integrable) Hamiltonian, which are
thermal.
The essential assumption is that for diagonal matrix element Onn = hOiMC (En ), i.e. eigenstates are thermal. This
can be understood a posteriori with naive two arguments: first, if thermalization holds for any initial condition it
should also hold if we pick as initial condition an exact eigenstate of the system, in which case the time evolution is
trivially hn|O(t)|ni = Onn . Hence in order this to be thermal one has to require that the eigenstate in the first place
are thermal. Another way of understand ETH ansatz is to consider this as an extreme limit of the microcanonical
13
ensemble, where the energy shell contains only a single eigenstate, so by equivalence of ensemble one should requre
Onn to be thermal. As the name suggest, ETH is an hypothesis, which has been to certain extent tested in numerical
calculations on (rather small) lattice quantum systems (see for example Refs. 27–29). We will come back to ETH at
the beginning of the second lecture and discuss more in detail its consequences, the numerical evidence in its support
and also discuss cases where it is known to fail.
1. More on ETH: Origin and Consequences
To be more precise ETH can be written as an ansatz on the matrix element of a "local" "few-body" operator in
the eigenbasis

hOiMC (Ē) m=n
Omn = −S(Ē)/2
Rmn fO (Ē, ω) e m 6= n
where Omc (Ē) is the microcanonical equilibrium average of the operator at the energy Ē = (En + Em )/2, fO (Ē, ω)
is a smooth function of the total energy Ē and of the energy difference ω = En − Em , S(E) is the thermodynamic
(microcanonical) entropy while on the off diagonal Rmn is a random (real/complex) matrix, such that hRnm idis =
2
0, hRnm idis = 1. Depending on whether the system satisfies time-reversal symmetry, i.e. has real or complex operators,
one obtain different properties such as
Rnm = Rmn → fO (Ē, −ω) = fO (Ē, ω) (63)

∗
Rnm = Rmn → fO∗ (Ē, −ω) = fO (Ē, ω) (64)
In order to grasp the origin of the ETH ansatz it is useful to recall the limit of an Hamiltonian described by Random
Matrix Theory. Let’s assume eigenvectors are random and let’s decompose an operator in its own eigenstates
X
O= Oα |αihα| (65)
α
Then for the matrix element

X ∗
Omn = Oα (ψαm ) ψαn (66)
α
Let’s average and let’s assume

1
hψ̄αm ψβn i = δmn δαβ (67)
D
we get
1 X
hOmm i = Oα ≡ Ō (68)
D α
hOmn i = 0 (69)
Fluctuations are
1 X 2 1 1
2
hOmm i − hOmm i2 = O ≡ O¯2 hOmn
2
i − hOmn i2 = O¯2 (70)
D2 α α D D
therefore we can say
O¯2
Omn = Ōδmn + Rmn (71)
D
where Rmn is some zero average unit variance random matrix.
14
a. Thermalization from ETH As we see ETH puts a lot of thermodynamic information into the matrix elements
of an operator. Let’s start discussing the first consequence of ETH, concerning the paradox of quantum evolution.
Let’s plug in the ETH ansatz in Eq. REF. We get
X Z
hOiDE = |Cn (0)|2 hOiMC (En ) = dEPDE (E)hOiMC (E) ≃ hOiMC (EQ ) (72)
n
where we have used the fact that in the thermodynamic limit the distribution Pde (E) is peaked around the value of
the conserved energy EQ . What about finite size fluctuations?We can expand Oeq (E) around E = EQ and plug it in
to get
1 ′′
hOide = Oeq (EQ ) + (δE)2 O (EQ ) (73)
2
i.e. the DE average match the MC average up to corrections related to the energy fluctuations of the DE,
Z
(δE)2 = dEPDE (E)(E − EQ )2 (74)
which vanish in the thermodynamic limit. Is that the end of the story? Not really!
While ETH is believed to be true in the thermodynamic limit for generic non-integrable systems, there are cases
in which we know for sure that ETH does not apply. The simplest example that we have already mentioned is the
case of integrable systems when one does not expect thermalization to occurr, due to the extensive set of conserved
quantities which highly constraint the dynamics. A less obvious case which has been recently recognized is the case
of strongly disordered quantum systems due to Anderson Localization. Finally, even for non-integrable systems we
expect deviation from ETH due to finite size effects and indeed there are several numerical studies which indicate lack
of thermalization after a quantum quench. In all these case it is interesting to understand why and in which sense
ETH could fail.
A key point raised by Ref28 concerns the effect of rare states which can strongly bias the diagonal ensemble
distribution. In order to appreciate this point it is crucial to study, as the size of the system is increased, the
distrubution of the matrix element Onn and eigenvalues En , namely
X
N (e, o) = δ(e − En /L)δ(o − Onn ) ∼ eS(e,O) (75)
n
√
Quite generically one can show28 that the width of such distribution shrinks in thermodynamic limit as 1/ L around
the value predicted by ETH, i.e. the microcanonical average. However this is not enough to ensure thermalization
since the crucial point is the behavior of the support of such distribution, namely if it remains finite or goes to zero as
well. In the former case there will remain rare states different from the ETH one, and therefore if the distribution of
|cn (0)|2 is strongly biased toward these states thermalization can be avoided. Indeed the diagonal ensemble average
involves the weighted distribution
X
Z(e, o) = |cn (0)|2 δ(e − En /L)δ(o − Onn ) ∼ eF (e,O) (76)
n
which can be in principle different from (75) , unless the weights |cn (0)|2 sample uniform the different eigenstates.
As opposite, if the support of the distribution N (e, o) also shrinks, namely if in the thermodynamic limit all states
satisfy ETH (also known as strong ETH), then thermalization is ensured. The results of Ref.28 show that such rare
states are indeed responsible for the absence of thermalization of integrable systems and might be at play as well in
finite size non-integrable models.
Another way of saying: there are two scenarios
• Weak ETH. S(e, o), F (e, o) peaked but the support doesn’t shirnk in the thermodynamic limit. Then one has
to be careful because they are in principle different functions (due to |cn (0)|2 ) with different saddle points:
oMC (e) such that ∂o S(e, o) = 0 and oMC (e) such that ∂o F (e, o) = 0. Thermalization emerges then only
if |cn (0)|2 samples states uniformly and result in an overall constant, which does not affect the maximum, i.e.
oMC (e) = oDE (e). Otherwise, if the initial condition bias the distribution one can avoid thermalization (example
of integrable systems).
• Strong ETH. S(e, o), F (e, o) peaked and the support shirnks in the thermodynamic limit. then the ansatz is
something like Z(e, o) ∼ θ(o − o∗ − δo) − θ(o − o∗ + δo) ∼ eF (e,O) with δo → 0 in the thermodynamic limit.
Then, all the states satisfy ETH and putative rare states do not play any role.
15
b. FDT from ETH Finally let’s show how from the ETH ansatz we can obtain the Fluctuation Dissipation
Relation26 . To this extent we introduce a dynamical correlation function of an operator O evaluated on a single
eigenstate |ni, i.e.
COn (t) = hn|O(t)O(0)|ni − hn|O(t)|nihn|O(0)|ni (77)
in terms of which we can write both the single-eigenstate retarded response function
Z ∞ Z ∞
χOn (ω) = i dteiωt h[O(t), O(0)]i = i dteiωt [CO (t) − CO (−t)] (78)
0 0
as well as the symmetric and antisymmetric (fluctuation and dissipation) correlation functions
Z Z
+ 1
Sn (ω) = dte hn| {O(t), O(0)} |nic = dteiωt (COn (t) + COn (−t))
iωt
(79)
2
Z Z
1
Sn− (ω) = dteiωt hn|[O(t), O(0)]|nic = dteiωt (COn (t) − COn (−t)) (80)
2
Using ETH we now derive the following result for COn (t), namely
Z
COn (t) = dωeβω/2 e−iωt |fO (En + ω/2, ω)|2 (81)
Let’s start from the definition of COn (t) and plug in a set of eigenstates to get
X
COn (t) = ei(En −Em )t |Onm |2 (82)
m6=n
Using ETH ansatz for Onm we get

X
COn (t) = e−S(Ē) e−iωt |fO (Ē, ω)|2 |Rnm |2 (83)
m6=n
where we have defined

En + Em ω
ω = Em − En Ē = = En + (84)
2 2
Let us now assume that summing over m average out the random contribution from the matrix R element Rmn , using
h|Rmn |2 i = 1 and that we can cast this summation over an integral in the continuum limit, using dΩδ(Ω+En −Em ) =
1 to get
X Z
COn (t) = dΩδ(Ω + En − Em )e−S(En +ω/2) e−iωt |fO (En + ω/2, ω)|2 (85)
m6=n
We now use the definition of microcanonical entropy as logarithm of number of available eigenstates namely S(E) =
log Γ(E), with
X X
Γ(E) = (θ(En − E − ∆E) − θ(En − E)) ≃ ∆E δ(E − En ) = ∆EN (E) (86)
n n
where N (E) is the many body density of states. From this we can write
X
δ(Ω + En − Em ) = ∆E exp S(En + Ω)
m
and therefore
Z
dΩ S(En +Ω)−S(En +Ω/2) −iΩt
COn (t) = e e |fO (En + Ω/2, Ω)|2 (87)
∆E
16
Finally we can expand the integrand for small Ω, i.e.

βΩ
S(En + Ω) − S(En + Ω/2) = (88)
2
(89)
with β = ∂S/∂E to get the result quoted in Eq (REF).

±
We can now plug this result into the definition of both SOn (ω) and χOn (ω) to obtain for the former
Z Z
1
S ± (ω) = dteiωt dΩeβΩ/2 |fO (En + Ω/2, Ω)|2 eiΩt ± e−iΩt = (90)
2

= π eβω/2 |fO (En + ω/2, ω)|2 ± e−βω/2 |fO (En − ω/2, −ω)|2 = (91)

= π cosh βω |fO (En + ω/2, ω)|2 ± |fO (En − ω/2, ω)|2 + (92)
2 2

+π sinh βω |fO (En + ω/2, ω)| ∓ |fO (En − ω/2, ω)| (93)
R
where we have used dteiωt = 2πδ(ω) and the symmetry properties of fO (E, ω). In particular we get for Sn+ (ω)

Sn+ (ω) = π cosh βω |fO (En + ω/2, ω)|2 + |fO (En − ω/2, ω)|2 + (94)
2 2

+π sinh βω |fO (En + ω/2, ω)| − |fO (En − ω/2, ω)| (95)
Similarly we can obtain, using

Z ∞
1
ei(ω+i0 )t = iπδ(ω) − P ( )
+
i (96)
0 ω
the following result for the imaginary part of χOn (ω)

ImχOn (ω) = π sinh βω |fO (En + ω/2, ω)|2 + |fO (En − ω/2, ω)|2 + (97)

+π cosh βω |fO (En + ω/2, ω)|2 − |fO (En − ω/2, ω)|2 (98)
We now notice that the functions in parenthesis are smooth and can be expanded for small ω to obtain

|fO (En + ω/2, ω)|2 + |fO (En − ω/2, ω)|2 ≃ 2|fO (En , ω)|2 (99)

|fO (En + ω/2, ω)|2 − |fO (En − ω/2, ω)|2 ≃ ω∂E |fO (En , ω)|2 (100)
Then we find

ω∂E |fO (En , ω)|2
Sn+ (ω) = 2π cosh βω/2|fO (En , ω)|2 1 + tanh βω/2 (101)
2|fO (En , ω)|2
and similarly

ω∂E |fO (En , ω)|2
Im χO (ω) = 2π sinh βω/2|fO (En , ω)|2 1 + coth βω/2 (102)
2|fO (En , ω)|2
From this we can get an effective distribution function, i.e.
Sn+ (ω) ∗
= Fdist (ω) (103)
ImχO (ω)
where the effective distribution function reads
2

1 + tanh βω/2 ω∂2|f
E |fO (En ,ω)|
O (E n ,ω)| 2
∗
Fdist (ω) = coth (βω/2) 2
(104)
1 + coth βω/2 ω∂2|f
E |fO (En ,ω)|
O (E n ,ω)| 2
Assuming that the energy dependence of |fO (E, ω)| is weak such that we can disregard the second term
we finally get
∗
Fdist (ω) ≃ coth (βω/2) (105)
17
D. Dephasing after a Sudden Quench: The example of Quantum Ising Chain
We now turn the previous discussion, which was done in full generality, to a concrete example, and compute the
dynamics after a quantum quench in a simple model, namely the Transverse Field Ising Chain (TFIC), which is
characterized by the following Hamiltonian
L
X L
X
H0 = −J σix σi+1
x
− Γ0 σiz (106)
i i
where L is the number of the spins in the chain and σiα (α=x,y,z) are the Pauli matrices relative to the i-th spin.
This model represents a paradigm solvable example of a quantum phase transition and it has been therefore
the subject of a large literature30 . In equilibrium at zero temperature and in the thermodynamic limit it exhibits
ferromagnetic (Γ0 < 1) and paramagnetic (Γ0 > 1) phases, separated by a quantum critical point at Γc = 1 (for choice
of J = 1). For Γ0 < 1 and L → ∞ there are two degenerate ground states related by the Z2 symmetry. Spontaneous
symmetry breaking selects unique ground state, in which spins align along the x-direction. On the other hand, for
fields Γ > 1 the ground state is non-degenerate and as the magnetic field Γ is increased, spins align more and more
along the z-direction. The order parameter for the quantum phase transition is the ground state expectation value
hσix i. Furthermore we refer to the ferromagnetic phase as the ordered phase, and to the paramagnetic one as the
disordered phase.
As discussed earlier,
Let’s discuss the quench protocol we are going to focus in the following. We consider the system initially prepared
at time t0 = 0 in the ground state |ψ0 i of Eq. (106). We then suddenly change the value of the transverse field,
Γ0 → Γ (global quench), so that for t > 0 the system evolves with the new Hamiltonian
L
X L
X
H = −J σix σi+1
x
−Γ σiz (107)
i i
The dynamics of the system after a sudden change of the transverse field, from Γ0 to Γ can be obtained exactly using
a Jordan-Wigner transformation and a time-dependent Bogolubov transformation, as we are going to show in the
next sections. As a result the calculation of simple observables, such as for example the transverse magnetization,
can be easily done analytically and we are going to present it in detail. The calculation of correlation functions is
a more challenging task for which recent developments have been obtained31–36 . In the next section we first review
briefly some basic equilibrium property of the TFIC and discuss its diagonalization which will be useful to discuss the
nonequilibrium dynamics. The reader which is familiar with this material can skip this section and move on directly
to XXX..
1. Diagonalization by Jordan-Wigner Transformation
The idea of mapping quantum spin 1/2 into (spinless) fermionic degrees of freedom has a long history in quantum
mechanics. It can be seen as a very natural one from the point of view of local Hilbert space, since the two up/down
states (eigenstates of σ z ) can be described in terms of presence/absence of a fermionic particle (eigenstates of the local
charge n = c† c) and analogously the spin flip operators σ ± as creation/annihilation of such a particle, i.e. σ + = f † ,
σ − = f and σ z = 1 − 2n. The problem with this naive identification comes from the statistics, namely from the fact
that spins on different sites are expected to commutes
[σiα , σjβ ] = 2iεαβγ δij σjγ (108)

while fermions to anticommute
n o
ci , c†j = δij (109)
In one dimension there is a way to overcome such a difficulty, namely to have a fermionic representation of the spin
which is consistent with the statistics, through the so called Jordan-Wigner transformation. The idea is to introduce
a fermionic representation of the quantum spin at each lattice site j of the form

σjx = Kj cj + c†j (110)
σjz = 1 − 2c†j cj (111)

18
where the so called string operator Kj , defined as

 
Y
Kj =  1 − 2c†l cl  , (112)
l<j
sets the proper sign by counting for the number of fermions preceding the j-th site considered by σjx . We notice that
Kj is an hermitian operator, such that [Ki , Kj ] = 0 and Ki2 = 1, One can see that these definitions are consistent
with fermionic anti-commutation relation/quantum spin algebra. In addition one can see that the definition of σiz is
consistent with this and does not involve the string operator, i.e.
σjz = 1 − 2σj+ σj− = 1 − 2c†j cj (113)
where we used the fact that Ki2 = 1. Under these transformations the TFIC Hamiltonian becomes in terms of fermions
L−1
X L
X
H = −J c†i ci+1 + c†i c†i+1 + hc + 2Γ c†i ci + (−1)NF c†L c1 + c†L c†1 + hc (114)
i=1 i=1
PL
where NF = i=1 c†i ci is the number of fermions in the chain. The last term can be accounted for by choosing either
periodic (PBC)/antiperiodic (ABC) boundary conditions for the fermions
cL+1 ≡ (−1)NF +1 c1 (115)
depending on whether their total number NF is odd/even. We have to focus to the fact that the Hamiltonian conserves
the parity of fermions and we restrict to the even sector which contains the ground state. Note that restricting to
one of the two sectors is justified only when one considers expectation values of operators which are defined in terms
of products of an even number of fermionic operators, i.e., such that they do not change the parity of the state that
they act on. We then perform a Fourier transform
1 X i kRj
cj = √ e ck (116)
L k
where the set of momenta k is given by

k = ±(2π/L) n n = 0, 1, · · · L/2 (P BC) (117)
k = ±(π/L) (2n + 1) n = 0, 1, · · · L/2 − 1 (ABC) (118)
(119)
as one can immediately see by setting cL = ±c1 in Eq. (116). (Since we have chosen NF even we take in consideration
k from ABC). The resulting Hamiltonian becomes
X X † †
H= εk c†k ck − J ck c−k eika + c−k ck e−ika (120)
k k
with εk = 2Γ − 2J cos k. We can simplify further the second term by noticing that
X † † X
ck c−k eika + c−k ck e−ika = cos k c†k c†−k + c−k ck +
k k
X X
+i sin k c†k c†−k − c−k ck = +i sin k c†k c†−k − c−k ck (121)
k k
where in the last step we have used the fact that

cos(−k) = cos(k)

c−k ck + ck c−k = − c†k c†−k + c−k ck
† †
(122)
so the first summation in Eq. (121) cancels out due to the contribution of positive and negative momenta. Finally we
have
X X
H= εk c†k ck − J i sin k c†k c†−k − c−k ck (123)
k k
19
We can write this expression in terms of the positive momenta only as

X X
H= εk c†k ck − c−k c†−k − 2J i sin k c†k c†−k − c−k ck (124)
k>0 k>0
To proceed with the diagonalization let’s introduce the Nambu-Gorkov spinor notation

ck
Ψk =
c†−k
to write the Ising Hamiltonian in the final form
X
H= Ψ̄k ĥk Ψk (125)
k>0
where ĥk is a 2 × 2 matrix which reads

εk −iγk
ĥk =
iγk −εk
where we have defined γk = 2J sin k. This formalism is particularly convenient to diagonalize the problem. Indeed
we can introduce the Pauli Matrices τi=x,y,z (to not be confused with the physical spin 1/2 operators!) and write

εk γk
ĥk = εk τz + γk τy = Ek τz − iτx = Ek τz e−2iθk τx = Ek e−iθk τx τz eiθk τx (126)
Ek Ek
Here we have defined the angle θk such as
εk γk
cos 2θk = sin 2θk = − (127)
Ek Ek
p
where Ek = ε2k + γk2 is the quasiparticle spectrum which reads
p
Ek = 2 J 2 + Γ2 − 2JΓ cos k (128)
From this result we see that quite generically the spectrum is gapped, i.e. the minimum amount of excitation energy
(obtained for k = 0) is finite and given by
p
∆g = 2 J 2 + Γ2 − 2JΓ (129)
but as a function of Γ/J there is a special point Γc = J where the gap vanishes continuously
∆g = 2J |1 − Γ/Γc | (130)
the nature of the spectrum changes and the system becomes gapless: the dispersion relation becomes linear at low
momenta.
√
Ek = 4 2J sin k/2 ∼ vk (131)
In order to complete the diagonalization process, we can now introduce a new set of fermionic fields

ηk iθk τx cos θk i sin θk ck
Φk ≡ † =e Ψk =
η−k i sin θk cos θk c†−k
or equivalently
Ψk = e−iθk τx Φk
such that the so called Bogolubov transformation reads
†
ck = cos θk ηk − i sin θk η−k (132)
c†−k = −i sin θk ηk + †
cos θk η−k (133)
which diagonalize the Hamiltonian
X X
H= Ek Φ̄k τz Φk = Ek ηk† ηk − η−k η−k
†
(134)
k>0 k>0
Alternatively, we could have obtained the same result by starting from the Hamiltonian (123), introducing a
new set of fermionic fields through the Bogolubov rotation of Eq.(132) and then substitute them into the Hamil-
tonian to finally impose all the anomalous terms of the form η † η † or η η to be absent in the quasiparticle Hamiltonian.
20
E. Equilibrium Transverse Magnetization and Total Energy
Here we use the mapping to fermions and the Bogolubov transformation to compute simple observables in thermal
equilibrium, both at zero temperature (corresponding to our initial state) and at finite temperature, which can be
relevant for comparison with the long-time steady state limit of the dynamics.
Transverse magnetization Let’s start with the definition and its expression in the fermionic language
1X z 2X
mz = hσi i = 1 − hnk i (135)
L i L
k
Then, we use the Bogolubov rotation (132) to write
hc†k ck i = cos2 θk hηk† ηk i + sin2 θk hη−k η−k

†
i = cos 2θk hηk† ηk i + sin2 θk (136)
and obtain
2X 2 2X
mz = 1 − sin θk − f (Ek ) cos 2θk (137)
L L
k k
where we have used the fact that the quasiparticle excitations are free fermions for which we have
hηk† ηk′ i = δk,k′ f (Ek ) (138)
Using the bisection identities we can write the result above also as
1X 1 X εk βEk
mz = cos 2θk (1 − 2f (Ek )) = tanh (139)
L L Ek 2
k k
At zero temperature we obtain the result

Z π
z 1 X εk dk Γ − J cos k
m (T = 0) = = √ (140)
L Ek −π 2π Γ2 + J 2 − 2JΓ cos k
k
We can easily check that this recovers the trivial limits, i.e. mz = 0 for Γ = 0 and mz = 1 for J = 0. The above
result could be re-derived also within the Nambu-Gorkov formalism.
Average Energy Here we compute the average energy at finite temperature T , within the Nambu-Gorkov formalism.
X X X
E(T ) = hHi = εk h c†k ck − c−k c†−k i − 2J i sin kh c†k c†−k − c−k ck i = hΨ̄k ĥk Ψk i (141)
k>0 k>0 k>0
The simplest way to proceed is to notice that an average is not affected by a unitary transformation. Therefore we
can insert in Eq. (141) the operator that diagonalize the Hamiltonian and obtain
X X βEk
E(T ) = Ek hΦ̄k τz Φk i = − Ek tanh (142)
2
k>0 k>0
1. Quantum Dynamics in the TFIC
We now consider the situation of interest, namely the case in which the system is taken out of equilibrium by a
quantum quench of the transverse field. We want to study averages of temporal dependent operators, whose dynamic
is governed by quenched Hamiltonian. Using Heisenberg notation:
hÔ(t)i = hψ0 |eiHt Ôe−iHt |ψ0 i (143)
Where the Hamiltonian H used for the dynamic of the operator O can not act directly on the ground state, because
it doesn’t commute with the equilibrium hamiltonian H0 . A convenient way to calculate the associated dynamics
correlations after the quench, consist in expressing the (time dependent) operators {ck (t), c†−k (t)}, in terms of
21
†
the quasi-particle operators {γk , γ−k } which diagonalize the original Hamiltonian H(Γ0 )34 . The merit of this pro-
cedure is evident when calculating expectation values over |ψ0 i, because γk operators, act trivially on their vacuum ψ0 .
We start from the initial and final Hamiltonian written in the fermionic language
X X
H0 = εk0 c†k ck − c−k c†−k − iγk c†k c†−k − c−k ck (144)
k>0 k>0
and
X X
H= εk c†k ck − c−k c†−k − iγk c†k c†−k − c−k ck (145)
k>0 k>0
where we have defined:
εk0 = Γ0 − 2 J cos k γk = J sin k (146)
and similarly for εk with Γ0 → Γ. The two quadratic Hamiltonians can be diagonalized in terms of two sets of
fermionic quasiparticles
X X
H0 = Ek0 ηk† ηk H= Ek γk† γk (147)
k>0 k>0
with energies
q q
Ek0 = ε2k0 + γk2 Ek0 = ε2k + γk2 (148)
A sudden change of the transverse field corresponds therefore to a sudden change of the gap.
It is useful to relate the quasi-particle operators ηk , ηk† of the initial hamiltonian H(Γ0 ) to quasi-particle operators
γk , γk† of the final hamiltonian H(Γ). To this extent we write the first Bogolubov rotation

ck cos θk0 −i sin θk0 ηk 0
−iθk τx ηk
= =e
c†−k −i sin θk0 cos θk0 †
η−k †
η−k
as well as the second one

ck cos θk −i sin θk γk γk
= = e−iθk τx
c†−k −i sin θk cos θk †
γ−k †
γ−k
Then, by combinining the two we finally get

γk i(θk −θk
0
) τx ηk
† =e †
γ−k η−k
which can be written as

γk cos δθk i sin δθk ηk
† = †
γ−k i sin δθk cos δθk η−k
where δθk = θk − θk0 and the Bogolubov angles θk0 , θk are defined respectively as
εk0 γk
cos 2θk0 = sin 2θk0 = − (149)
Ek0 Ek0
εk0 γk
cos 2θk = sin 2θk = − (150)
Ek Ek
Then we compute the time evolution of the fermionic operators
Using this result we can obtain the full time-dependence of the fermionic operators which is needed to evaluate
dynamical averages

ck (t) ηk
= M (t)
c†−k (t) k †
η−k
22
where the dynamical matrix reads in compact form
Mk (t) = cos θk0 cos Ek t − i cos(θk + δθk ) sin Ek t τ z +

−i cos Ek t sin θk0 τ x − i sin(θk + δθk ) sin Ek t τ y
n o
with τα=x,y,z the Pauli matrices. In order for the anti commutation relation for c†k1 , ck2 = δk1 ,k2 to be conserved,
we could easily check that the matrix obtained for the dynamical transformation, has Det = 1. Thus we have
found the unitary transformation, which allows us to compute average quantities, starting from an out of equilibrium
Hamiltonian.

cos θk cos ∆θk e−iEk t + sin θk sin ∆θk eiEk t i cos θk sin ∆θk e−iEk t − i sin θk cos ∆θk eiEk t
M(Γ, Γ0 ) =
−i sin θk cos ∆θk e−iEk t + i cos θk sin ∆θk eiEk t sin θk sin ∆θk e−iEk t + cos θk cos ∆θk eiEk t
namely
 
cos θk0 cos Ek t − i cos(θk + ∆θk ) sin Ek t −i cos Ek t sin θk0 + sin(θk + ∆θk ) sin Ek t
M(Γ, Γ0 ) =  −i cos Ek t sin θk0 − sin(θk + ∆θk ) sin Ek t 
cos θk0 cos Ek t + i cos (θk + ∆θk ) sin Ek t
F. Results for the TFIC Quench Dynamics
Out of Equilibrium Transverse Magnetization As we have done in thermal equilibrium, we now focus our attention
on the transverse magnetization and compute its time evolution after the quench using the time-dependent Bogolubov
transformation we have obtained. We notice that while this is a particularly straightforward observable to evaluate,
this is not true for the order parameter, hσix (t)i which mixes the two parity sectors of the Hilbert space and its
calculation requires a more sophisticated analysis. We could refer to a large literature where this quantity was widely
studied, In particular31,32,34 where quantities such that one-point hσix (t)i and two-point hσix (t) σi+l
x
(t)i correlation
functions of the order parameter, where studied.
We compute the transverse magnetization taking into account the dynamical matrix obtained in Eq.(??). Thus we
transform the expression below
1X z 2X † 2 X †
mz (t) = hσi (t)i = 1 − hck (t)ck (t)i = − hck (t)ck (t)i − hc−k (t)c†−k (t)i (151)
L i L L
k k>0
Using the matrix of the dynamical transformation defined above and taking the averages at T = 0, the resulting
expression is:
2X
mz (t) = [cos 2θk cos 2∆θk + sin 2θk sin 2∆θk cos (2Ek t)] (152)
L
k>0
We can manipulate Eq. (152) with the transformation in Eq 127 and in the continuum limit it becomes
Z π
dk 1
mz (t) = 2 2 ǫk ǫk ǫk0 + γk2 + γk2 (ǫk0 − ǫk ) cos (2Ek t) (153)
0 2π Ek Ek0
The quantity in Eq.(153), shows dephasing effect in the thermodynamic limit. Infact looking at Fig.(3), we could
observe quantum revivals effect: increasing the size of the system, revivals are moved later in time. Taking the
continuum limit in Eq.(153) (thus L → ∞), quantum recurrence "disappears", and we could observe just dephasing.
Dephasing effect is due to the locality of the global transverse magnetization in real space: infact if we consider the
quantity hnk i, we could state that it doesn’t dephase beacuse it is defined for a fixed momentum k (just one wave
function), thus
P using Fourier transformation, it is completely delocalized in real space. But if we integrate over all
momenta L1 k hnk i, superposition of many oscillatory curves, each one of them having a slightly different frequency,
with time progressing, they get out of phase and so their average value decays.
23
Γ0=0.5 Γ=0.8
0.4
0.35
0.3
0.25
0.2
0.15 L=300
0.1
0 100 200 300 400 500
0.4
0.35
0.3
0.25
0.2
mz(t)
0.15 L=500
0.1
0 100 200 300 400 500
0.4
0.35
0.3
0.25
0.2
0.15 L=1000
0.1
0 100 200 300 400 500
0.4
0.35
0.3
0.25
0.2 L=10000
0.15
0.1
0 100 200 300 400 500
t
Figure 3: Plot of the global transverse magnetization mz (t) for fixed values of the quench parameter at Γ0 = 0.5 and Γ = 0.8
(within the ferromagnetic case); for each of the four panels, we fixed the size L of the system increasing it from the top to the
bottom.
Quench Energy In order to check energy conservation, let’s compute the mean value of the Hamiltonian.
X † X †
H= εk ck (t)ck (t) − c−k (t)c†−k (t) − i γk ck (t)c†−k (t) − c−k (t)ck (t) (154)
k>0 k>0
Finally, the solution for the average energy could be written as the sum of two terms:
hĤi = Eq + E(t) (155)
The former is time independent and corresponds to the total amount of energy given by the quench while the latter
is time dependent and is zero. Indeed
X
E(t) = [ǫk (− sin 2θk sin 2∆θk cos(2Ek t)) − γk (sin 2θk cos 2∆θk cos(2Ek t))] (156)
k>0
X
= cos(2Ek t) sin 2∆θk (−ǫk sin 2θk − γk cos 2θk ) = 0 (157)
| {z }
k>0
=0 f rom Eq (127)
and
X
Eq = [ǫk (− cos 2θk cos 2∆θk ) + γk sin 2θk cos 2∆θk ] (158)
k>0
X
= cos 2∆θk (γk sin 2θk − ǫk cos 2θk ) (159)
k>0
X ǫk ǫ0 + γ 2 .
k k
=− = Eq (Γ0, Γ) (160)
Ek Ek0
k>0
G. Generalized Gibbs Ensemble
As we have argued above average values of local operators in the stationary state after the quench in the TFIC are
not described by the standard thermal equilibrium ensemble. The reason, as we have mentioned, is that the TFIC is an
exactly solvable, i.e. integrable, model with an extensive set of local integral of motions. For this class of systems one
can nonetheless describe the long-time stationary properties in terms of a generalized statistical ensemble, which take
into account the extra conserved quantities. The idea is to construct an ensemble which maximizes the entropy under
24
the constraints imposed by the conserved quantities. Let’s indicate with In a set of mutually commuting integrals of
motion
[In , Im ] = 0 [In , H] = 0 (161)
Then let’s introduce the functional of density matrix ρ

X
F [ρ] = λn (TrρIn − I0n ) + Trρ log ρ (162)
n
where I0n = hΨ0 |In |Ψ0 i is the initial value of the integral of motion which is conserved, and let’s require its stationarity
with respect to ρ and λn to obtain
!
X
∗
ρ ∼ exp λn In (163)
n
Trρ∗ In = I0n (164)
This so called Generalized Gibbs ensemble (introduced firstly in the context of classical statistical mechanics by
Jaynes in the late fities37,38 and then more recently in the context of quantum quenches39 ) describes well the long-time
properties of several integrable quantum systems, in particular those that can be mapped onto quadratic Hamiltonians.
Its general validitiy is however matter of debate, that we won’t have the time to discuss in detail, mainly because as
we mentioned already, the definition of conserved charge is a bit tricky in quantum mechanics and the understanding
of which conserved quantity has to be included in the GGE description to reproduce correctly the long-time steady
state is a delicate question40 .
III. SHORT-TO-INTERMEDIATE TIME DYNAMICS: PRETHERMALIZATION AND DYNAMICAL

TRANSITIONS
Here we discuss some non trivial aspects of the short-to-intermediate time dynamics of isolated interacting quantum
systems. In certain case thermalization can be a fast process, while in other situations the approach to equilibrium can
involve multiple time scales and proceed as a two-stage process, where on intermediate time scales a quasistationary
state is reached which eventually decay away toward thermal equilibrium. Such a phenomenon has been dubbed
prethermalization 41 and appears ubiquitously in non-equilibrium dynamics of a variety of quantum systems and
recently also observed experimentally42,43 .
Another interesting effect which has been discussed extensively recently is the possibility of some sort of dynamical
critical behavior of dynamical transition in the time evolution of isolated quantum systems. Such an issue
was raised first in the context of Fermi-Hubbard model solved with Dynamical Mean Field Theory44,45 but later was
recognized to appear in many other models close to the mean field limit, including fermions within the time-dependent
Gutzwiller approximation46,47 , fully connected Ising spins in transverse field and Bose-Hubbard models48 , interacting
field theories49–51 and Fermi-Hubbard models with antiferro-magnetic correlations52,53 .
A. Prethermalization in the Fermi Hubbard Model
To discuss prethermalization let us consider the Fermi Hubbard model describing electrons hopping on a lattice
and interacting with an onsite Coulomb repulsion. Since we are interested in quench dynamics we will consider a time
dependent interaction that we will suddenly switch on,
X X 2
H =− tij c†iσ cjσ + U (t) (ni − 1) (165)
ij i
Let’s start at t = 0 with a Fermi sea and let’s ask the following basic and simple question: How does the distribution
function evolves in time after a weak perturbation of this form? This problem was studied in 2008 by Kehrein and
Moeckel54 using a real-time RG-perturbative approach and later using Keldysh perturbation theory55 .
The result for the time dependent quasiparticle weight reads
Z
sin2 (ωt/2)
Zneq (t) = 1 − 4 dω ΓkF (ω) (166)
ω2
25
where Γk (ω) is the interaction matrix element and has the Fermi-golden rule structure
X
Jk (ω) = U 2 δ(k + p = k ′ + p′ )δ(ω + εp = εk′ + εp′ ) [nk np (1 − nk′ )(1 − np′ ) − nk′ np′ (1 − nk )(1 − np )] (167)
which can also be related to the imaginary part of the self-energy for the fermions, giving their lifetime.
It turns out to be a much more general phenomenon, known under the name of prethermalization. How can we
understand this trapping into a nonthermal state after a small quantum quench? First remark. since we are talking
about weak perturbations breaking integrability one might be tempted to use real-time perturbation theory, for
example in the Keldysh framework. In order to do so one has to be careful about how to proceed. In general non
self-consistent perturbation theory might lead to singuar terms diverging in time. Why: think about a perturbative
expansion of some exponential decay e−γ(U)t with γ ∼ U 2 for small U. Then if you expand this in power of U you
would get 1 − U 2 t + .... These are trivial singularity or secular terms which can be avoided using either self-consistent
resummation schemes or as we are going to see unitary transformations56.
Let’s consider an Hamiltonian of the form
H = H0 + gH1 (168)
P
where H0 = α εα Iα is diagonal in terms of some set of integrals of motion
[Iα , Iβ ] = 0 [H, Iα ] = 0 (169)
which are not conserved charges anymore at finite g. We would like to compute the dynamics of an observable O
under the action of H, in particular we will focus on an observable that (i) we will assume to commute with all the
conserved quantities (i.e. some linear combination of them) and that further we assume (ii) to satisfy O|ψ(0)i = 0. In
the example of the Hubbard model we see that this observable could be the momentum distribution (which is indeed
conserved by H0 ) minus the average over the initial Fermi sea (such that A|Ψ(0)i = 0), i.e.
O = nk − hnk i0
In the limit of a weak perturbation, g ≪ 1, we can compute things using a unitary transformation
Let’s introduce a unitary transformation Ω = e−S generated by an antiunitary operator S † = −S that we expand
formally in power of g
g2
S = S0 + gS1 + S2 + o(g 3 ) (170)
2
The effective Hamiltonian, Hef f = Ω† HΩ = eS He−S can be also expanded in series of g
Hef f = H0 + g (H1 + [S1 , H0 ]) + g 2 () (171)
and we will require that, order by order, this Hamiltonian becomes diagonal in the same conserved quantities, i.e.
(1) (2)
Hef f = H0 + gHdiag + g 2 Hdiag + O(g 3 ) (172)
(i) P (i)
with Hdiag = n En |nihn|. Such a condition will translate into equation for the matrix elements of Si as well as for
(i)
the correction to the eigenvalues En . We get for the linear order
hn|H1 |mi
hn|S1 |mi = (1 − δnm ) (173)
En − Em
En(1) = hn|H1 |ni (174)
while for the term order g 2

(1)
hn|[S1 , H0 + Hdiag ]|mi
hn|S2 |mi = (1 − δnm ) (175)
En − Em
X |hm|H1 |ni|2
En(2) = (176)
(En − Em )
m6=n
26
Let’s now compute the time-dependent average of an operator O
hOit = hΨ(0)|eiHt Oe−iHt |Ψ(0)i (177)
and let’s insert the identity 1 = e−S eS using the fact that [Hef f , O] = 0 up to o(g 2 ), to obtain
hOit = hΨ(0)|e−S eS(t) Oe−S(t) eS |Ψ(0)i (178)
where we have defined eS(t) = eiHef f t eS e−iHef f t . We now expand the inner and outer exponential and use the Baker
Hausdorff formula
1
eS X e−S = X + [S, X] + [S, [S, X]] + ... (179)
2
to obtain to second order
1 1
hOit = hΨ(0)|O + [S(t) − S, O] − [S, [2S(t) − S, O]] + [S(t), [S(t), O]]|Ψ(0)i + .. =(180)
2 2
1 1
hOi0 + h(S(t) − S)O − O(S(t) − S)i0 − h[S, (2S(t) − S)O − O(2S(t) − S)]i0 + h[S(t), (S(t)O − OS(t)]i0 + ...
2 2
In evaluating these commutators we now use the fact that the operator O annihilates the initial state, i.e. O|Ψ(0)i = 0
to simplify the result that becomes,
1 1
hOit = h(2S(t) − S)OSi0 + hSO(2S(t) − S)i0 − hS(t)OS(t)i0 = (181)
2 2
= hS(t)OSi0 + hSOS(t)i0 − 2hSOSi0 = 2g 2 RehS1 (t)OS1 i0 − 2g 2 hS1 OS1 i0 (182)
where in the last step we used the fact that the result is second order in S, therefore to get hOit to order g 2 we can
set S = gS1 . The first term can be written as
Z
X |h0|H1 |ni|2 J(ω)
hS1 (t)OS1 i0 = e−i(En −E0 )t hn|O|ni 2
= − dω 2 e−iωt (183)
n
(En − E0 ) ω
where we have introduced the spectral density

X
J(ω) = − |h0|H1 |ni|2 hn|O|niδ(ω − (En − E0 )) (184)
n
Evaluating this result at t = 0 and combining it in Eq. REF we finally get

Z Z
J(ω) J(ω)
hOit = −2g 2 dω 2 (1 − cos(ωt) = −4g 2 dω 2 sin2 (ωt/2) (185)
ω ω
Back now to the Fermi Hubbard model
Z
J(ω)
hNk it = hNk i0 + 4g 2
dω sin2 (ωt/2) (186)
ω2
In this case we can interpret this directly for the jump and we get
Z
J(ω)
Zneq (t) = hNk=k− it − hNk=k+ it = 1 − 4g 2 dω sin2 (ωt/2) (187)
F F ω2
the result that we get for the is then the same obtained by Moeckel and Kehrein where one can show after some
algebra that the kernel J(ω) is given by nothing that the Fermi-Golden rule one. The long time limit of this quantity
is controlled by the low frequency behavior of the phase space factor, which we now from Fermi Liquid theory goes
as ω 2 at low frequency with some finite energy cutoff. If we use a form
J(ω) = θ(Λ − ω)ω 2 /Λ (188)
we obtain
Z Λ
2 2 1 2 1
Zneq (t) = 1 − 4g dω sin (ωt/2) = 1 − 4g Λt/2 − sin Λt ≃ 1 − 2g 2 + o(1/t) (189)
0 Λt 2
27
Two remarks are in order here: first, the result is clearly different from what one would get at finite temperature,
where the step would be vanishing. Quite interestingly this is not even related to zero temperature we would get
for the renomalization factor (or wave function renormalization) mismatch at the final value of the interaction. It is
instructive to compare this with the analog finite temperature expression obtained in equilibrium which would read
X |hm|H1 |ni|2
hOieq = g 2 ρ0m hn|O|ni (190)
nm
(En − Em )2
from which we would get at zero temperature

Z Λ
J(ω)
Zeq = 1 − g 2 dω (191)
0 ω2
Bottomline - the metastable fermi liquid is more correlated than its zero temperature counterpart and has still a sharp
fermi surface! More recent works have investigate the properties of this metastable Fermi Liquid for what concerns
for example its response functions (structure factor)? .
B. Dynamical Transitions in the Fully Connected Transverse Field Ising
Let’s consider our favourite Ising model in a transverse field, but let’s now consider the limit of a fully connected
model, i.e. a model where the exchange interaction involves all pairs of spins
J X x x X
H =− σi σj − Γ σiz (192)
N ij i
We can write this in terms of the components of the total spin S = (S x , S y , S z ) where we have defined
1 X α
Sα = σ (193)
2N i i
We get
4J x 2
H =− (S ) − 2ΓS z (194)
N
from which we immediately conclude that the total spin, S · S = (S x )2 + (S y )2 + (S z )2 is conserved in this fully
connected model
[H, S · S] = 0 (195)
Using the spin algebra
[S α , S β ] = iεαβγ S γ (196)
we can obtain the equations of motion of the various spin components evolving under the action of H, i.e. Ṡ α =
−i[S α , H], which read
Ṡ x = 2Γ S y (197)
4
Ṡ = − (S z S x + S x S z )
y
(198)
N
4
Ṡ z = − (S y S x + S x S y ) (199)
N
We now take the average over the initial state and use the fact that, in the thermodynamic limit which also corresponds
to the large spin limit, the spin behaves semiclassically which allows us to factorize the correlations
hS α S β i ≃ hS α ihS β i
28
Corrections to REF can be shown to vanish in the thermodynamic limit (REF). This can be seen by noticing that if
we define the reduced ("intensive") spin operators mα = S α /N , the commutation relations for these operators read
i αβγ γ
[mα , mβ ] = ε m (200)
N
,i.e. in the thermodynamic limit these objects behave as classical (commuting) variables. We then get the mean field
equations
hṠ x i = 2ΓhS y i (201)

8
hṠ y i = hS z ihS x i − 2ΓhS x i (202)
N
8
hṠ z i = − hS y ihS x i (203)
N
which can be easily written as classical equations of motion using the parametrization
N
hṠ x i = cos θ (204)
2
N
hṠ x i = sin θ sin φ (205)
2
N
hṠ z i = sin θ cos φ (206)
2
(207)
From the S z equation we immediately get
θ̇ = −2Γ sin φ (208)

(209)
while from the equations for S x , S y , by multiplying for cosφ and sin φ and subtracting we get
sin θφ̇ = 4 sin θ cos θ − 2Γ cos θ cos φ (210)
These equations conserve the total energy of the systems, which play the role of the classical effective Hamiltonian
and reads
ε(θ, φ) = − cos2 θ − Γ sin θ cos φ (211)
Anagously one can show that this amounts to consider a coherent evolution of the form

1 iϕ 1 −iϕ
|θ, ϕi = exp θe S− − θe S+ |S, M i (212)
2 2
and making stationary the action
Z
S= dthθ, φ|i∂t − H|θ, φi (213)
Using the conservation of energy we can write the problem in terms of an effective one dimensional classical
dynamics. Indeed if we define m = S x /N = cos θ/2 we get
1
ṁ = − sin θθ̇ = Γ sin θ sin φ (214)
2
and using the fact that
s 2
p ε + cos2 θ
sin φ = 1 − cos2 φ = 1− (215)
Γ sin θ
we finally get
p
ṁ = ± f (m) (216)
29
with the effective potential

2
f (m) = Γ2 (1 − 4m2 ) − ε + 4m2 (217)
We also notice that in terms of the variable m, φ the classical energy reads
p
E(m, φ) = −4m2 − Γ 1 − 4m2 cos φ (218)
Before turning into the discussion of the resulting dynamics, let’s conclude by presenting another (more rigorous)
derivation of the semiclassical dynamics where corrections 1/N are explicitly present. Let’s go back to the fully
connected TFI Hamiltonian,
4J x 2
H =− (S ) − 2ΓS z (219)
N
and let’s study the dynamics at finite N . We already mentioned that the total spin is conserved, [H, S2 ] = 0, so one
can block diagonalize the Hamiltonian in sectors of the Hilbert space at fixed S. Similarly one can study the dynamics
in these subspaces. To this extent let’s decompose the wave function in the basis states at fixed S, M , where M is
the eigenvalue of S x
S
X
|Ψ(t)i = ψM (t)|S, M i (220)
M=−S
and apply the time dependent Schroedinger equation i∂t |Ψ(t)i = H|Ψ(t)i to get
X 4J X X
ψM (t)(S x )2 |S, M i − 2Γ
i∂t ψM (t)|S, M i = − ψM (t)S z |S, M i =
N
M M M
4J X 2 X p p
=− M ψM (t)|S, M i − 2Γ ψM (t) (S − M )(S + M + 1)|S, M + 1i + (S + M )(S − M + 1)|S, M − 1i (221) =
N
M M
If we now project back on the state |S, M ′ i we get an equation for the amplitudes
4JM 2 Xp
i∂t ψM (t) = − ψM (t) − 2Γ S(S + 1) − M (M + α)ψM+α (t) (222)
N α=±
We now introduce the reduced magnetization m = 2M/N and the reduced spin S = N (1 − 2k)/2 such that we can
re-write this equation as

1 p X
i∂t ψm (t) = −Jm2 ψm (t) − 2Γ (1 − 2k)2 − m2 ψm+α (t) (223)
N α
We now can use the fact that ψm+a = ea∂m ψm and finally obtain

1 p
i∂t ψm (t) = −Jm2 ψm (t) − 2Γ (1 − 2k)2 − m2 cos (∂m ) ψm (t) = Hef f ψm (t) (224)
N
which shows how the dynamics at finite N is quantum mechanical with an effective planck constant of order 1/N .
If one wants to study the limit of large N this corresponds to do a semiclassical analysis (WKB) of the quantum
problem, i.e. solving a classical dynamics for the degrees of freedom m and its conjugateφ ≡ ∂m with an Hamiltonian
p
Hef f = −Jm2 − 2Γ (1 − 2k)2 − m2 cos φ (225)
which matches the result we have obtained with the mean field decoupling.
1. Dynamical Restoration of the Symmetry
Let’s now specify the quench and the initial condition. Let’s assume to start from Γ0 = 0, corresponding to an indeal
classical Ising ferromagnet and to switch on a finite Γ. The initial values of the angles θ, φ therefore corresponds to
30
θ = 0, with arbitrary φ, and the conserved energy reads ε = −1. Then one can easily show that the effective potential
has, for Γ < 1 four roots
r
1 − Γ2
meq = ±1/2 m∗ = ± (226)
2
where meq = 1/2 is the initial equilibrium condition which is always a stationary point of the dynamics. The motion
of this one dimensional problem is confined in the regions which are classically allowed, corresponding to f (m) > 0
which are nothing but the regions [m− , 1/2]. The motion is periodic and one can easily compute the period of these
oscillations
Z 1/2
dm
T = q (227)
m− 16 (1/4 − m2 ) (m2 − m2− )
As the transverse field approaches the value Γdyn

c = 1 the root m− → 0 and the dynamics changes qualitatively. Right
at the critical point the order parameter relaxes to zero exponentially
Z 1/2
dm
dt = p → m(t) ∼ exp(−γt) (228)
0 m 1/4 − x2
Above the threshold, for Γ > 1, the shape of the effective potential changes since now only two roots are present,
namely meq = ±1/2 which always remain a stationary point. In other words in this regime the order parameter
oscillates freely around zero, i.e. the symmetry is dynamically broken.
It is interesting to notice the relation between this dynamical critical point and the quantum phase transition
in equilibrium. Indeed if one solves for the model at T = 0 using the same mean field decoupling one finds the
self-consistent equation
2Jm p
m= √ tanh β 16J 2 m2 + Γ2 (229)
16J 2 m2 + Γ2
which has a quantum critical point at Γc = 2J. Therefore the dynamical transition, corresponding to a restoration of
the symmetry, happens well within the ordered phase.
C. The Role of Dephasing in the Dynamical Transition: Quenches in the AFM Hubbard Model
We have seen that at the mean field level the dynamical transition is sharp but lacks for example some basic feature
like the effect of damping. Now we discuss how one can improve this result by including some sort of fluctuations. To
see this point it is useful to go back again to the Hubbard model
X X 2
H=− tij c†iσ cjσ + U (ni − 1) (230)
ij i
While earlier we considered the Fermi Liquid regime, where the state remains fully symmetric, the HM has a wide range
of broken symmetry phases. We focus on the more conventional one, the antiferromagnetic (AFM) pattern, which
typically arises in hypercubic lattice (cubic lattice in d dimension). The critical temperature has the characteristic
bell shape, starting small at weak coupling (BCS-like) and going as J ∼ t2ij /U at large interaction.
We are now going to study quenches in this weak coupling regime using time-dependent Hartree Fock. The idea is,
as previously, to decouple the interaction with a time-dependent and spin dependent average density profile, i.e.
1
hniσ it = + m(t)σ(−1)i (231)
2
i.e. in other words the order parameter reads
hni↑ it − hni↓ it = (−1)i m(t) (232)
The resulting Hamiltonian reads

X X X X †
Hef f = − tij c†iσ cjσ − m(t)U (−1)i (ni↑ − ni↓ ) = εkσ c†kσ ckσ − U m(t) ck↑ ck+Q↑ − c†k↓ ck+Q↓ (233)
ij i kσ k
31
where we have used the fact that eiQRi = (−1)i and we have to satisfy the self-consistency condition
X
m(t) = (−1)i [hni↑ it − hni↓ it ] (234)
i
(to complete: derive equations of motion for the momentum distributions and show they admit an extensive set of
conserved quantities, basically the lenght of a pseudospin).
IV. PERIODICALLY DRIVEN ISOLATED QUANTUM SYSTEMS
A. Intro to Floquet Theory - First formulation
Let’s now discuss a first formulation of Floquet theory. Let us consider an isolated quantum system which is
periodically driven, such that its Hamiltonian is periodic
H(t) = H(t + T )
According to the Floquet theorem the time-dependent Schroedinger solution admits a complet set of (quasi-)periodic
solutions (Floquet Basis)
|Ψα (t)i = e−iεα t |Φα (t)i (235)
where the Floquet mode is periodic |Φα (t + T )i = |Φα (t)i and εα are the so called Floquet quasi-energies. To see this
we can apply the operator (H(t) − i∂t ) to a given Floquet mode to get
(H(t) − i∂t ) |Φα (t)i = εα |Φα (t)i + eiεα t (H − i∂t ) |Ψα i (236)
from which it immediately follows that |Ψα i is solution provided we impose
(H(t) − i∂t ) |Φα (t)i = εα |Φα (t)i (237)
Given a Floquet mode |Φα (t)i one can obtain another one by a phase transformation
|Φ̃α (t)i = e−inΩt |Φα (t)i (238)
this new mode satisfy the same Schroedinger equation with quasi-energy
εα − nΩ (239)
As a result the quasi-energies are defined modulo an integer multiple of Ω which means we can restrict ourselves to
the interval [−Ω/2, Ω/2]. Such a restriction is completely similar to the Bloch theorem for quasi-periodic real-space
wave functions.
Since the Floquet modes are periodic it is useful to expand them in harmonics
X
|Φα (t)i = e−inΩt |Φnα i (240)
n
and to write the eigenvalue equation for the quasi-energies in this basis to get firts
X X
(H(t) − nΩ) |Φnα i−inΩt = εα |Φnα i−inΩt (241)
n n
then if one introduces a proper scalar product and use the ortogonality condition
Z
1 T
dt exp(i(m − n)t) = δm,n (242)
T 0
finally ends up with the result
X
(Hmn − δmn mΩ) |Φnα i = εα |Φnα i (243)
n
with
Z T
1
Hmn = dt ei(m−n)Ωt H(t) (244)
T 0
The previous equation takes the form of a conventional eigenvalue equation in an enlarged Hilbert space that now
contains the space of harmonics as well.
32
B. Floquet Theory, second formulation
We now further elaborate on the Floquet theory, considering it now in the Heisenberg rather than Schroedinger
picture, where it will take the form of a specific ansatz on the evolution operator. For the sake of simplicity we assume
to switch on the periodic driving at t = 0 and to consider the evolution operator up to time t which is of the form
Z t
U (t, 0) = T exp −i dt′ H(t′ ) (245)
0
and satisfies as well the SE, i∂t U (t) = H(t)U (t) with initial condition U (0, 0) = 1. A general property of the periodic
evolution operator is that if one evaluates it at times which are multiples of the period, t = kT , then one can easily
prove
k
U (kT, 0) = (U (T, 0)) (246)
this result follows immediately from the definition above. Indeed we have
Z ! k−1
!
kT X Z (n+1)T
U (kT, 0) = T exp −i dt′ H(t′ ) = T exp −i dt′ H(t′ ) = (247)
0 n=0 nT
k−1
! k−1 Z
!
X Z (n+1)T X T
′ ′ ′′ ′′
= T exp −i dt H(t ) = T exp −i dt H(t + nT ) = (248)
n=0 nT n=0 0
k−1
! !
XZ T Y Z T Y k
′′ ′′ ′′ ′′
T exp −i dt H(t ) =T exp −i dt H(t ) = U (T, 0) = (U (T, 0)) (249)
n=0 0 n 0 n
This result is very suggestive, since it essentially says that if one observe the system stroboscopically, that is at integer
multiples k of the fundamental period, the evolution operator is equivalent to applying k times the same operator
U (T, 0) which acts therefore as it was a time independent Hamiltonian. In other words this suggests to define a
time-independent Floquet Hamiltonian as
U (T, 0) ≡ e−iHF T (250)
But there’s more to the role of the evolution in a single cycle. If one takes the Schroedinger equation with a time-
periodic Hamiltonian it is easy to see that if |Ψα (t)i is a solution, then also |Ψα (t + T )i it is a solution, since
i∂t |Ψα (t + T )i = H(t)|Ψα (t + T )i → (251)

→ i∂t+T |Ψα (t + T )i = H(t + T )Ψα (t + T )i (252)
Therefore there should be an operator which connects these two solutions. It is easy to see using the semigroup
properties of the unitary operator that
|Ψα (t + T )i = U (t + T, 0)|Ψα (0)i = U (t + T, 0)U † (t, 0)U (t, 0)|Ψα (0)i = U (t + T, t)|Ψα (t)i (253)
namely the two solutions at any time t are connected by the evolution operator during a cycle T . If we plug now the
Floquet ansatz for the solution of the t.d. Schroedinger equation we have
U (t + T, t)|Ψα (t)i = U (t + T, t)|Φα (t)ie−iεα t = (254)

−iεα (t+T ) −iεα (t+T )
e |Φα (t + T )i = e |Φα (t)i (255)
from which we conclude that the Floquet modes are the eigenstates of the evolution operators during a period T with
the eigenvalues given by e−iεα T , i.e.
U (t + T, t)|Φα (t)i = e−iεα (t+T ) |Φα (t)i (256)
which suggests the decomposition

X
U (t + T, t) = e−iεα T |Φα (t)ihΦα (t)| (257)
α
33
which evaluated at time t = 0 gives

X
U (T, 0) = e−iHF T = e−iεα T |Φα (0)ihΦα (0)| (258)
α
In addition, one can obtain for the evolution operator at time t within the first cycle the result
X
U (t, 0) = e−iεα t |Φα (t)ihΦα (0)| (259)
α
which can be easily verified by inspection. Indeed if we take the time derivative of both sides we get
X X X
i∂t U (t, 0) = εα e−iεα t |Φα (t)ihΦα (0)| + e−iεα t i∂t |Φα (t)ihΦα (0)| = H e−iεα t |Φα (t)ihΦα (0)| (260)
α α α
From this result we see that Floquet eigenmodes and quasi-energies play for HF a very similar role to the standard
eigenstates and eigenvalues of a static Hamiltonian in terms of a natural basis onto which one can expand any initial
state and study easily the stroboscopic dynamics.The above results also suggest practical numerical approaches to
compute solve the Floquet problem.
One possibility involves computing the evolution operator up to time T , solving the equation of motion in matrix
form from t = 0 up to T and use Eq. (REF) to diagonalize it and find the Floquet modes at time t = 0 and the
quasi-energies. Then one can just propagate these states to get them in the full first cycle. Otherwise one can solve
for the evolution operator in the full first cycle U (t, 0) and then diagonalize it (i.e. get right and left eigenvectors) to
obtain directly the Floquet modes over the first cycle.
C. Floquet Approach and the Transformation to Rotating Frame
Let’s start from some very general consideration about time-dependent unitary transformations. Let’s assume to
have a system satisfying the td Schroedinger equation
i∂t |Ψ(t)i = H(t)|Ψ(t)i (261)
Let’s introduce a time-dependent unitary transformation
|Ψ(t)i = Ω(t)|Φ(t)i (262)
and we would like to ask what kind of equation of motion is satisfied by the state |Φ(t)i. This can be easily found by
computing the time-derivative of the state |Φ(t)i = Ω† (t)|Ψ(t)i to get

i∂t |Φ(t)i = i∂t Ω† |Ψ(t)i + Ω† i∂t |Ψ(t)i = i∂t Ω† Ω + Ω† HΩ |Φ(t)i (263)
from which we get that the transformed Hamiltonian reads

H̃ = i∂t Ω† Ω + Ω† HΩ (264)
It is also interesting to define a transformed evolution operator, such that
Ũ (t, 0) = Ω† (t)U (t, 0)Ω(0) (265)
which can be easily shown to satisfy the equation
i∂t Ũ (t, 0) = H̃ Ũ (t, 0) (266)
as one can immediately see from inspection and using the definition of Ũ (t, 0).
Let’s now consider the following periodic Hamiltonian
λ
H = H0 (S 2 , Sz ) + λ (Sx cos ωt − Sy sin ωt) = H0 (S 2 , Sz ) + S+ eiωt + S− e−iωt (267)
2
describing a quantum spin S, with components Sx,y,z driven by circularly polarized light. Here H0 includes the
rotationally invariant part of the Hamiltonian as well as the part diagonal in Sz . Let’s now consider a time-dependent
unitary transformation generated by Sz of the form
Ω = eiSz ωt (268)
34
Under the action of this operator we know that
Ω† Sx Ω = Sx cos ωt + Sy sin ωt (269)

†
Ω Sy Ω = −Sx sin ωt + Sy cos ωt (270)
†
Ω Sz Ω = Sz (271)
as one can easily show using the spin algebra [Sα , Sβ ] = −iεαβγ Sγ . Clearly the rotation is such that the rotational
invariant part, function of S 2 = Sx2 + Sy2 + Sz2 remains constant. The rotated Hamiltonian becomes
H̃ = H0 + λ (cos ωt (Sx cos ωt + Sy sin ωt) − sin ωt (−Sx sin ωt + Sy cos ωt)) + ωSz = H0 + ωSz + λSx (272)
namely in the transformed frame the problem becomes time-dependent. The evolution operator in the rotated frame
is therefore just Ũ (t) = e−iH̃t , and when transformed back we get
U (t, 0) = Ω(t)e−iH̃t Ω† (0) = eiSz ωt e−iH̃t (273)
which assumes the Floquet form of a periodic operator times an evolution with a static (time-independent) Hamilto-
nian.
We can cast the Floquet approach in this language, which is another angle to the Floquet problem. Let’s consider
a periodic Hamiltonian H(t + T ) = H(t) and let’s look for a unitary "gauge" transformation
Ω(t) = eiF (t) (274)
generated by some periodic hermitian operator F (t + T ) = F (t) such that in the new gauge the transformed Hamil-
tonian i.e.

H̃ = i∂t e−iF eiF + e−iF HeiF (275)
is time indpendent. If we assume this operator F exists and if therefore H̃ is time-independent, it is immediate to
conclude that the Floquet theorem holds. Indeed let’s then call the eigenfunctions of the new Hamiltonian H̃ with
|Φα (0)i, their dynamics induced by the effective Hamiltonian is just a trivial phase with an eigenvalue εα
e−iH̃t |Φα (0)i = e−iεα t |Φα (0)i (276)
while in the original frame the time evolution corresponds to the action of the operator eiF , i.e. in other words
|Ψα (t)i = eiF (t) e−iεα t |Φα (0)i = e−iεα t |Φα (t)i (277)
where we can now identify the Floquet state with the periodic function eiF (t) |Φα (0)i. Interesting we have that the
unitary operator in the original frame reads
U (t, 0) = eiF (t) e−iH̃t e−iF (0) (278)
1
L. D. Landau and E. M. Lifshitz, Statistical Physics, Part I (1995).
2
S. K. Ma, Statistical Mechanics (World Scientific, Singapore, 1985).
3
R. P. Feynmann, Statistical Mechanics: A Set of Lectures (Westview Press, 1998).
4
A. L. Fetter and J. D. Walecka, Quantum Theory of Many Particle Systems (Dover, 2003).
5
L. V. Keldysh, Sov. Phys. JETP 20, 1018 (1964).
6
L. P. Kadanoff and G. Baym, Quantum Statistical Mechanics (Benjamin, New York, 1962).
7
J. Schwinger, Journal of Mathematical Physics 2, 407 (1961).
8
I. Bloch, J. Dalibard, and W. Zwerger, Rev. Mod. Phys. 80, 885 (2008).
9
T. Langen, R. Geiger, and J. Schmiedmayer, Annual Review of Condensed Matter Physics 6, 201 (2015),
http://dx.doi.org/10.1146/annurev-conmatphys-031214-014548.
10
B. Rauer, T. Schweigler, T. Langen, and J. Schmiedmayer, ArXiv e-prints (2015), 1504.04288.
11
E. Altman, ArXiv e-prints (2015), 1512.00870.
12
I. Carusotto and C. Ciuti, Rev. Mod. Phys. 85, 299 (2013).
13
H. Deng, H. Haug, and Y. Yamamoto, Rev. Mod. Phys. 82, 1489 (2010).
35
14
A. Tomadin and R. Fazio, Journal of the Optical Society of America B 27, A130 (2010).
15
A. A. Houck, H. E. Türeci, and J. Koch, Nat. Phys. 8, 292 (2012), ISSN 1745-2473, URL http://www.nature.com/nphys/
journal/v8/n4/abs/nphys2251.html.
16
K. Le Hur, L. Henriet, A. Petrescu, K. Plekhanov, G. Roux, and M. Schiró, ArXiv e-prints (2015), 1505.00167.
17
H. Ritsch, P. Domokos, F. Brennecke, and T. Esslinger, Rev. Mod. Phys. 85, 553 (2013).
18
P. Calabrese and J. Cardy, Phys. Rev. Lett. 96, 136801 (2006).
19
A. Silva, Phys. Rev. Lett. 101, 120603 (2008).
20
J. Kurchan, eprint arXiv:cond-mat/0007360 (2000), cond-mat/0007360.
21
M. Campisi, P. Hänggi, and P. Talkner, Rev. Mod. Phys. 83, 771 (2011).
22
C. Jarzynksi, Phys. Rev. Lett. 78, 120603 (1997).
23
G. E. Crooks, Phys. Rev. E 60, 2721 (1999).
24
M. Srednicki, Phys. Rev. E 50, 888 (1994).
25
J. M. Deutsch, Phys. Rev. A 43, 2046 (1991).
26
L. D’Alessio, Y. Kafri, A. Polkovnikov, and M. Rigol, ArXiv e-prints (2015), 1509.06411.
27
M. Rigol, V. Dunjko, and M. Olshanii, Nature 452 (2008).
28
G. Biroli, C. Kollath, and A. M. Läuchli, Phys. Rev. Lett. 105, 250401 (2010).
29
H. Kim, T. N. Ikeda, and D. A. Huse, Phys. Rev. E 90, 052105 (2014).
30
S. Sachdev, Quantum Phase Transitions (Cambridge University Press, 1998).
31
D. Rossini, A. Silva, G. Mussardo, and G. E. Santoro, Phys. Rev. Lett. 102, 127204 (2009).
32
P. Calabrese, F. H. L. Essler, and M. Fagotti, Phys. Rev. Lett. 106, 227203 (2011).
33
L. Foini, L. F. Cugliandolo, and A. Gambassi, Phys. Rev. B 84, 212404 (2011).
34
L. Foini, L. F. Cugliandolo, and A. Gambassi, Journal of Statistical Mechanics: Theory and Experiment 2012, P09011
(2012).
35
P. Calabrese, F. H. L. Essler, and M. Fagotti, Journal of Statistical Mechanics: Theory and Experiment 2012, P07016
(2012).
36
P. Calabrese, F. H. L. Essler, and M. Fagotti, Journal of Statistical Mechanics: Theory and Experiment 2012, P07022
(2012).
37
E. T. Jaynes, Phys. Rev. 106, 620 (1957).
38
E. T. Jaynes, Phys. Rev. 108, 171 (1957).
39
M. Rigol, V. Dunjko, V. Yurovsky, and M. Olshanii, Phys. Rev. Lett. 98, 050405 (2007).
40
E. Ilievski, J. De Nardis, B. Wouters, J.-S. Caux, F. H. L. Essler, and T. Prosen, Phys. Rev. Lett. 115, 157201 (2015).
41
J. Berges, S. Borsányi, and C. Wetterich, Phys. Rev. Lett. 93, 142002 (2004).
42
T. Kitagawa, A. Imambekov, J. Schmiedmayer, and E. Demler, New Journal of Physics 13, 073018 (2011).
43
M. Gring, M. Kuhnert, T. Langen, T. Kitagawa, B. Rauer, M. Schreitl, I. Mazets, D. A. Smith, E. Demler, and J. Schmied-
mayer, Science 337, 1318 (2012).
44
M. Eckstein, M. Kollar, and P. Werner, Phys. Rev. Lett. 103, 056403 (2009).
45
H. Aoki, N. Tsuji, M. Eckstein, M. Kollar, T. Oka, and P. Werner, Rev. Mod. Phys. 86, 779 (2014).
46
M. Schiró and M. Fabrizio, Phys. Rev. Lett 105, 076401 (2010).
47
M. Schiró and M. Fabrizio, Phys. Rev. B 83, 165105 (2011).
48
B. Sciolla and G. Biroli, Phys. Rev. Lett 105, 220401 (2010).
49
B. Sciolla and G. Biroli, Phys. Rev. B 88, 201110 (2013), URL http://link.aps.org/doi/10.1103/PhysRevB.88.201110.
50
P. Smacchia, M. Knap, E. Demler, and A. Silva, Phys. Rev. B 91, 205136 (2015).
51
A. Maraga, A. Chiocchetta, A. Mitra, and A. Gambassi, Phys. Rev. E 92, 042151 (2015).
52
N. Tsuji, M. Eckstein, and P. Werner, Phys. Rev. Lett. 110, 136404 (2013).
53
M. Sandri and M. Fabrizio, Phys. Rev. B 88, 165113 (2013).
54
M. Moeckel and S. Kehrein, Phys. Rev. Lett. 100, 175702 (2008).
55
M. Moeckel and S. Kehrein, New Journal of Physics 12, 055016 (2010).
56
M. Kollar, F. A. Wolf, and M. Eckstein, Phys. Rev. B 84, 054304 (2011).
57
http://www.nature.com/nphys/insight/non-equilibrium-physics/index.html
58
What should be called a non trivial conserved quantity is already quite tricky, since for example all powers of the Hamiltonian
H k will be conserved

Quench Good Review

Uploaded by

Copyright:

Available Formats

Quench Good Review

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Quench Good Review

Uploaded by

Copyright:

Available Formats

Dynamics of Quantum Many Body Systems Far From Thermal Equilibrium – DRAFT

I. Introduction, Motivations and Physical Examples 1

II. Dynamics of Closed Quantum Systems: Intro 7

III. Short-to-Intermediate Time Dynamics: Prethermalization and Dynamical Transitions 24

IV. Periodically Driven Isolated Quantum Systems 31

I. INTRODUCTION, MOTIVATIONS AND PHYSICAL EXAMPLES

A. Falling Gently Out of Equilibrium

CAB (t, t′ ) = Tr (ρeq A(t)B(t′ )) = Tr (ρeq A(t − t′ )B(0)) = CAB (t − t′ ) (5)

where the linear response or retarded susceptibility is defined as

CBA (−ω) = e−βω CAB (ω) (10)

to obtain the so called Fluctuation-Dissipation theorem

and obtaines the fermionic version of FDT relation

CAB (τ ) = −hTτ A(τ )B(0)i A(τ ) = eHτ A e−Hτ

H(t) = H0 + e−|ε|t Hint (15)

Then one typically focus on multi-point correlations functions of the form

G(t1 , · · · , tn ) = hΨ|T (A(t1 ) · · · A(tn )) |Ψi (16)

G(t1 , · · · , tn ) = h+∞|S[+∞, −∞]T (A(t1 ) · · · A(tn )) | − ∞i (17)

| + ∞i = eiδ |∞i eiδ = h−∞| + ∞i (18)

B. Why Quantum Nonequilibrium?

C. Outline of the course

D. General Readings (to be completed)

II. DYNAMICS OF CLOSED QUANTUM SYSTEMS: INTRO

A. Quantum Dynamics, Generalities

E(t) = hΨ(t)|H(t)|Ψ(t)i = hΨ(0)|U † (t) H(t) U (t)|Ψ(0)i (29)

1. The simplest nonequilibrium protocol: a "Quantum Quench"

H0i |φin i = Eni |φin i (31)

We notice that P (W ) has a threshold, i.e.

hW i = EQ − E0i = E0 − E0i + (EQ − E0 ) = δE + Wexc (33)

B. Time-Independent Hamiltonian, Isolated Systems

|Ψ(t)i = e−iHt |Ψ(0)i (40)

and the energy is conserved

E(t) = hΨ(0)|H|Ψ(0)i ≡ EQ (41)

H|Φn i = En |Φn i (42)

Then it is useful to decompose the initial state in this basis to obtain

Then we can write the (conserved) energy

as well as the energy fluctuation

hIit = hΨ(t)|I|Ψ(t)i = I0 (49)

Λnm (t) = ei(En −Em )t Λnm (50)

as well as the spectral density

C. Quantum Thermalization, Eigenstate Thermalization Hypothesis

where we have introduced the equilibrium micro-canonical (MC) ensemble average

1. More on ETH: Origin and Consequences

Rnm = Rmn → fO (Ē, −ω) = fO (Ē, ω) (63)

Then for the matrix element

Let’s average and let’s assume

therefore we can say

COn (t) = hn|O(t)O(0)|ni − hn|O(t)|nihn|O(0)|ni (77)

Using ETH ansatz for Onm we get

where we have defined

Finally we can expand the integrand for small Ω, i.e.

with β = ∂S/∂E to get the result quoted in Eq (REF).

Similarly we can obtain, using

D. Dephasing after a Sudden Quench: The example of Quantum Ising Chain

1. Diagonalization by Jordan-Wigner Transformation

[σiα , σjβ ] = 2iεαβγ δij σjγ (108)

σjz = 1 − 2c†j cj (111)

where the so called string operator Kj , defined as

σjz = 1 − 2σj+ σj− = 1 − 2c†j cj (113)

where the set of momenta k is given by