in driven quantum systems
Zur Erlangung des akademischen Grades eines
Doktors der Naturwissenschaften
der MathematischNaturwissenschaftlichen Fakult¨ at
der Universit¨ at Augsburg vorgelegte
Dissertation
von
Dipl.Phys. Sigmund Kohler
aus
Ehingen (Donau)
Augsburg, im Februar 1999
Erster Berichter: Prof. Dr. Peter H¨ anggi
Zweiter Berichter: Prof. Dr. Thomas Dittrich
Tag der m¨ undlichen Pr¨ ufung: 5. M¨ arz 1999
Contents
1 Introduction 1
2 Driven quantum systems and Floquet theory 5
2.1 Discrete timetranslation and Floquet ansatz . . . . . . . . . . . . . . 5
2.2 Composite Hilbert space . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 Properties of Floquet states . . . . . . . . . . . . . . . . . . . . . . . 9
2.4 The propagator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.5 Numerical computation of Floquet states . . . . . . . . . . . . . . . . 11
2.5.1 Floquetmatrix methods . . . . . . . . . . . . . . . . . . . . . 12
2.5.2 Propagator methods . . . . . . . . . . . . . . . . . . . . . . . 12
3 Quantum dissipation and Markov approximation 15
3.1 The systembath model . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.2 Quantum Langevin equation . . . . . . . . . . . . . . . . . . . . . . . 16
3.3 Inﬂuence functional . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.4 Markovian master equation . . . . . . . . . . . . . . . . . . . . . . . 20
4 Driving and dissipation: FloquetMarkov theory 23
4.1 Simple inclusion of the driving . . . . . . . . . . . . . . . . . . . . . . 23
4.2 An improved Markovian master equation . . . . . . . . . . . . . . . . 24
4.3 Decomposition into Floquet basis . . . . . . . . . . . . . . . . . . . . 25
4.3.1 Matrix elements . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.3.2 Rotatingwave approximation . . . . . . . . . . . . . . . . . . 27
4.4 The dissipative quantum map and its numerical implementation . . . 28
5 The parametrically driven harmonic oscillator 31
5.1 The model and its classical dynamics . . . . . . . . . . . . . . . . . . 31
5.2 Floquet states in stable regimes . . . . . . . . . . . . . . . . . . . . . 34
5.3 FloquetMarkov description in full RWA . . . . . . . . . . . . . . . . 36
5.4 Basisindependent description beyond RWA . . . . . . . . . . . . . . 38
5.4.1 Wigner representation and FokkerPlanck equation . . . . . . 39
5.4.2 WignerFloquet solutions . . . . . . . . . . . . . . . . . . . . . 40
5.4.3 Inﬂuence of the driving on the master equation . . . . . . . . 41
5.5 Asymptotics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.5.1 The conservative limit . . . . . . . . . . . . . . . . . . . . . . 43
ii Contents
5.5.2 The hightemperature limit . . . . . . . . . . . . . . . . . . . 44
5.6 Numerical results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
5.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
6 The harmonically driven doublewell potential 51
6.1 The model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
6.1.1 Symmetries . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
6.1.2 Tunneling, driving, and dissipation . . . . . . . . . . . . . . . 55
6.1.3 The onset of chaos . . . . . . . . . . . . . . . . . . . . . . . . 56
6.2 Chaotic tunneling near singletdoublet crossings . . . . . . . . . . . . 58
6.2.1 Threelevel crossings . . . . . . . . . . . . . . . . . . . . . . . 58
6.2.2 Dissipative chaosassisted tunneling . . . . . . . . . . . . . . . 65
6.2.3 Asymptotic state . . . . . . . . . . . . . . . . . . . . . . . . . 67
6.3 Signatures of chaos in the asymptotic state . . . . . . . . . . . . . . . 72
6.3.1 Classical attractor . . . . . . . . . . . . . . . . . . . . . . . . 72
6.3.2 Quantum attractor . . . . . . . . . . . . . . . . . . . . . . . . 75
7 Summary and outlook 79
A The harmonic oscillator 81
A.1 Number states as a basis set . . . . . . . . . . . . . . . . . . . . . . . 82
A.2 Coherent states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
A.3 Quasiprobabilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
A.3.1 Wigner function . . . . . . . . . . . . . . . . . . . . . . . . . . 84
A.3.2 Husimi function and Wehrl entropy . . . . . . . . . . . . . . . 85
B The density operator 87
B.1 Lindblad form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
B.2 Coherence and entropy . . . . . . . . . . . . . . . . . . . . . . . . . . 88
C Solution of the FokkerPlanck equation 89
References 93
Acknowledgment 101
1 Introduction
The interplay of classical chaos and dissipation in a quantum system bears inter
esting eﬀects at the border between classical and quantum mechanics like, e.g., the
suppression of classical chaos by quantum interference [1] or its restauration by dissi
pation [2]. While the mutual inﬂuence of quantum coherence and classical chaos has
been an extensive ﬁeld of research since many years, the additional eﬀects caused
by coupling the chaotic system to an environment, namely dissipation and decoher
ence, have been studied only rarely. A reason may be the fact that by including
dissipation, the computational eﬀort grows drastically, since one has to deal with
density matrices instead of wave functions.
In classical Hamiltonian systems, the transition from regular motion to chaos
is most clearly visible in the change of the phasespace structure: With increasing
nonlinearity, regular tori start to dissolve in a chaotic layer which grows in size
until it covers the whole phase space. While the motion along regular tori is stable
and predictable for long times, chaotic dynamics is characterized by a sensitive
dependence on the initial conditions: Neighboring phasespace points start to diverge
exponentially in time and a completely deterministic system evolves in a practically
diﬀusive manner on a chaotic sea [3].
On a quantum level, the positionmomentum uncertainty does not allow for
the arbitrarily ﬁne classical phasespace structures and results in coarsegraining
over an area which is given by Planck’s quantum of action. Thus, the classical
dynamics leaves in the corresponding quantum system, at most, its signatures like,
e.g., scars along unstable periodic orbits in the wave functions [4], or the centering of
Husimi functions on classical manifolds [5]. Another characteristic quantum feature
is the discreteness of the energy levels in bounded systems. In complex systems,
eigenenergies are eﬀectively random numbers whose statistical properties depend on
the integrability of the corresponding classical dynamics [6–8]. In the fully chaotic
case, the eigenenergies are anticorrelated and the inverse of their mean spacing
deﬁnes a time scale, the socalled break time, after which the quantum dynamics
becomes quasiperiodic and thus, classical chaos is suppressed [1]. This suppression of
chaos relies on the perfect coherence of a superposition which remains for arbitrary
long times. Therefore, any disruption of coherence, like it occurs due to the coupling
to an environment, restores the characteristics of classical features at least to some
extent [2].
One of the most intriguing quantum eﬀects is tunneling, the coherent transport
through a potential barrier. It was originally proposed by Hund [9] to explain
the ammonium spectrum and studied since then in various modiﬁcations. A generic
setting for the observation of tunneling is a symmetric bistable potential whose wells
are separated by a static energy barrier. A timedependent external ﬁeld acting on
2 Introduction
such a system may entail dramatic consequences for the quantum dynamics, even
if its eﬀect is barely visible in the classical phase space. Depending on the driving
amplitude and frequency, an external driving can modify the tunnel rate by orders
of magnitude or even bring tunneling to a complete standstill [10]. Tunneling is
particularly sensitive to any disruption of coherence—in presence of dissipation it
becomes a transient eﬀect that fades out on a ﬁnite time scale [11, 12].
Driving the doublewell potential with a frequency near the classical resonances
results in even more signiﬁcant consequences. They are apparent already in the
classical phase space since chaos comes into play and the separatrix which encloses
the wells is replaced by a chaotic layer. In the corresponding quantum system,
we therefore observe chaotic tunneling—coherent transport between regular islands
which are separated by a chaotic layer, rather than by a static barrier. The small
but ﬁnite overlap of the tunnel doublets with the chaotic states, i.e., with states
which are localized in the chaotic layer, typically increases the tunnel splittings
and, consequently, the tunnel rates—the essence of chaosassisted tunneling [13–15].
As soon as the chaotic layer grows in size and attains a signiﬁcant overlap with
the tunnel doublets, the tunnel splittings become of the order of the mean level
spacing [16] and tunneling is replaced by chaotic diﬀusion [16–18].
The most successful approach to dissipation in quantum mechanics, consistent
with the fundamental laws of quantum mechanics, is based on the coupling of the
conservative system to external degrees of freedom. Probably the ﬁrst proof that
such a systembath scheme results in dissipative quantum mechanics was given by
Magalinski˘ı [19] for a harmonic oscillator. Using a perturbative approach, Zwanzig
[20] derived from this model a Markovian master equation for a general classical
system subject to weak dissipation. Master equations of this kind have been applied
to various problems in solid state physics, quantum optics, and chemistry. Later,
Caldeira and Leggett eliminated the bath exactly [11, 21], which enabled studying
dissipative quantum systems, beyond a weakcoupling limit. However, even a par
tially analytical solution of the resulting pathintegral expression is only feasible
for the simplest systems, like harmonic potentials or twolevel systems—the inves
tigation of dissipative systems with complex dynamics requires to fall back to the
weakcoupling regime.
Thus, for the description of strongly driven, nonlinear systems subject to weak
dissipation, it is desirable to combine a Markovian approach to quantum dissipation,
leading to a master equation for the density operator, with the Floquet formalism
that allows to treat timeperiodic forces of arbitrary strength and frequency [22].
While the Floquet formalism is exact and essentially amounts to using an optimal
representation for the treatment of timeperiodic problems [23–25], the simpliﬁcation
brought about by the Markovian description is achieved only at the expense of
accuracy. Here, a subtle technical diﬃculty lies in the fact that the truncation of
the longtime memory introduced by the bath, and the inclusion of the driving, do
not commute. This implies that the result of the Markov approximation depends
on whether the driving is considered in its derivation or not [26, 27].
Within the present work, we will implement a Markovian approach to quan
Introduction 3
tum dissipation based on the Floquet formalism to the investigation of two diﬀerent
systems, for each of which we have, besides other interesting aspects, one central
question in mind: The parametrically driven harmonic oscillator, will serve pre
dominantly to test diﬀerent approximation schemes for the FloquetMarkov master
equation and to study the modiﬁcation of its dissipative part brought about by the
driving. For this linear system, all approximative steps can be reliably checked since
an exact solution is at hand [28]. Besides being an exactly solvable model with yet
nontrivial dynamics, this system is interesting in its own right, since it describes the
motion of an ion in a Paul trap. These traps have gained new interest very recently,
since they form the central system in a scheme for a quantum computer [29] whose
experimental realization is currently attempted. Thereby the main obstacle is, be
sides the preparation of the ground state, the loss of coherence once the computation
has started.
The harmonically driven quartic doublewell potential , a system which exhibits
complex nonlinear dynamics, will be used as a working model for the investigation
of chaotic tunneling in presence of dissipation. Recent studies of nondissipative
chaotic tunneling suggest that tunneling is accelerated by the inﬂuence of chaotic
states, replacing a doublet structure by a threelevel dynamics [30–33]. The bath,
in turn, couples these states indirectly to all other states of the system and, thus,
we expect to observe a novel dissipative tunnel scenario which is on the one hand
richer than the conservative dynamics and on the other hand substantially diﬀerent
from the familiar twostate tunneling.
This thesis is organized as follows: In Chapter 2 we give an introduction to
Floquet theory for quantum systems with periodic timedependence. A brief review
of the systembath model for quantum dissipation and a derivation of a Markovian
master equation is provided in Chapter 3 and combined with Floquet theory in
Chapter 4 to obtain a Markovian description of periodically driven quantum systems
subject to weak dissipation. Within this FloquetMarkov approach, we investigate
the dynamics of the parametrically driven harmonic oscillator and the driven double
well potential in Chapters 5 and 6, respectively. Chapter 7 serves to summarize the
main results. A number of merely technical issues is deferred to the appendix. Parts
of this thesis have already been published in Refs. [27, 34].
4
2
Driven quantum systems
and Floquet theory
Interactions of quantum systems with strong laser ﬁelds are characterized by two
properties of the ﬁeld: On the one hand, the inﬂuence of the ﬁeld on the system is
typically so strong that a treatment beyond perturbation theory becomes necessary,
but the backaction of the system on the ﬁeld is negligible. On the other hand,
the ﬁeld is in a coherent state with large mean photon number and, thus, can
be described adequately by its expectation value, given by a function harmonic
in time. This implies that an explicit time dependence of the Hamiltonian serves
as a substitute for a canonical degree of freedom and raised interest in a theory
for quantum systems with explicit periodic time dependence, thus an extension of
Floquet theory [35] from classical to quantum mechanics. Onedimensional driven
systems also play an important role as models for (quantum) chaos: Their “one and
a half degrees of freedom” represent the minimal requirement for nonintegrable
dynamics [36]. Thus, they exemplify the simplest quantum systems with chaotic
classical counterpart.
In this chapter we give an introduction to Floquet theory for quantum systems
with periodic time dependence [12, 23–25, 36, 37], where we put strong focus on the
properties of Floquet states and numerical methods which we use in subsequent
chapters.
2.1 Discrete timetranslation and Floquet ansatz
To reduce the complexity of a physical system, its symmetries are analyzed to ob
tain a proper ansatz for the symmetryreduced solutions. In quantum mechanics,
symmetry is expressed by an operator o which leaves the Schr¨ odinger equation
H(t) −i
∂
∂t
[ψ(t)` = 0 (2.1)
invariant, i.e., commutes with the operator H(t) − i∂
t
. Thus, the solutions of the
Schr¨ odinger equation are, besides a timedependent phase factor, also eigenfunctions
of the symmetry operator [38].
For a Hamiltonian with Tperiodic time dependence,
H(t) = H(t +T), T =
2π
Ω
, (2.2)
the related symmetry operation is a discrete time translation by one period of the
driving,
o
T
: t →t +T. (2.3)
6 Driven quantum systems and Floquet theory
As symmetry operations have to conserve the norm of any wavefunction, the eigen
values of o are pure phase factors and we may assume for an eigenfunction [ψ(t)`
the eigenvalue exp(−iθ), θ ∈ R,
o
T
[ψ(t)` = [ψ(t +T)` = e
−iθ
[ψ(t)`. (2.4)
By inserting this eigenvalue equation into the ansatz
[ψ(t)` = e
−it/
[φ(t)`, = θ/T, (2.5)
we obtain the condition
[φ(t)` = [φ(t +T)`, (2.6)
which means that [φ(t)` is periodic in time, alike the Hamiltonian. Thus for a
system which obeys discrete timetranslational symmetry, there exists a complete
set ¦[ψ
α
(t)`¦ of solutions of the Schr¨ odinger equation which have Floquet structure,
i.e., they are of the form
[ψ
α
(t)` = e
−iαt/
[φ
α
(t)`, (2.7)
[φ
α
(t)` = [φ
α
(t +T)`. (2.8)
However, a general solution of the Schr¨ odinger equation (2.1) is given by a super
position of many Floquet states,
[ψ(t)` =
¸
α
u
α
e
−iαt/
[φ
α
(t)`, (2.9)
and is in general not of the form (2.5). The Floquet states [φ
α
(t)` are, in contrast to
the [ψ
α
(t)`, not solutions of the Schr¨ odinger equation. The
α
have the dimension
energy and in periodically driven systems play a role analogous to the eigenenergies
in timeindependent systems. In analogy to the quasimomentum of electrons in
spatially periodic systems, they are called quasienergies. We emphasize that the
Tperiodic timedependence of the Floquet states is only relevant for the dynamics
within a period of the driving, whereas the longtime dynamics is governed by the
phase factors exp(−i
α
t/).
Inserting (2.5) into the Schr¨ odinger equation yields the eigenvalue equation for
the Floquet states [23, 39, 40]
H(t)[φ(t)` = [φ(t)` (2.10)
with the Hermitian Floquet Hamiltonian [40]
H(t) = H(t) −i
∂
∂t
. (2.11)
Technically, the determination of the Floquet states from (2.10) is one of the main
tasks in dealing with periodically timedependent systems.
2.2 Composite Hilbert space 7
From a grouptheoretical point of view, each Floquet state [φ
α
(t)` belongs to
an irreducible representation of an Abelian group, characterized by the Floquet
exponent θ
α
=
α
T/ [40]. This exponent allows for an interpretation as a Berry
phase [41].
Solutions of Floquet structure are found for dynamical systems that can be
described by diﬀerential equations with periodically timedependent coeﬃcients [35,
42]. We also use this fact for the solution of classical equations of motion and for the
solution of FokkerPlanck equations in subsequent chapters. In these cases, however,
the eigenvalue equation which corresponds to (2.10) is in general nonHermitian, thus
the Floquet indices may be complex.
2.2 Composite Hilbert space
The state [ψ(t)` of a system, as well as the Floquet states [φ
α
(t)`, are elements of
a Hilbert space 1, which describes the system’s degrees of freedom. For a bounded
particle moving in a potential, 1 is the space of squareintegrable functions [43]. In
many cases, 1 can be approximated by a Hilbert space with ﬁnite dimension.
It is possible to describe the time dependence of the Floquet states within the
framework of a Hilbert space theory. According to (2.8), the Floquet states are
elements of the space of Tperiodic functions, denoted by T [40]. An inner product
on T is deﬁned by
(f, g) =
1
T
T
0
dt f
∗
(t) g(t), (2.12)
and a set of orthonormalized basis functions reads [43]
ϕ
n
(t) = e
−inΩt
, Ω =
2π
T
, n ∈ Z. (2.13)
For a basis independent notation, we deﬁne the vectors [n`
T
by
ϕ
n
(t) = 't[n`
T
. (2.14)
To avoid confusion with elements of conﬁguration space 1, we mark these vectors
by an index T . The basis set ¦ϕ
n
¦ is orthonormalized and complete [43],
(ϕ
n
, ϕ
n
) = δ
n,n
, (2.15)
1
T
¸
n
ϕ
∗
n
(t) ϕ
n
(t
) = δ
T
(t −t
), (2.16)
where δ
T
denotes the Tperiodic delta function.
We combine the periodic time dependence of the Floquet states with their spatial
degrees of freedom and interpret them as elements of a composite Hilbert space
1⊗T . The inner product (2.12) is extended accordingly,
''φ[φ
`` =
1
T
T
0
dt 'φ(t)[φ
(t)`. (2.17)
8 Driven quantum systems and Floquet theory
The elements of this composite Hilbert space, written in “time representation,” are
Tperiodic states,
't[φ`` ≡ [φ(t)` = [φ(t +T)`. (2.18)
By this introduction of a Hilbert space structure for the time dependence, we
formally traced back the computation of Floquet states to the computation of eigen
states of a timeindependent Hamiltonian with an additional degree of freedom. The
methods known for the computation of energy eigenstates of a timeindependent
Hamiltonian, like e.g., perturbation theory, can be applied accordingly [39, 40].
The decomposition of a state [φ(t)` into the set of basis functions (2.13) is equi
valent to its representation as a Fourier series,
[φ
α
(t)` =
¸
n
e
−inΩt
[c
α,n
`, (2.19)
[c
α,n
` =
1
T
T
0
dt e
inΩt
[φ
α
(t)`. (2.20)
The Fourier modes in this context are also called Floquet channels.
Semiclassical interpretation of the Floquet states
A timedependent Hamiltonian is usually obtained from a timeindependent theory
by substituting a part of the system by its classical limit [25]. This allows for a
semiclassical interpretation of the vectors [n`
T
and the Floquet states [40]. We
restrict ourselves to the case of a linearly coupled driving ﬁeld with cosine shape.
A system S, which couples via dipole interaction to a singlemode laser with
frequency Ω, can be described by the Hamiltonian [44]
H = H
S
+µx(a +a
+
) +Ωa
+
a. (2.21)
We assume in the semiclassical limit that the state of the laser ﬁeld is a coherent
one (see Appendix A) and that it possesses a very high mean photon number,
[z` = [
√
n
0
exp(iΩt)` , n
0
1. (2.22)
Under this condition, the description of the system can be simpliﬁed in two ways:
1. We replace the operators a and a
+
by their expectation values (see Ap
pendix A) and obtain a driven system with a timedependent Hamiltonian.
The corresponding Floquet Hamiltonian reads
H = H
S
+ 2µx
√
n
0
cos(Ωt) +Ωn
0
−i∂
t
, (2.23)
decomposed into the basis set ¦[n`
T
¦,
H
n,n
= H
S
δ
n,n
+µx
√
n
0
(δ
n,n
+1
+δ
n,n
−1
) +Ω(n
0
−n)δ
n,n
. (2.24)
2.3 Properties of Floquet states 9
2. We decompose the Hamiltonian H, whose eigenfunctions are the socalled
dressed states, into the number states (A.10) of the laser mode to obtain
H
n,n
= H
S
δ
n,n
+µx
√
n + 1 δ
n,n
+1
+
√
nδ
n,n
−1
+Ωnδ
n,n
. (2.25)
If the state of the laser ﬁeld is the highly exited coherent state (2.22), we get
relevant contributions only for n ≈ n
0
1. The prefactors
√
n and
√
n + 1
in this limit become
√
n
0
+O(n
−1/2
0
).
The Floquet Hamiltonian (2.24) agrees—besides a shift in the index—with the
Hamiltonian (2.25). Therefore the basis states [n`
T
allow for an interpretation as
the semiclassical limit of the number states of the laser ﬁeld and the Floquet states
as the semiclassical limit of the dressed states.
2.3 Properties of Floquet states
Equivalent representations
Assuming that [φ(t)` is an eigenvector of H(t) with eigenvalue ,
H(t) [φ(t)` = [φ(t)`, (2.26)
the state
[φ
(n)
(t)` = e
inΩt
[φ(t)` (2.27)
obeys
H(t) [φ
(n)
(t)` = (H(t) +∂
t
) e
inΩt
[φ(t)` (2.28)
= ( +nΩ) e
inΩt
[φ(t)`. (2.29)
This means that [φ
(n)
(t)` is also an eigenvector of the Floquet Hamiltonian H(t),
i.e., a Floquet state, but with eigenvalue
(n)
= +nΩ. (2.30)
The respective solutions of the Schr¨ odinger equation,
[ψ
(n)
(t)` = e
−i(+nΩ)t/
[φ
(n)
(t)` (2.31)
= [ψ(t)` (2.32)
are identical. Thus, there exists a class of equivalent Floquet states whose quasi
energies diﬀer only by integer multiples of Ω. They all describe the same physical
state. Therefore, it is suﬃcient to take only those Floquet states into account, whose
quasienergies lie within a single Brillouin zone ω
BZ
≤ < (ω
BZ
+ Ω).
In the following, we denote by ¦[φ
α
(t)`¦ a complete set of Floquet states with
corresponding quasienergies ¦
α
¦. They are orthonormalized with respect to the
inner product (2.17),
''φ
α
[φ
α
`` = δ
α,α
. (2.33)
10 Driven quantum systems and Floquet theory
Orthonormalization on 1
The inner product of two nonequivalent Floquet states on 1obeys the Tperiodicity
of the Floquet states and can be written as a Fourier series,
'φ
α
(t)[φ
α
(t)` =
¸
n
κ
n
e
−inΩt
. (2.34)
The Fourier coeﬃcients read
κ
n
=
1
T
T
0
dt
e
inΩt
'φ
α
(t
)[φ
α
(t
)` (2.35)
= ''φ
α
[φ
(n)
α
`` = δ
α,α
δ
n,0
, (2.36)
where the time integration has been expressed by the inner product (2.17). Thus,
we get
'φ
α
(t)[φ
α
(t)` = δ
α,α
. (2.37)
This means that from the orthonormalization of the Floquet states with respect to
the inner product (2.17) on 1 ⊗ T we obtain orthonormalization with respect to
the inner product on 1 at equal times. Here however, caution is appropriate: The
orthonormalization on 1 is in general only valid for equal times and is in particular
not valid for the Fourier components (2.20).
Mean energy
Due to the Brillouinzone structure (2.30), quasienergies do not allow for global
ordering. The instantaneous energies
E
α
(t) = 'ψ
α
(t)[H(t)[ψ
α
(t)` (2.38)
= 'φ
α
(t)[H(t)[φ
α
(t)` (2.39)
do not either, since they vary with time. A quantity that is deﬁned on the full real
axis and therefore does allow for a complete ordering is the mean energy [12, 23–25]
E
α
=
1
T
T
0
dt E
α
(t) (2.40)
=
α
+ i''φ
α
[
∂
∂t
[φ
α
``, (2.41)
which results from averaging over one period of the driving. By use of the Fourier
representation (2.19) we obtain
E
α
=
¸
n
(
α
+nΩ)'c
α,n
[c
α,n
` (2.42)
Thus the nth Floquet channel gives a contribution
α
+nΩ, weighted by the squared
modulus 'c
α,n
[c
α,n
` of the corresponding Fourier coeﬃcient.
2.4 The propagator 11
2.4 The propagator
The time evolution of a quantum system can be written by use of a unitary operator
U(t, t
), which is a solution of the Schr¨ odinger equation,
i
∂
∂t
U(t, t
) = H(t) U(t, t
), (2.43)
U(t, t) = 1. (2.44)
A formal integration yields
U(t, t
) = Texp
−
i
t
t
dt
H(t
)
, (2.45)
where T denotes time ordering. Due to the time dependence of the Hamiltonian,
U(t, t
) depends explicitly on both times t and t
, not only on their diﬀerence.
Expressed in terms of the Floquet states, the propagator reads
U(t, t
) =
¸
α
e
−iα(t−t
)/
[φ
α
(t)`'φ
α
(t
)[, (2.46)
as this expression obviously solves the Schr¨ odinger equation and the initial condition
(2.44) is ensured by the completeness of the Floquet states.
The propagator U(T, 0) deﬁnes a quantum map for the propagation over a full
period of the driving,
U(T, 0) =
¸
α
e
−iαT/
[φ
α
(0)`'φ
α
(0)[, (2.47)
U(nT, 0) =
¸
α
e
−inαT/
[φ
α
(0)`'φ
α
(0)[ (2.48)
= [U(T, 0)]
n
, (2.49)
To obtain the last line, we used the Tperiodicity of the Floquet states and their com
pleteness and orthogonality at equal times. The propagator U(T, 0) is indispensable
for the investigation of the longtime dynamics of driven quantum systems [23, 25].
The Floquet states at time t are instantaneous eigenstates of the oneperiod
propagator U(t +T, t),
U(t +T, t)[φ
α
(t)` = e
−iαT/
[φ
α
(t)`, (2.50)
as can easily be seen by inserting the Floquetstate representation (2.46) of the
propagator.
2.5 Numerical computation of Floquet states
Among the methods for the computation of Floquet states of bounded systems, we
essentially discern two classes [37]: The ﬁrst class consists of methods based directly
12 Driven quantum systems and Floquet theory
on the solution of the eigenvalue equation (2.10) of the Floquet Hamiltonian. A sec
ond class of methods starts with the computation of the Floquet propagator U(T, 0),
followed by the solution of the eigenvalue equation (2.50) for the propagator. In the
present work, we treat systems subject to a cosineshaped driving. Accordingly, we
elucidate the numerical methods for the case of a Hamiltonian of the structure
H(t) = H
0
+ 2H
1
cos(Ωt), (2.51)
where we have introduced a factor 2 for ease of notation. They can be generalized
straightforwardly.
2.5.1 Floquetmatrix methods
The Floquet Hamiltonian for (2.51) decomposed into the basis ¦[n`
T
¦ reads
H
n,n
= (H
0
+nΩ)δ
n,n
+H
1
(δ
n,n
+1
+δ
n,n
−1
), (2.52)
or in matrix notation,
H =
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
H
0
+ 2Ω H
1
0 0 0
H
1
H
0
+Ω H
1
0 0
0 H
1
H
0
H
1
0
0 0 H
1
H
0
−Ω H
1
0 0 0 H
1
H
0
−2Ω
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. (2.53)
The eigenvectors of (2.53) are the Fourier components [c
α,n
` of the Floquet states,
as the decomposition into ¦[n`
T
¦ corresponds to Fourier representation. Due to
the Brillouinzone like structure, it is suﬃcient to compute all eigenvectors whose
eigenvalues lie in an interval of size Ω.
As a basis set for the Hilbert space 1, one commonly uses M eigenstates of the
undriven Hamiltonian H
0
, which itself has been decomposed into the eigenfunctions
of the harmonic oscillator (see Appendix A). Thus, for N Floquet channels the
dimension of the Floquet matrix is NM and the computational eﬀort for the matrix
diagonalization is proportional to (NM)
3
.
A further eﬃcient method for the computation of eigenvectors of the tridiagonal
matrix (2.53) are matrix continued fractions [25,45]. We shall not apply this method.
2.5.2 Propagator methods
The quasienergies and the Floquet states at time t = 0 can be extracted from the
oneperiod propagator by use of the eigenvalue equation (2.50) of the unitary oper
ator U(T, 0). In numerical calculations, however, it is advantageous to diagonalize
the Hermitian operator
V = i
1 +U(T, 0)
1 −U(T, 0)
. (2.54)
2.5 Numerical computation of Floquet states 13
Being a function of U(T, 0), V possesses the same eigenvectors as U(T, 0), namely
the Floquet states [φ
α
(0)`. It is straightforward to show that the corresponding
eigenvalues read cot(
α
T/2).
For the computation of the mean energies and to determine the coeﬃcients
of the master equation for the dissipative dynamics (see next chapter), it is nec
essary to know the Floquet states’ Fourier coeﬃcients [c
α,n
`. They are obtained
by propagating the [φ
α
(0)` over one period of the driving, which yields [ψ
α
(t)` =
exp(−i
α
t/)[φ
α
(t)`, for t in the range [0, T]. The [c
α,n
` result from Fourier decom
position, according to their deﬁnition (2.20). The propagation can be performed in
various ways. In the following, we sketch the methods implemented in this work.
Direct integration of the Schr¨ odinger equation
The most simple method for the computation of the propagator is the direct in
tegration of the Schr¨ odinger equation by use of a RungeKutta routine, where the
initial condition is the unit matrix. An extension of this method to other shapes of
driving is rather easy.
It emerges that the numerical eﬀort for the propagation is proportional to NM
3
,
however with a much larger prefactor compared to the diagonalization of the Floquet
matrix. Therefore, computing the Floquet states by direct integration is wellsuited
if a large number of Floquet channels is required.
The (t, t
)formalism
A very eﬃcient numerical method for the computation of the propagator for a Hamil
tonian of the form (2.51) is derived from the (t, t
)formalism [38, 46, 47]. There, the
Schr¨ odinger is extended by a second time coordinate to read
i
∂
∂t
[ψ(t, t
)` =
H(t
) −i
∂
∂t
[ψ(t, t
)`. (2.55)
The time t
is treated formally like an additional canonical coordinate of a time
independent problem. We postulate Tperiodic boundary conditions in t
, which
enables decomposition into the basis set (2.13). Being a solution of (2.55), [ψ(t, t
)`,
on the cut t
= t where ∂t
/∂t = 1, obeys
i
∂
∂t
[ψ(t, t)` = i
∂
∂t
+
∂
∂t
[ψ(t, t
)`
t
=t
(2.56)
= H(t)[ψ(t, t)`. (2.57)
Thus [ψ(t, t)` is a solution of the “true” Schr¨ odinger equation (2.1). In an analogous
way, from the propagator
(t −t
0
) = e
−iH(t−t
0
)/
(2.58)
of the extended Schr¨ odinger equation (2.55) one can extract the “true” propagator.
It reads
U(t, t
0
) =
T
'0[(t −t
0
)[t
`
T
[
t
=t
(2.59)
14 Driven quantum systems and Floquet theory
=
¸
n
T
'0[(t −t
0
)[n`
T
e
inΩt
. (2.60)
This is so because on the one hand it fulﬁlls the initial condition
U(t
0
, t
0
) =
¸
n
T
'0[1
R⊗T
[n`
T
e
inΩt
0
= 1
R
, (2.61)
and on the other hand solves the Schr¨ odinger equation,
i
∂
∂t
U(t, t
0
) = i
∂
∂t
+
∂
∂t
T
'0[(t −t
0
)[t
`
T
t
=t
(2.62)
= H(t)U(t, t
0
). (2.63)
Here the time ordering, which we have to consider explicitly in (2.45), is intrinsic.
By Taylor expansion of the extended propagator  one obtains for the time step
from t to t +τ
U(t + τ, t) =
¸
n
e
inΩ(t+τ)
T
'0[(τ)[n`
T
(2.64)
=
¸
n
e
inΩ(t+τ)
∞
¸
ν=0

(ν)
0,n
(τ), (2.65)
with

(ν)
0,n
(τ) =
1
ν!
−
iτ
ν
T
'0[H
ν
[n`
T
. (2.66)
For a suﬃciently small time step τ, it is possible to truncate the sum over ν after
N+1 terms. Due to the tridiagonal structure of H, in the sum over n all terms with
[n[ > N vanish. Typically, already a few Floquet channels are suﬃcient to obtain
numerical convergence [47].
In the special case N = 1 we obtain U(t +τ, t) = 1−iH(t +τ)τ/, the ﬁrst term
of the Taylor expansion of the timeordered exponential (2.45). For larger N, the
time ordering results in a more complicated expression.
3
Quantum dissipation and
Markov approximation
Within the framework of classical mechanics, dissipation can be introduced phe
nomenologically just by adding a velocityproportional friction force. Although an
extension of the Lagrange formalism to this model of dissipation is possible [48],
quantization results in unphysical properties, e.g., a timedependent mass, or doesn’t
handle the uncertainty relation properly [49].
The most successful approach to dissipation in quantum mechanics, consistent
with the fundamental laws of quantum mechanics, is based on the coupling of the
conservative system to external degrees of freedom. Probably the ﬁrst proof that
such a systembath scheme results in dissipative quantum mechanics was given by
Magalinski˘ı [19] for a harmonic oscillator. Zwanzig generalized this concept within
the framework of classical stochastic processes to arbitrary potentials and derived
a Markovian master equation for the dynamics of the dissipative system by the
socalled projector formalism [20]. By similar approaches, master equations for
quantum systems [50–52] were derived and applied in laser physics [50] and to nuc
lear magnetic resonance and electronspin resonance. Later, Caldeira and Leggett
rediscovered the systembath model in the context of dissipative tunneling [11] and,
in a pathintegral formulation, eliminated the bath exactly [21,53]. This enabled the
investigation of dissipative quantum systems, beyond a weakcoupling limit. Strong
systembath correlations result in interesting eﬀects, among them most prominently
the algebraic decay of correlation functions at zero temperature [54, 55].
However, as soon as nonlinear forces come into play, the pathintegral approach
requires to resort to extensive and sophisticated numerics, such as MonteCarlo
calculations [56–58], with their own shortcomings. Thus, for the description of
nonlinear systems subject to weak dissipation, it is desirable to to treat the inﬂuence
of the bath in perturbation theory, leading to a Markovian master equation for the
density operator [50–52]. In this chapter, we introduce the systembath model and
derive a Markovian master equation for the reduced density operator for the case of
a static central system.
3.1 The systembath model
To achieve a microscopic model of dissipation, we couple the system bilinearly to
a bath of noninteracting harmonic oscillators with masses m
ν
, frequencies ω
ν
, mo
menta p
ν
, and coordinates x
ν
, with the coupling strength c
ν
[11, 19, 59]. The total
Hamiltonian of system and bath is then given by
H = H
S
+H
SB
+H
B
, (3.1)
16 Quantum dissipation and Markov approximation
where H
S
denotes the Hamiltonian of the central system and
H
B
=
¸
ν
p
2
ν
2m
ν
+
1
2
m
ν
ω
2
ν
x
2
ν
, (3.2)
H
SB
= −x
¸
ν
c
ν
x
ν
+x
2
¸
ν
c
2
ν
2m
ν
ω
2
ν
, (3.3)
describe the heat bath and its coupling to the system. The second term in H
SB
,
which depends only on the position x of the system, serves to cancel a renormaliza
tion of the potential due to the coupling [49, 53, 59].
For the time evolution we choose an initial condition of the FeynmanVernon
type: at t = t
0
, the bath is not correlated to the system and canonically distributed
with respect to the free bath Hamiltonian, i.e., the density operator W of system
plus bath reads
W(t
0
) = (t
0
) ⊗
e
−H
B
/k
B
T
tr e
−H
B
/k
B
T
, (3.4)
where is the density operator of the system and k
B
T denotes Boltzmann’s constant
times temperature. Although this choice is somewhat artiﬁcial, it is favorable due
to its technical simplicity. Other initial conditions, like e.g. the canonical ensemble
of the whole system including the coupling [60], are more realistic. However, below
we will deal with driven systems where specifying a more sophisticated preparation
is not meaningful without specifying an onset of the driving.
Due to the bilinearity of the bath and its coupling to the system, one can elim
inate the bath variables to get an exact, closed integrodiﬀerential equation for the
dynamics of the central system, subject to dissipation. The elimination can be
performed in two ways, which are the subjects of the following sections.
3.2 Quantum Langevin equation
From the systembath Hamiltonian (3.1) we derive the Heisenberg equations of mo
tion for the system and the bath operators and solve the latter formally. This results
in a dissipative diﬀerential equation for the Heisenberg position operator of the sys
tem, which is driven by an operatorvalued stochastic force. Although in general,
this quantum Langevin equation cannot be solved exactly and thus is of limited
practical use, it oﬀers a possibility for interpretations.
The Heisenberg equations of motion for the position operators of the system and
of the bath oscillators read
¨ x +
1
m
V
(x) =
1
m
¸
ν
c
ν
x
ν
−
c
ν
m
ν
ω
2
ν
x
, (3.5)
¨ x
ν
+ω
2
ν
x
ν
=
c
ν
m
ν
x. (3.6)
3.2 Quantum Langevin equation 17
Equation (3.6) is easily integrated to yield the formal solution
x
ν
(t) = x
ν
(t
0
) cos ω(t −t
0
) +
p
ν
(t
0
)
m
ν
ω
ν
sin ω
ν
(t −t
0
)
+
c
ν
m
ν
ω
ν
t
t
0
dt
sin ω
ν
(t −t
) x(t
). (3.7)
After integration by parts, inserting into (3.5) results in the socalled quantum
Langevin equation [61–64]
¨ x(t) +
t
t
0
dt
γ(t −t
) ˙ x(t
) +
1
m
V
(x(t)) =
1
m
ξ(t) −γ(t)x(t
0
) (3.8)
with the damping kernel
γ(t) =
1
m
¸
ν
c
2
ν
m
ν
ω
2
ν
cos ω
ν
(t −t
0
) (3.9)
and the operatorvalued ﬂuctuating force
ξ(t) =
¸
ν
c
ν
x
ν
(t
0
) cos ω
ν
(t −t
0
) +
p
ν
(t
0
)
m
ν
ω
ν
sin ω
ν
(t −t
0
)
. (3.10)
The last term in (3.8) gives rise to an initial slip due to the sudden coupling of the
system and the bath at time t
0
[19, 61, 64]. It will be omitted in the following as we
will not study preparation eﬀects within this framework. The inﬂuence of the ﬂuc
tuating force on the system is fully characterized by its symmetric autocorrelation
function, the noise kernel
K(t −t
) =
1
2
'ξ(t)ξ(t
) +ξ(t
)ξ(t)`, (3.11)
=
¸
ν
c
2
ν
2m
ν
ω
ν
coth
ω
ν
2k
B
T
cos ω
ν
(t −t
). (3.12)
To obtain the last line, we have made use of the equilibrium expectation values
1
2
m
ν
ω
2
ν
'x
ν
x
ν
` =
1
2m
ν
'p
ν
p
ν
` =
ω
ν
4
coth
ω
ν
2k
B
T
δ
νν
(3.13)
for the bath operators in the canonical ensemble. As the systembath Hamiltonian
(3.1) is bilinear in the bath coordinates x
ν
, the Gaussian property holds, i.e., we
can express moments and correlations of higher order by products of K’s. The
correlation function K(τ) decays within a time
τ
B
= /k
B
T, (3.14)
which also marks the time scale below which correlations between system and bath
are relevant. In the limit of zero temperature, τ
B
diverges and these correlations
play a dominant role [54, 55].
18 Quantum dissipation and Markov approximation
At this point it is convenient to introduce the spectral density of the systembath
coupling
I(ω) = π
¸
ν
c
2
ν
2m
ν
ω
ν
δ(ω −ω
ν
). (3.15)
In a continuum limit for the heat bath we assume I(ω) to be a smooth function.
The damping and the noise kernel can be expressed by the spectral function to read
γ(t) =
2
πm
∞
0
dω
I(ω)
ω
cos ωt (3.16)
K(t) =
1
π
∞
0
dω I(ω) coth
ω
2k
B
T
cos ωt. (3.17)
Both are not independent of each other since they obey the socalled second ﬂuc
tuationdissipation relation [49], which in Fourier representation reads
K(ω) =
1
2
mωγ(ω) coth
ω
2k
B
T
. (3.18)
In the classical limit k
B
T ω, Eq. (3.18) reads K(ω) = mγ(ω)k
B
T and the
quantum Langevin equation becomes in the longtime limit formally equivalent to
the corresponding classical Langevin equation [49, 62, 63].
As a prototypical model for damping, we use the Ohmic friction kernel γ(t) =
2γδ(t), where the memory of the friction in (3.8) drops to zero. This corresponds to
the Ohmic spectral density I(ω) = mγω. An Ohmic spectral density is often used as
an approximation to a more complicated one and therefore in literature sometimes
appears as “ﬁrst Markov approximation” [50]. The assumption of an increasing
spectral density for arbitrarily high frequencies, however, is not only somewhat
artiﬁcial, but also results in divergent integrals. We regularize them, if required, by
a cutoﬀ in the spectral density,
I(ω) = mγω
ω
2
D
ω
2
+ω
2
D
, (3.19)
which deﬁnes the Drude model. The cutoﬀ frequency ω
D
introduces a short but
ﬁnite memory τ
D
= 1/ω
D
for the friction.
3.3 Inﬂuence functional
Despite the fact that the quantum Langevin equation (3.8) appears quite simple, its
practical use is limited to the very rare cases where it can be integrated directly. A
more useful approach is the elimination of the heat bath in the equation of motion
for the full density operator W, which results in an equation of motion for the
reduced density operator = tr
B
W of the central system subject to dissipation,
where tr
B
denotes the trace over the bath variables. For an exact elimination of the
3.3 Inﬂuence functional 19
heat bath, the pathintegral formulation of quantum mechanics has proved to be
more convenient than operator notation [11, 53].
We start with the time evolution of the full density matrix,
W(t) = e
−iH(t−t
0
)/
W(t
0
)e
iH(t−t
0
)/
, (3.20)
which in position representation reads
W(x
f
, x
f
, x
f
, x
f
, t) ≡ 'x
f
, x
f
[W(t)[x
f
, x
f
` (3.21)
=
dx
0
dx
0
dx
0
dx
0
U(x
f
, x
f
, t; x
0
, x
0
, t
0
) (3.22)
U
∗
(x
f
, x
f
, t; x
0
, x
0
, t
0
)W(x
0
, x
0
, x
0
, x
0
, t
0
). (3.23)
The propagator U(x, x, t; x
0
, x
0
, t
0
) of the system plus the bath is given by the path
integral expression [11, 49, 65]
U(x
f
, x
f
, t; x
0
, x
0
, t
0
) =
x(t)=x
f
x(t
0
)=x
0
Tx
x(t)=x
f
x(t
0
)=x
0
Txexp
i
S[x] +
i
S
B
[x, x]
.
(3.24)
The variable x is a shorthand for all bath coordinates x
ν
and Tx denotes path
integration over all of them. The actions
S[x] =
t
t
0
dt
m
2
˙ x
2
(t
) −V (x(t
))
, (3.25)
S
B
[x, x] =
¸
ν
t
t
0
dt
m
ν
2
˙ x
ν
(t
)
2
−
1
2
m
ν
ω
2
ν
x
ν
(t
) −
c
ν
m
ν
ω
2
ν
x(t
)
2
, (3.26)
correspond to the Hamiltonian H
S
of the central system and H
B
+ H
SB
for the
bath plus systembath coupling, respectively. We insert the initial condition (3.4)
and evaluate the path integral over the bath variables. After tracing out the bath
variables by integrating over all the bath coordinates x
f
, we obtain [11, 21, 49]
(x
f
, x
f
, t) =
dx
0
dx
0
J(x
f
, x
f
, t; x
0
, x
0
, t
0
)(x
0
, x
0
, t
0
), (3.27)
J(x
f
, x
f
, t; x
0
, x
0
, t
0
) =
x(t)=x
f
x(t
0
)=x
0
Tx
x
(t)=x
f
x
(t
0
)=x
0
Tx
exp
i
S[x] −
i
S[x
]
exp
−
1
φ
FV
[x, x
]
. (3.28)
The propagator J(x
f
, x
f
, t; x
0
, x
0
, t
0
) describes the timeevolution of the dissipative
system. The entire inﬂuence of the bath is subsumed in the socalled inﬂuence
functional [21] φ
FV
[x, x
],
Re φ
FV
[x, x
] =
t
t
0
dt
t
t
0
dt
x(t
) −x
(t
)
K(t
−t
)
x(t
) −x
(t
)
, (3.29)
20 Quantum dissipation and Markov approximation
Imφ
FV
[x, x
] = −
m
2
t
t
0
dt
t
t
0
dt
x(t
) −x
(t
)
γ(t
−t
)
˙ x(t
) + ˙ x
(t
)
−
m
2
t
t
0
dt
x(t
) −x
(t
)
γ(t
)
x(t
0
) +x
(t
0
)
. (3.30)
To obtain (3.30), we have integrated Imφ
FV
by parts, canceling the potential renor
malization in (3.26). The last term of the imaginary part gives the initial slip,
known from Eq. (3.8), and is omitted in the following. The real part of the inﬂuence
functional describes the noise, whereas the imaginary part gives rise to friction [49].
3.4 Markovian master equation
By perturbation theory for the propagator (3.28) up to lowest nontrivial order in the
systembath coupling, we derive a master equation of Markovian type, i.e., without
memory. The steps to introduce this Markov approximation are usually performed
in operator notation, starting from the full systembath Hamiltonian (3.1). Here, we
give a derivation from the pathintegral expression (3.28). The present derivation
requires essentially the same approximations as the standard projection technique
approach, but has some advantages. First, one can here distinguish more clearly
between the inﬂuence of friction and noise, because each of them is easily identiﬁed
in the path integral expression (3.28) [49]. A second beneﬁt is the exact cancellation
of the potential renormalization in (3.30). And last but not least, one can show that
for the case of an Ohmic spectral density, the friction part in the Markovian master
equation becomes exact.
In standard perturbation theory for path integrals [65, 66], the exponent of the
inﬂuence functional is approximated by a Taylor series,
exp
−
1
φ
FV
[x, x
]
≈ 1 −
1
φ
FV
[x, x
]. (3.31)
The small parameter in this approximation is the eﬀective coupling strength γ, which
means that γ has to be the smallest frequency scale in the problem. Thus,
γ <1/τ
B
= k
B
T/, (3.32)
γ <∆/, (3.33)
where τ
B
is the correlation time of the bath and ∆ denotes any energy diﬀerence in
the spectrum of the conservative problem.
The propagator for the density matrix is at order zero in the perturbation given
by the ﬁrst line of (3.28). It can be separated into two parts, one depending only
on x, the other only on x
. They are easily identiﬁed as the propagator for the
Schr¨ odinger equation of the pure system and its complex conjugate,
J
0
(x
f
, x
f
, t; x
0
, x
0
, t
0
) = U
0
(x
f
, t; x
0
, t
0
) U
∗
0
(x
f
, t; x
0
, t
0
). (3.34)
3.4 Markovian master equation 21
In ﬁrst order of perturbation (which is already second order in the coupling con
stants c
ν
), the inﬂuence functionals (3.29) and (3.30) only yield contributions at
times t
and t
. Thus we can dissect the path integral into an explicit integration
over x
1
= x(t
) and x
2
= x(t
) and free time evolution [65, 66] to get
(x
f
, x
f
, t) =
dx
0
dx
0
J
0
(x
f
, x
f
, t; x
0
, x
0
, t
0
)(x
0
, x
0
, t
0
)
−
1
t
t
0
dt
t
t
0
dt
dx
1
dx
1
dx
2
dx
2
J
0
(x
f
, x
f
, t; x
1
, x
1
, t
) (x
1
−x
1
)
J
0
(x
1
, x
1
, t
; x
2
, x
2
, t
)K(t
−t
) (x
2
−x
2
)(x
2
, x
2
, t
)
+
im
2
t
t
0
dt
t
t
0
dt
dx
1
dx
1
dx
2
dx
2
J
0
(x
f
, x
f
, t; x
1
, x
1
, t
) (x
1
−x
1
)
J
0
(x
1
, x
1
, t
; x
2
, x
2
, t
) γ(t
−t
) ( ˙ x
2
+ ˙ x
2
)(x
2
, x
2
, t
), (3.35)
where we have assumed that path integration commutes with the integrals over t
and t
. By use of (3.27) and (3.34), we can express (t
) in zeroth order of the
perturbation by (t),
(x
2
, x
2
, t
) =
dx
0
dx
0
J
0
(x
2
, x
2
, t
; x
0
, x
0
, t)(x
0
, x
0
, t). (3.36)
We insert into (3.35), diﬀerentiate with respect to t, and obtain the master equation
˙ (x
f
, x
f
, t)
= −
i
H(x
f
) −H(x
f
)
(x
f
, x
f
, t)
−
1
t
t
0
dτ K(τ)
dx
1
dx
1
dx
2
dx
2
(x
f
−x
f
) U
0
(x
f
, t; x
2
, t −τ)U
∗
0
(x
f
, t; x
2
, t −τ)
(x
2
−x
2
) U
0
(x
2
, t −τ; x, t) U
∗
0
(x
2
, t −τ; x
, t) (x, x
, t)
+
im
2
t
t
0
dτ γ(τ)
dx
1
dx
1
dx
2
dx
2
(x
f
−x
f
) U
0
(x
f
, t; x
2
, t −τ)U
∗
0
(x
f
, t; x
2
, t −τ)
( ˙ x
2
+ ˙ x
2
) U
0
(x
2
, t −τ; x, t) U
∗
0
(x
2
, t −τ; x
, t) (x, x
, t), (3.37)
where the free propagator J
0
for the density matrix has been substituted by the
propagator U
0
of the Schr¨ odinger equation and the integration variable t
by τ =
t − t
. This master equation is Markovian since ˙ (t) depends only on (t), i.e., at
equal times, not on the history of .
In the following chapters, we will solve the master equation in energy basis, in
Floquet basis, or in Wigner representation, respectively. In all these cases, an opera
tor notation is more convenient than a position representation. Deriving from (3.37)
22 Quantum dissipation and Markov approximation
the representationfree form is straightforward and yields
d
dt
= −
i
[H
S
, ] −
1
∞
0
dτ K(τ) [x, [x
H
(t −τ, t), ]]
+
i
2
∞
0
dτ γ(τ) [x, [p
H
(t −τ, t), ]
+
], (3.38)
with the anticommutator [A, B]
+
= AB + BA. The Heisenberg position and mo
mentum operators x
H
and p
H
are deﬁned according to
O
H
(t, t
) = U
†
0
(t, t
) OU
0
(t, t
), (3.39)
where U
0
(t, t
) = exp(−iH
S
(t − t
)/) denotes the propagator of the conservative
system. We have assumed further that the integration kernel K(τ) is practically
zero for τ > τ
B
[67] and extended the upper integration limit in (3.38) to inﬁnity.
This implicitly moved the preparation time t
0
→ −∞, thus the master equation
(3.38) describes only the system dynamics suﬃciently close to equilibrium.
For an Ohmic spectral density γ(τ) = 2γδ(τ), the integration in the second line
of (3.38) can be evaluated and we obtain the Markovian master equation
d
dt
= −
i
[H
S
, ] +
L
friction
+L
0
noise
. (3.40)
The commutator in (3.40) gives the coherent dynamics, whereas the superoperators
L
friction
= −
iγ
2
[x, [p, ]
+
], (3.41)
L
0
noise
= −
1
[x, [Q, ]], (3.42)
describe the inﬂuence of the the bath: friction and noise. The operator
Q =
∞
0
dτ K(τ) x
H
(t −τ, t), (3.43)
is qualitatively the Heisenberg position operator x
H
of the system in Fourier re
presentation. Therefore L
0
noise
depends on the conservative dynamics (superscript
0
),
thus on the energy spectrum of the central system. Note that Q is time independent,
since for a static Hamiltonian x
H
(t −τ, t) = x
H
(−τ).
The Markovian master equation (3.40) together with (3.41) and (3.42), does
not exhibit Lindblad form (B.3), thus the positivity of the density operator is not
guaranteed for all possible initial states. The violation of positivity due to a master
equation in this case, however, is a transient eﬀect which only arises for preparations
far from equilibrium [68–71], where the conditions under which the master equation
has been derived, are not fulﬁlled. (See Appendix B.1 for a more detailed discussion).
4
Driving and dissipation:
FloquetMarkov theory
For a dissipative quantum system subject to external driving, even a partially analyt
ical solution within the pathintegral approach is feasible only for the very simplest
systems, in particular, for the periodically driven, damped harmonic oscillator [28],
or for driven dissipative twolevel systems [72, 73]. Thus, for the description of
strongly driven systems subject to weak dissipation, it is desirable to combine a
Markovian approach to quantum dissipation, leading to a master equation for the
density operator, with the Floquet formalism that allows to treat timeperiodic
forces of arbitrary strength and frequency. While the Floquet formalism amounts
essentially to using an optimal representation and is exact [23], the simpliﬁcation
brought about by the Markovian description is achieved only at the expense of ac
curacy. Here, a subtle technical diﬃculty lies in the fact that the truncation of the
longtime memory introduced by the bath and the inclusion of the driving do not
commute: As pointed out in Refs. [26, 27], the result of the Markov approximation
depends on whether it is made with respect to the eigenenergy spectrum of the
central system without the driving, or with respect to the quasienergy spectrum ob
tained from the Floquet solution of the driven system. In the second case it cannot
be treated as a system with proper eigenstates and eigenenergies. Figure 4.1 depicts
the two diﬀerent possibilities for including driving and dissipation to the description
of a quantum system. Both approaches yield a Markovian master equation, but
diﬀer quantitatively. We will investigate this diﬀerence in detail for the case of a
parametrically driven harmonic oscillator in Chapter 5.
A Floquet theory for dissipative driven systems based on the energy spectrum
has been worked out and applied to intenseﬁeld excitations of atoms in Refs. [37,74];
a quasienergy spectrum approach has been implemented in recent work on driven
Rydberg atoms [22, 75] and coherent destruction of tunneling [76–78].
4.1 Simple inclusion of the driving
A simple Markovian approach to dissipative driven quantum systems results directly
from the master equation for the undriven system: We replace in (3.40) the static
Hamiltonian H
S
by the timedependent Hamiltonian
H
S
(t) = H
0
+H
F
(t) (4.1)
which yields
d
dt
= −
i
[H
S
(t), ] +
L
friction
+L
0
noise
. (4.2)
24 Driving and dissipation: FloquetMarkov theory
S
Floquet theory
−−−−−−−−−−−−−−−−−−−→
for Schr¨ odinger equation
S +D
Markov
(energy spectrum)
Markov
(quasienergy spectrum)
S +B
Floquet theory
−−−−−−−−−−−−−−−→
for master equation
S +D+B
Figure 4.1: Successive inclusion of the driving (D) and the inﬂuence of a heat bath (B)
to the description of a quantum system (S). The horizontal arrows denote exact Floquet
treatment, whereas the vertical arrows mark an approximate step, namely the truncation
of the longtime memory. The result depends on the route taken.
Here, the driving enters only the coherent part of the master equation, whereas
L
0
noise
has been derived from the undriven Hamiltonian H
0
. Thus, we refer to this
approach as the Markovian approach with respect to the unperturbed spectrum.
For a periodically timedependent driving, H
F
(t) = H
F
(t +T), the master equation
(4.2) allows for a Floquet treatment [37].
4.2 An improved Markovian master equation
We pointed out in Section 3.4, that the coherent dynamics of the central system
plays an important role in the derivation of the Markovian master equation (3.40).
This means that for a driven system the Markovian master equation depends on
whether the driving is considered in its derivation or not.
To obtain an improved master equation whose dissipative kernel accounts for
the inﬂuence of the driving, we start anew from the full systembath Hamiltonian
including the driving. Performing the same steps as in the preceeding chapter,
but for an explicitly timedependent Hamiltonian H
S
(t), we obtain the Markovian
master equation
d
dt
= −
i
[H
S
(t), ] + (L
friction
+L
noise
) . (4.3)
Friction and noise are described by the superoperators
L
friction
= −
iγ
2
[x, [p, ]
+
], (4.4)
L
noise
= −
1
[x, [Q(t), ]]. (4.5)
Whereas L
friction
is the same for both Markovian approaches, L
noise
has acquired a
time dependence which stems from the operator
Q(t) =
∞
0
dτ K(τ) x
H
(t −τ, t), (4.6)
4.3 Decomposition into Floquet basis 25
where
x
H
(t, t
) = U
†
(t, t
) xU(t, t
) (4.7)
is the Heisenberg position operator of the driven system which depends explicitly on
both times, t and t
, not only on their diﬀerence. Therefore L
noise
is time dependent
and does—in contrast to L
0
noise
—not depend on the energy spectrum of the undriven
system, but on the quasienergy spectrum of the driven system.
Since the role of the eigenenergies is now taken over by the quasienergies, we refer
to this master equation as the Markovian approach with respect to the quasienergy
spectrum. The inﬂuence of a driving force on L
noise
will be studied in detail for the
case of a parametrically driven harmonic oscillator in Chapter 5.
4.3 Decomposition into Floquet basis
So far, we did not specify the time dependence of the system Hamiltonian in the
derivation of the master equation. By assuming a Tperiodic Hamiltonian, we are
able to make use of the Floquet theorem and expand the reduced density operator
into the timeperiodic Floquet states [φ
α
(t)` of the isolated driven system. They
form a welladapted basis for the case of weak dissipation. A master equation for
the matrix elements
αβ
= 'φ
α
(t)[[φ
β
(t)` (4.8)
is derived from the basisindependent improved master equation (4.3).
4.3.1 Matrix elements
To decompose the master equation (4.3), we need to know the matrix elements of
the operators x, p and Q(t) in the Floquet basis. They all are Tperiodic and can
be expressed as a Fourier series,
X
αβ
(t) ≡ 'φ
α
(t)[x[φ
β
(t)` =
¸
n
e
inΩt
X
αβ,n
, (4.9)
P
αβ
(t) ≡ 'φ
α
(t)[p[φ
β
(t)` =
¸
n
e
inΩt
P
αβ,n
, (4.10)
Q
αβ
(t) ≡ 'φ
α
(t)[Q(t)[φ
β
(t)` =
¸
n
e
inΩt
Q
αβ,n
. (4.11)
The Fourier coeﬃcients of the position matrix elements read
X
αβ,n
=
1
T
T
0
dt e
−inΩt
'φ
α
(t)[x[φ
β
(t)` (4.12)
= ''φ
α
(t)[x[φ
(−n)
β
(t)``. (4.13)
Next, we will express the Fourier coeﬃcients P
αβ,n
and Q
αβ,n
in terms of X
αβ,n
.
26 Driving and dissipation: FloquetMarkov theory
For a Hamiltonian of the form H = p
2
/2m + V (x, t), the momentum operator
can be expressed by a commutator,
p =
m
i
[H, x] =
m
i
[H, x], (4.14)
where H = H −i∂/∂t denotes the Floquet Hamiltonian. Thus we get
P
αβ,n
=
m
i
''φ
α
[p[φ
(−n)
β
`` (4.15)
=
m
i
(
α
−
β
+nΩ)X
αβ,n
. (4.16)
To obtain the last line, we made use of the eigenvalue equation (2.29) for the Floquet
states after inserting (4.14).
The Fourier coeﬃcients of the timedependent matrix element Q
αβ
(t) read
Q
αβ,n
=
1
T
T
0
dt e
−inΩt
∞
0
dτ K(τ)'φ
α
(t)[x
H
(t −τ, t)[φ
β
(t)` (4.17)
=
∞
0
dτ
mγ
π
∞
0
dω ω coth
ω
2k
B
T
cos(ωτ)e
−i(α−
β
+nΩ)τ/
X
αβ,n
,
where we have inserted the spectral representation (3.17) of the noise kernel and
made use of (4.12). The τintegration is evaluated by using
∞
0
dτ exp(iωτ) =
πδ(ω) + iP(1/ω), where P denotes Cauchy’s principal part. We end up with
Q
αβ,n
=
mγ
2
(
α
−
β
+nΩ) coth
α
−
β
+nΩ
2k
B
T
X
αβ,n
. (4.18)
The contributions of the principal part result in quasienergy shifts of the order γ,
the socalled Lamb shifts [50, 51], and have been neglected.
By use of the Fourier representations (4.9)–(4.11) we obtain from Eq. (4.3) the
FloquetMarkov master equation [22, 27, 75]
˙
αβ
(t) =
d
dt
'φ
α
(t)[(t)[φ
β
(t)`
= −
i
(
α
−
β
)
αβ
(t)
+
¸
α
β
nn
e
i(n+n
)Ωt
(N
αα
,n
+N
ββ
,−n
)X
αα
,n
α
β
X
β
β,n
(4.19)
−N
β
α
,n
X
αβ
,n
X
β
α
,n
α
β
−N
α
β
,−n
αβ
X
β
α
,n
X
α
β,n
.
Note that the coeﬃcients of this diﬀerential equation are periodic in time with the
period of the driving. The N
αβ,n
are given by
N
αβ,n
= N(
α
−
α
+nΩ), N() =
mγ
2
n
th
(), (4.20)
with the thermal occupation number
n
th
() =
1
e
/k
B
T
−1
=
1
2
¸
coth
2k
B
T
−1
. (4.21)
For k
B
T, N() approaches zero.
4.3 Decomposition into Floquet basis 27
4.3.2 Rotatingwave approximation
We used the Floquet basis to formally eliminate a driving force of arbitrary strength
from the coherent part of the master equation. However, the coeﬃcients of the
dissipative part are still time dependent and complicate the solution of the master
equation. Here, we explore the conditions under which these coeﬃcients can be
replaced by their time average. This step eﬀectively amounts to a rotatingwave
approximation (RWA).
Moderate rotatingwave approximation
Assuming that dissipative eﬀects are relevant only on a time scale much larger than
the period 2π/Ω of the driving, we average the likewise 2π/Ωperiodic coeﬃcients
of the master equation (4.19) over one period of the driving and end up with the
equation of motion
˙
αβ
(t) = −
i
(
α
−
β
)
αβ
(t) +
¸
α
β
L
αβ,α
β
α
β
(t), (4.22)
with the dissipative transition rates
L
αβ,α
β
=
¸
n
(N
αα
,n
+N
ββ
,n
) X
αα
,n
X
β
β,−n
(4.23)
−δ
ββ
¸
β
,n
N
β
α
,n
X
αβ
,−n
X
β
α
,n
−δ
αα
¸
α
n
N
α
β
,n
X
β
α
,−n
X
α
β,n
.
The timeindependence of its coeﬃcients reﬂects that the inﬂuence of the driving has
been formally absorbed by decomposing into the Floquet basis. Note that diagonal
and oﬀdiagonal elements of the density matrix are not decoupled. It has also to
be stressed that the rotatingwave approximation introduced here is less restrictive
than the one in Refs. [22, 75], as detailed in the next paragraph.
Full rotatingwave approximation
In some cases one can even go one step further. We solve the coherent part of the
master equation (4.22) by the ansatz
αβ
(t) = e
−i(α−
β
)t/
σ
αβ
(t), (4.24)
a transformation to the Heisenberg picture of the central system plus the driving.
Inserting into (4.22) yields
˙ σ
αβ
(t) =
¸
α
β
e
i(α−
β
−
α
+
β
)t/
L
αβ,α
β
σ
α
β
(t) (4.25)
If dissipative eﬀects are only relevant on a time scale much longer than all ﬁnite
times 2π/(
α
−
β
−
α
+
β
), we are allowed to replace the coeﬃcients in (4.25)
by their time average. Thus only the L
αβ,α
β
which fulﬁll the fullRWA condition
α
−
β
=
α
−
β
, (4.26)
28 Driving and dissipation: FloquetMarkov theory
remain in (4.25) or (4.22), respectively. This condition is, however, much more
restrictive than the one in the previous paragraph, since here, we have averaged
over a longer time scale. Therefore the applicability of a full RWA is limited to
very rare cases like, e.g., harmonic potentials with their equidistant (quasi) energy
levels.
Moreover, one can assume that for the case of a completely irregular spectrum
where all quasienergies are eﬀectively random numbers [8, 79], the quasienergy dif
ferences have no degeneracy at all. Then the fullRWA condition (4.26) results
in [22, 75]
α = α
, β = β
or α = β, α
= β
. (4.27)
Inserting into (4.25) yields two decoupled sets of equations for the diagonal and the
oﬀdiagonal matrix elements,
˙ σ
αα
(t) =
¸
α
L
αα,α
α
σ
α
α
(t), (4.28)
˙ σ
αβ
(t) = L
αβ,αβ
σ
αβ
(t), α = β. (4.29)
The second equation results in an exponential decay of the oﬀdiagonal matrix ele
ments. Therefore in the asymptotic limit, the density matrix becomes diagonal in
the Floquet basis.
We will, however, ﬁnd in Section 6.3 that even in a case where the dynamics
of the system is fully chaotic and thus, a full RWA seems to be appropriate, the
oﬀdiagonal matrix elements play an important role for the asymptotic state.
4.4 The dissipative quantum map and its numerical imple
mentation
The master equation (4.3) generates a dynamical semigroup for the time evolution
of the density operator. Its coeﬃcients share the Tperiodicity of the driven system
Hamiltonian H
S
(t), i.e., Eq. (4.3) meets the conditions for a Floquet treatment.
Therefore, it is possible to deﬁne a dissipative quantum map ((T) [25, 74, 80]—
the analogue of the onecycle propagator U(T, 0) in the conservative case—which
describes the stroboscopic dissipative time evolution of the density operator,
(nT) = [((T)]
n
(0). (4.30)
As the dynamics generated by (4.30) is dissipative, it converges in the longtime
limit to an asymptotic state
∞
, the “quantum attractor” which is the ﬁxed point
of the dissipative quantum map ((T).
Decomposing into the Floquet basis ¦[φ
α
(t)`¦ yields the onecycle propagation
of the density matrix elements
αβ
((n + 1)T) =
¸
α
β
(
αβ,α
β
(T)
α
β
(nT). (4.31)
4.4 The dissipative quantum map and its numerical implementation 29
An equation of motion for the dissipative map,
˙
(
αβ,α
β
(t) = −
i
(
α
−
β
)(
αβ,α
β
(t) δ
αα
δ
ββ
+
¸
α
β
L
αβ,α
β
(
α
β
,α
β
(t) (4.32)
follows straightforwardly from (4.22). This form enables a numerical treatment of
the master equation: We integrate (4.32) over one period of the driving T to obtain
the dissipative map ((T). The time evolution of the density operator results from
iteration according to (4.31).
30
5
The parametrically driven
harmonic oscillator
In this chapter we investigate the properties and the quality of the diﬀerent Markov
ian approaches to damped periodically driven quantum dynamics for a linear system
where an exact pathintegral solution is still available: The parametrically driven,
damped harmonic oscillator allows for a very transparent and wellcontrolled investi
gation of the diﬀerent approximation schemes introduced in Chapters 3 and 4. Here,
their quality can be reliably checked since in this system, the quasienergy spectrum
is suﬃciently diﬀerent from the unperturbed energy spectrum [81,82] (this feature is
in contrast to the additively driven harmonic oscillator where the diﬀerence of two
quasienergies does not depend on the driving parameters [83]), and a comparison
with the known quantum pathintegral solution [28] is possible.
Moreover, by switching to a phasespace representation such as the Wigner func
tion, it is possible to elucidate the relationship of the quantal results to the corre
sponding classical FokkerPlanck dynamics. Since this relation is particularly close
in the case of linear systems, this provides an additional consistency check. There
fore, a strong emphasis of this chapter is on the testing and thorough understanding
of the available methods.
Forming a convenient “laboratory animal” due to its simplicity and linearity, the
parametrically driven harmonic oscillator still shows nontrivial behavior, interesting
in its own right. We give a brief review of the model, its classical dynamics, and its
coherent quantum dynamics in Sections 5.1 and 5.2. In Section 5.3 we present the
solution of the dissipative dynamics in FloquetMarkov description. A reﬁned inves
tigation within a basisindependent description, which allows for a detailed analysis
of the inﬂuence of the driving on the dissipative terms of the master equation,
is given in Section 5.4. Section 5.5 is devoted to a discussion of the asymptotics
of the quantal solutions, such as the conservative and the hightemperature lim
its. Section 5.6 contains numerical results for a number of characteristic dynamical
quantities as obtained for the alternative Markovian approaches, and the compar
ison to the pathintegral solution. A summary of the various representations and
levels of Markovian description, with their interrelations, is given in Section 5.7. A
merely technical issue, the solution of a FokkerPlanck equation by the method of
characteristics, is deferred to Appendix C.
5.1 The model and its classical dynamics
For a particle with mass m moving in a harmonic potential with timedependent
frequency, the Hamiltonian is given by
H
S
(t) =
p
2
2m
+
1
2
k(t)x
2
, (5.1)
32 The parametrically driven harmonic oscillator
where k(t) = mω
2
(t) is a periodic function with period T. An initial phase of the
driving can be taken into account by a proper time translation. A special case is
the Mathieu oscillator, where
ω
2
(t) = ω
2
0
+ε cos Ωt, Ω = 2π/T. (5.2)
This is an experimentally important case in view of the fact that it describes the
Paul trap [84]. Depending on its frequency and amplitude, the driving can stabilize
or destabilize the undriven oscillation. Figure 5.1 shows the zones of stable and
unstable motion, respectively, for the Mathieu oscillator, in the ω
2
0
ε plane.
The equation of motion for a classical particle with Ohmic (i.e., velocitypropor
tional) dissipation in the potential given in (5.1) reads
¨ x +γ ˙ x +
1
m
k(t)x = 0. (5.3)
By substituting x = ξ exp(−γt/2), we can formally remove the damping to get an
undamped equation with a modiﬁed potential
¨
ξ +
ω
2
(t) −γ
2
/4
ξ = 0. (5.4)
Already here, on the level of the classical equations of motion, we can apply the Flo
quet theorem for secondorder diﬀerential equations with timeperiodic coeﬃcients.
It asserts [42, 85] that Eq. (5.4) has two solutions of the form
ξ
1
(t) = e
iµt
ϕ(t), ξ
2
(t) = ξ
∗
1
(t), ϕ(t +T) = ϕ(t). (5.5)
The solution ξ
2
(t) is related to ξ
1
(t) by the fact that the coeﬃcients in the diﬀerential
equation (5.4) are real. Being periodic in time, the classical Floquet function ϕ(t)
can be represented as a Fourier series,
ϕ(t) =
∞
¸
n=−∞
c
n
e
inΩt
. (5.6)
The Floquet index µ depends on the shape of the driving k(t) and is deﬁned only
modΩ. There exist driving functions for which µ is complex so that one of the
solutions ξ
i
(t) becomes unstable (cf. Fig. 5.1). In stable regions µ is real. On the
border between a stable and an unstable region, µ becomes a multiple of Ω/2 and the
solutions ξ
1
(t) and ξ
2
(t) are not linearly independent. For given k(t), the functions
ϕ(t), ξ
i
(t) and the Floquet index µ still depend on the damping γ. We denote the
limit γ →0 of ϕ(t), ξ
i
(t), µ by ϕ
0
(t), ξ
0
i
(t), µ
0
.
The normalization of the c
n
is chosen such that the Wronskian J, which is a
constant of the motion, is given by
J =
˙
ξ
1
(t)ξ
2
(t) −ξ
1
(t)
˙
ξ
2
(t) = 2i, (5.7)
resulting in the sum rule
∞
¸
n=−∞
c
2
n
(µ +nΩ) = 1. (5.8)
5.1 The model and its classical dynamics 33
0 5 10 15
5
0
5
10
15
20
25
30
ε [Ω
2
/2]
ω
2 0
[
Ω
2
/
4
]
stable stable
unstable
Figure 5.1: Stability of equation (5.3) with γ = 0 for the case of a Mathieu oscillator. In
the white areas the Floquet index µ is real, which corresponds to stable solutions. In the
shaded areas µ is complex and therefore one of the fundamental solutions (5.5) is unstable.
On the borderlines, µ becomes a multiple of Ω/2 and the motion is marginal stable.
Returning to the original xcoordinate, we ﬁnd that the fundamental solutions
of (5.3) read
f
i
(t) = e
−γt/2
ξ
i
(t), i = 1, 2. (5.9)
For constant frequency of the oscillator, k(t) = const = mω
2
0
, the Floquet index
and the periodic function become µ = (ω
2
0
− γ
2
/4)
1/2
and ϕ(t) = (ω
2
0
− γ
2
/4)
−1/2
,
respectively, which reproduces the results for a damped harmonic oscillator without
driving.
The Green function for Eq. (5.3) is constructed using Eqs. (5.6) and (5.7),
G(t, t
) = e
−γ(t−t
)/2
ξ
1
(t)ξ
2
(t
) −ξ
2
(t)ξ
1
(t
)
/2i (5.10)
= e
−γ(t−t
)/2
¸
n,n
c
n
c
n
sin
µ(t −t
) + Ω(nt −n
t
)
. (5.11)
In terms of this function, the solution of (5.3) with initial conditions x(t
0
) = x
0
and
p(t
0
) = p
0
, reads
x(t, t
0
) = −x
0
∂G(t, t
0
)
∂t
0
+
p
0
m
G(t, t
0
). (5.12)
Since the potential breaks continuous timetranslational invariance, this solution
depends explicitly on the initial time t
0
.
34 The parametrically driven harmonic oscillator
5.2 Floquet states in stable regimes
It can be shown by group theoretical methods that the quantum mechanical quasi
energy spectrum of a parametrically driven harmonic oscillator in a stable regime is
equivalent to the energy spectrum of an undriven harmonic oscillator [83]. In unsta
ble zones or on the borderlines, the quasienergy spectrum is equivalent to the energy
spectrum of a parabolic barrier or of a free particle, respectively. The latter cases
result in a continuous quasienergy spectrum. We restrict ourselves to the motion in
stable regions.
In these regions of the parametrically driven harmonic oscillator (5.1), the Flo
quet solutions for the Schr¨ odinger equation are derived in the literature in various
ways [81, 86–90]. Here we sketch a derivation in the spirit of Ref. [86].
A solution of the classical equation of motion (5.3) in the nondissipative case
γ = 0 reads
x(t) =
2m
A
∗
ξ
0
1
(t) +Aξ
0
2
(t)
, (5.13)
where A and A
∗
are complex normal coordinates. In a quantized version, they are
replaced by the conjugate pair of operators
A(t) =
i
√
2m
ξ
0
1
(t)p −m
˙
ξ
0
1
(t)x
, (5.14)
A
+
(t) = −
i
√
2m
ξ
0
2
(t)p −m
˙
ξ
0
2
(t)x
, (5.15)
which satisfy the canonical commutation relation
[A(t), A
+
(t)] = 1. (5.16)
In the limit of zero driving amplitude they reduce to the familiar shift operators
(A.3), (A.4) of the timeindependent harmonic oscillator.
The parametrically driven harmonic oscillator (5.1) possesses a Tperiodic Her
mitian invariant operator, the socalled Lewis invariant [86]
I(t) = A
+
(t)A(t) =
1
2
x ˙ r(t) −p r(t)
2
+
x
2
r
2
(t)
, (5.17)
r(t) =
ξ
0
1
(t) ξ
0
2
(t) = [ϕ
0
(t)[. (5.18)
The instantaneous eigenstates ψ
α
(x, t) of this invariant coincide—besides a time
dependent phase factor—with the Floquet states of the system [86, 87, 91]. They
can be constructed in analogy to the energy eigenstates of the timeindependent
harmonic oscillator: From the commutation relation (5.16) one obtains
A(t) ψ
α
(x, t) =
√
αψ
α−1
(x, t), (5.19)
A
+
(t) ψ
α
(x, t) =
√
α + 1 ψ
α+1
(x, t). (5.20)
5.2 Floquet states in stable regimes 35
Solving A(t)ψ
0
(x, t) = 0 and iterating according to (5.20), we ﬁnd for I(t) the
eigenfunctions
ψ
α
(x, t) =
(A
+
(t))
α
√
α!
ψ
0
(x, t) (5.21)
=
ξ
0
2
(t)
ξ
0
1
(t)
α/2
m/π
2
α
α!ξ
0
1
(t)
H
α
x
m/ξ
0
1
(t)ξ
0
2
(t)
exp
im
2
˙
ξ
0
1
(t)
ξ
0
1
(t)
x
2
,
where H
α
is the αth Hermite polynomial, α = 0, 1, 2, . . . . These states are solutions
of the timedependent Schr¨ odinger equation [86] and in the undriven limit reduce
to the position representation of the familiar eigenstates (A.10). Separating ψ
α
(x, t)
into a 2π/Ωperiodic function and an exponential prefactor, one ﬁnds the Floquet
states
φ
α
(x, t) =
m/π
2
α
α!ϕ
0
(t)
H
α
x
[ϕ
0
(t)[
m
exp
im
2
˙
ξ
0
1
(t)
ξ
0
1
(t)
x
2
. (5.22)
The corresponding quasienergies
α
= µ
0
(α + 1/2) (5.23)
are chosen such that in the undriven limit they reduce to the eigenenergies of the
harmonic oscillator. Thus they do not lie within a single Brillouin zone.
The matrix elements of the position operator x with the Floquet states [φ
α
(t)`,
which we will need to obtain the coeﬃcients of the master equation, read
X
αβ
(t) = 'φ
α
(t)[x[φ
β
(t)` (5.24)
=
¸
n
e
inΩt
X
αβ,n
, (5.25)
X
αβ,n
=
1
T
T
0
dt e
−inΩt
'φ
α
(t)[x[φ
β
(t)`. (5.26)
To obtain Eqs. (5.25) and (5.26), the periodicity of the Floquet states [φ
α
(t)` has
been used. The Fourier components X
αβ,n
are preferably evaluated in the spatial
representation,
X
αβ
(t) =
∞
−∞
dxφ
α
(x, t) xφ
β
(x, t) (5.27)
=
2m
β ϕ
0
(−t)δ
α,β−1
+
√
αϕ
0
(t)δ
α,β+1
, (5.28)
by inserting the Fourier expansion (5.6) for ϕ
0
(t), to give
X
αβ,n
=
2m
β c
−n
δ
α,β−1
+
√
αc
n
δ
α,β+1
. (5.29)
36 The parametrically driven harmonic oscillator
5.3 FloquetMarkov description in full RWA
In the full rotatingwave approximation (RWA) introduced in Section 4.3.2, we
neglect all contributions with
α
−
β
=
α
−
β
in Eq. (4.25). Thus in the
present case of an equidistant quasienergy spectrum, we have to keep all terms
with (α −β) = (α
−β
).
Substituting Eq. (5.29) in Eq. (4.23), we obtain from (4.25) the timeindependent
master equation
˙ σ
αβ
=
γ
2
(N + 1)
2
(α + 1)(β + 1)σ
α+1,β+1
−(α +β)σ
αβ
+N
2
αβσ
α−1,β−1
−(α +β + 2)σ
αβ
¸
. (5.30)
The eﬀective thermalbath occupation number
N =
¸
n
c
0
n
2
(µ
0
+nΩ) n
th
(µ
0
+nΩ) (5.31)
reduces to N = n
th
(ω
0
) in the undriven limit.
Formally, this master equation coincides with the one for the undriven dissipa
tive harmonic oscillator in rotatingwave approximation [50]. It has the stationary
solution
σ
∞
αβ
=
∞
αβ
=
1
N + 1
N
N + 1
α
δ
αβ
. (5.32)
The density operator of the asymptotic solution is diagonal in this representation
and reads
∞
(t) =
∞
¸
α=0
∞
αα
[φ
α
(t)`'φ
α
(t)[. (5.33)
The basis ¦[φ
α
(t)`¦ corresponds to the “generalized Floquet states” introduced in
Ref. [26], i.e., they are centered on the classical asymptotic solution and diagonalize
the asymptotic density operator.
To get the variances of (5.33), we switch to the Wigner representation. There,
the asymptotic state reads
W
∞
(x, p, t) =
∞
¸
α=0
∞
αα
W
α
(x, p, t), (5.34)
where
W
α
(x, p, t) =
(−1)
α
π
e
−z
2
L
α
(2z
2
), (5.35)
z
2
=
1
m
˙
ξ
0
1
(t)
˙
ξ
0
2
(t)x
2
−
˙
ξ
0
1
(t)ξ
0
2
(t) +ξ
0
1
(t)
˙
ξ
0
2
(t)
px +ξ
0
1
(t)ξ
0
2
(t)p
2
/m
,
is the Wigner function corresponding to [φ
α
(t)` [89], with the αth Laguerre polyno
mial L
α
. Using the sum rule [92]
∞
¸
α=0
κ
α
L
α
(x) = (1 −κ)
−1
exp
xκ
κ −1
, (5.36)
5.3 FloquetMarkov description in full RWA 37
we obtain the asymptotic solution in Wigner representation as
W
∞
(x, p, t) =
1
π(2N + 1)
e
−z
2
/(2N+1)
. (5.37)
It is a Gaussian with the variances
σ
xx
(t) =
m
(N + 1/2)ξ
0
1
(t)ξ
0
2
(t), (5.38)
σ
xp
(t) = (N + 1/2)
˙
ξ
0
1
(t)ξ
0
2
(t) +ξ
0
1
(t)
˙
ξ
0
2
(t)
/2, (5.39)
σ
pp
(t) = m(N + 1/2)
˙
ξ
0
1
(t)
˙
ξ
0
2
(t). (5.40)
To enable a comparison between the diﬀerent equations of motions for the dissi
pative quantum system, we also give for the master equation in RWA, Eq. (5.30), the
corresponding partial diﬀerential equation in Wigner representation. For a deriva
tion, we use the shift properties (5.19) and (5.20) of the operators A and A
+
, to
obtain the corresponding basisfree operator equation from the master equation
(5.30) for the density matrix elements σ
αβ
˙ = −
i
[H
S
(t), ]
+
γ
2
(N + 1)
2AA
+
−A
+
A −A
+
A
(5.41)
+N
2A
+
A −AA
+
−AA
+
¸
.
The dissipative part of this equation is the same as for the undriven dissipative
harmonic oscillator [50], but with the shift operators for Floquet states instead of
the usual creation and annihilation operators. Obviously, this master equation is of
Lindblad form [93] (see Appendix B.1).
By substituting (5.14), (5.15), we get an operator equation which only consists
of position and momentum operators. Applying the transformations (A.25)–(A.28)
yields for the Wigner function the diﬀerential equation
∂
t
W(x, p, t) = L
RWA
(t) W(x, p, t), (5.42)
with the diﬀerential operator
L
RWA
(t) = −
1
m
p∂
x
+
γ
2
(∂
x
x +∂
p
p) +k(t)x∂
x
+
γ
2
D
xx
(t)∂
2
x
+D
xp
(t)∂
x
∂
p
+D
pp
(t)∂
2
p
(5.43)
and the diﬀusion coeﬃcients
D
xx
(t) = ξ
0
1
(t)ξ
0
2
(t)(N + 1/2)/m, (5.44)
D
xp
(t) =
˙
ξ
0
1
(t)ξ
0
2
(t) +ξ
0
1
(t)
˙
ξ
0
2
(t)
(N + 1/2), (5.45)
D
pp
(t) = m
˙
ξ
0
1
(t)
˙
ξ
0
2
(t)(N + 1/2). (5.46)
38 The parametrically driven harmonic oscillator
The fact that there are also dissipative terms in Eq. (5.42) containing derivatives
with respect to x is a consequence of the RWA: Its eﬀect is equivalent to using the
coupling Hamiltonian H
RWA
SB
=
¸
ν
g
ν
(ab
†
ν
+ a
†
b
ν
) instead of (3.1), where a and b
ν
are the usual annihilation operators of the system and the bath, respectively. This
introduces an additional coupling term ∝ pp
ν
. In the next section we show how to
avoid this RWA, by returning to the original Markov approximation, Eq. (4.3).
5.4 Basisindependent description beyond RWA
In the present case of a bilinear system, driven or not, for which the classical motion
is integrable, the knowledge of the classical dynamics opens a more direct access
also to the quantal time evolution. Speciﬁcally, the Heisenberg position operator
x
H
(t, t
) for the corresponding undamped quantum system is given by the solution
of the classical equation of motion in the limit γ →0, indicated by the superscript
0
.
In our case the classical solution is given by (5.12). The corresponding interaction
picture position operator reads
x
H
(t, t
) = −x
∂G
0
(t, t
)
∂t
+
p
m
G
0
(t, t
), (5.47)
where x and p now denote the position and the momentum operator. Inserting this
operator into Eq. (4.6), leads to the master equation
˙ = −
i
[H
S
(t), ] −
i
2
γ [x, [p, ]
+
]
−
γ
2
D
pp
[x, [x, ]] +
γ
2
D
xp
[x, [p, ]], (5.48)
with the periodically timedependent transport coeﬃcients
D
pp
(t) = −
γ
∞
0
dτ K(τ)
∂G
0
(t −τ, t
)
∂t
t
=t
, (5.49)
D
xp
(t) = −
mγ
∞
0
dτ K(τ) G
0
(t −τ, t). (5.50)
This form of the master equation does not produce a positive semideﬁnite diﬀusion
matrix. It consequently does not exhibit Lindblad form [93] (see Appendix B.1).
Note that within a Markov approximation, the master equation is periodic with the
driving period T = 2π/Ω. This is in contrast to the nonMarkovian exact master
equation [28]. In this latter case, the eﬀective master equation has the structure
of (5.48) with coeﬃcients D
xp
and D
pp
which depend in a nonperiodic way on the
time elapsed since the preparation at t
0
. In Wigner representation, this corresponds
to a timedependent diﬀusion coeﬃcient, see Eq. (5.56), below.
To evaluate these expressions, we substitute the undamped limit of Eq. (5.11),
G
0
(t, t
) =
∞
¸
n,n
=−∞
c
0
n
c
0
n
sin
µ
0
(t −t
) + Ω(nt −n
t
)
. (5.51)
5.4 Basisindependent description beyond RWA 39
The explicit time dependence in G(t, t
) results in a 2π/Ωperiodic time dependence
of the coeﬃcients D
pp
and D
xp
. Averaging the transport coeﬃcients over one period
of the driving is equivalent to the moderate rotatingwave approximation introduced
in Section 4.3.2.
After inserting the noise kernel (3.17) and assuming an Ohmic bath, I(ω) = mγω,
we ﬁnd for D
pp
in an average over one period of the driving,
D
pp
=
1
2
m
∞
¸
n=−∞
c
0
n
(µ
0
+nΩ)
2
coth
(µ
0
+nΩ)
2k
B
T
. (5.52)
This form makes explicit that the diﬀusion D
pp
accounts for the quasienergies (µ
0
+
nΩ). Thus the quasienergy spectrum approach is reﬂected solely by a driving
induced modiﬁcation of the momentum diﬀusion D
pp
.
The evaluation of the cross diﬀusion D
xp
is more complex. Its logarithmic diver
gence is regularized by a Drude cutoﬀ to obtain
D
xp
= −
2π
∞
¸
n=−∞
P
∞
−∞
dω coth
ω
2k
B
T
ω
ω
2
−(µ
0
+nΩ)
2
ω
2
D
ω
2
+ω
2
D
, (5.53)
where P denotes Cauchy’s principal part. The integral in Eq. (5.53) is solved by
contour integration in the upper half plane. Expressing the resulting sums by the
psi function ψ(x) = d ln Γ(x)/dx [92], we obtain
D
xp
= −
π
¸
ψ
1 +
ω
D
2πk
B
T
+C
, (5.54)
where C is the Euler constant. We have neglected terms of the order (µ
0
+nΩ)/ω
D
,
i.e., we have to choose the cutoﬀ ω
D
much larger than the relevant frequencies
µ
0
+nΩ.
Interestingly enough, mγD
xp
coincides with the Drude regularized divergent part
of the stationary momentum variance of a dissipative harmonic oscillator [54,55,60].
In contrast to the FokkerPlanck equation with RWA in the last subsection, the
terms with ∂
x
x and ∂
2
x
are now absent. In addition, the cross diﬀusion D
xp
in
(5.50) is completely diﬀerent, and unrelated to the one in the RWA case (5.45). It
originates from a principal part that has been neglected in the derivation of the
FloquetMarkov equation in (4.18).
5.4.1 Wigner representation and FokkerPlanck equation
In order to achieve a description close to the classical phasespace dynamics, we
discuss the time evolution of the density operator in Wigner representation. Apply
ing the transformations (A.25)–(A.28) to the master equation (5.48), we obtain a
cnumber equation of motion,
∂
t
W(x, p, t) = L(t)W(x, p, t), (5.55)
40 The parametrically driven harmonic oscillator
with the diﬀerential operator
L(t) = −
1
m
p∂
x
+γ∂
p
p +k(t)x∂
p
+γD
pp
∂
2
p
+γD
xp
∂
x
∂
p
. (5.56)
Equation (5.56) has the structure of an eﬀective FokkerPlanck operator. How
ever, for any nonzero D
xp
, the diﬀusion matrix is not positive semideﬁnite; corre
spondingly the FokkerPlancklike equation (5.55) with Eq. (5.56) has no equivalent
Langevin representation.
As is the case for the master equation from which it has been derived, the
coeﬃcients of the FokkerPlanck operator retain the periodicity of the driving, so
that (5.55) has solutions of Floquet form. This fact will be exploited in the following
subsection to construct the solutions.
5.4.2 WignerFloquet solutions
The FokkerPlanck equation for the density operator in Wigner representation,
Eq. (5.55) with Eq. (5.56), oﬀers the opportunity to make full use of the wellknown
and intuitive results for the corresponding classical stochastic system. In particular,
a solution of the FokkerPlanck equation can be obtained directly by solving the
equivalent Langevin equation [45, 94], or by using the formula for the conditional
probability of a Gauss process [94]. In the present case, however, the fact that the
diﬀusion matrix of (5.56) is not positive semideﬁnite requires to take a diﬀerent
route.
Since Eq. (5.55) with Eq. (5.56) represents a diﬀerential equation with time
periodic coeﬃcients, it complies with the conditions of the Floquet theorem. Con
sequently, there exists a complete set of solutions of the form
W
α
(x, p, t) = e
−µαt
u
α
(x, p, t), u
α
(x, p, t) = u
α
(x, p, t +T), (5.57)
henceforth referred to as WignerFloquet functions.
We construct a solution for (5.55) of this form with µ
00
= 0 by the method of
characteristics [95] in Appendix C. In the limit t
0
→−∞, the terms in the ﬁrst line
of (C.18), which contain the initial condition, vanish and we obtain the asymptotic
solution
W
00
(x, p, t) =
1
2π
σ
xx
(t) σ
xp
(t)
σ
xp
(t) σ
pp
(t)
−1/2
exp
−
1
2
x
p
σ
xx
(t) σ
xp
(t)
σ
xp
(t) σ
pp
(t)
−1
x
p
¸
(5.58)
with the variances
σ
xx
(t) =
2γD
pp
m
2
t
−∞
dt
[G(t, t
)]
2
, (5.59)
σ
xp
(t) =
2γD
pp
m
t
−∞
dt
G(t, t
)
∂
∂t
G(t, t
), (5.60)
σ
pp
(t) = −mγD
xp
+ 2γD
pp
t
−∞
dt
¸
∂
∂t
G(t, t
)
2
. (5.61)
5.4 Basisindependent description beyond RWA 41
Note that in (5.59)–(5.61) the diﬀerence in using D
pp
and D = D
pp
+ γD
xp
[see
Eq. (C.14)] is meaningless, since it is a correction of order γ. By inserting the Fourier
representation (5.11) for G(t, t
), one ﬁnds that the variances are asymptotically time
periodic.
Starting from W
00
, we construct further WignerFloquet functions: By solving
the characteristic equations (see Appendix C), we ﬁnd the two timedependent dif
ferential operators
Q
1+
(t) = f
1
(t)∂
x
+m
˙
f
1
(t)∂
p
, (5.62)
Q
2+
(t) = f
2
(t)∂
x
+m
˙
f
2
(t)∂
p
, (5.63)
where the solutions f
i
(t) of the classical equation of motion are given by (5.9). The
operators Q
i+
(t) have the properties
[L(t) −∂
t
, Q
1+
(t)] = [L(t) −∂
t
, Q
2+
(t)] = 0 (5.64)
and
Q
1+
(t +T) = e
(−γ/2+iµ)T
Q
1+
(t), (5.65)
Q
2+
(t +T) = e
(−γ/2−iµ)T
Q
2+
(t). (5.66)
Taking the commutation relation (5.64) into account, the functions
W
nn
(x, p, t) = Q
n
1+
(t) Q
n
2+
(t) W
00
(x, p, t), n, n
= 0, 1, 2, . . . (5.67)
also solve Eq. (5.55).
Due to Eqs. (5.65), (5.66) they are of Floquet structure with the Floquet spec
trum
µ
nn
= (n +n
)γ/2 −i(n −n
)µ. (5.68)
This spectrum is independent of the diﬀusion constants, as expected for an oper
ator of type (5.56) [96], and therefore is the same as in the case of the classical
parametrically driven Brownian oscillator [97].
The expression for the eigenfunctions in the hightemperature limit of the (un
driven) classical Brownian harmonic oscillator in Refs. [96, 98] is also of the struc
ture (5.67). We can recover this solution by inserting the classical diﬀusion constant
mk
B
T and the undriven limit ε →0 for the classical solution, given in Section 5.1.
5.4.3 Inﬂuence of the driving on the master equation
The master equation in operator notation (5.48) and the FokkerPlanck equation
(5.55) given in this section result from a Markov approximation with respect to the
quasienergy spectrum. Nevertheless, they are formally independent of the Floquet
basis. This allows for a detailed analysis of the diﬀerence between the Markovian
approach with respect to the unperturbed spectrum and the quasienergy spectrum
approach beyond mere diﬀerences in representation.
42 The parametrically driven harmonic oscillator
Parametrical driving
The Markov approximation with respect to the unperturbed spectrum can be ob
tained from the (in general more complicated) quasienergy spectrum approach by
replacing the coeﬃcients of friction and diﬀusion by their corresponding limits for
zero driving amplitude ε. We obtain a master equation of the form (5.48) and accord
ingly a FokkerPlanck equation of the form (5.55), where the momentum diﬀusion
coeﬃcient D
pp
is replaced by its limit for ε →0,
D
pp
= lim
ε→0
D
pp
=
1
2
mω
0
coth
ω
0
2k
B
T
. (5.69)
In general D
pp
= D
pp
, which we verify by numerical studies in Section 5.6. Thus
parametric driving of a dissipative harmonic oscillator modiﬁes the momentum dif
fusion in the master equation.
Additional additive driving
The Markovian master equation within the quasienergy spectrum approach under
goes a further modiﬁcation when the parametric oscillator is subject to an additional
additive driving −xF(t), i.e.,
H
F
(t) = H
S
(t) −xF(t). (5.70)
With H
S
(t) being a timeindependent harmonic oscillator, i.e., k(t) = mω
2
0
, the
corresponding Markovian master equation in RWA for the dissipative system was
already given in [26]. Herein we generalize these results for the combined time
dependent system Hamiltonian in (5.70).
It is known that the only eﬀect of the driving force F(t) on the (quasi) energy
spectrum of a parametrically driven harmonic oscillator is an overall level shift
[82]. Thus the level separations remain unaﬀected and we expect no change in the
dissipative part of the master equation (5.48).
The classical equation of motion, which is also obeyed by the interactionpicture
position operator, now reads
m¨ x +k(t)x = F(t), (5.71)
and can be integrated to yield the interactionpicture operators
x
H
(t, t
) = −x
∂G
0
(t, t
)
∂t
+
p
m
G
0
(t, t
) +
1
m
t
t
dt
G
0
(t, t
)F(t
), (5.72)
p
H
(t, t
) = −xm
∂
2
G
0
(t, t
)
∂t ∂t
+p
∂G
0
(t, t
)
∂t
+
t
t
dt
∂G
0
(t, t
)
∂t
F(t
). (5.73)
Thus we obtain a cnumber correction to the interactionpicture position operator
(5.47), given by the third term. After inserting (5.73) into (3.38), the Markovian
5.5 Asymptotics 43
master equation emerges as
˙ = . . . +
i
F(t)[x, ] (5.74)
+
i
∞
0
dτ γ(τ) [x, ]
2
m
t−τ
t
dt
G
0
(t −τ, t
)F(t
). (5.75)
The dots denote the old result for F(t) ≡ 0, given by the right hand side of Eq. (5.48).
The term in the ﬁrst line stems from the reversible part of the master equation (4.3);
the second one is a correction of the driving force due to interaction with the bath.
Thus the equation of motion for the density operator has the structure
˙ = . . . +
i
˜
F(t)[x, ] (5.76)
with an eﬀective total driving force
˜
F(t) = F(t) +
∞
0
dτ γ(τ)
t−τ
t
dt
∂G
0
(t −τ, t
)
∂t
F(t
). (5.77)
Note that the dissipative parts of (5.76) are not aﬀected by the additive driving
force F(t). This makes explicit, that we must use a parametric timedependence to
study diﬀerences in the dissipative parts resulting from the Markov approximation
with respect to the energy spectrum versus the Markov approximation with respect
to the quasienergy spectrum.
With an Ohmic bath, γ(τ) = 2γδ(τ), the inner integral in (5.77) vanishes and
we obtain
˜
F(t) = F(t). Thus in contrast to an explicit parametric time dependence
k(t) in the quadratic part of the Hamiltonian, the time dependence of an additive
force, in this case, does not change the Markovian master equation of the dissipative
system.
5.5 Asymptotics
5.5.1 The conservative limit
In contrast to the Markov approximation with RWA in Section 5.3, the variances in
both Markov approximations without RWA still depend on the friction γ. To obtain
the conservative limit γ →0 of these, we insert the Green function (5.11) into (5.59)
and get
σ
xx
(t) = −
γD
pp
2m
2
¸
n,n
c
n
c
n
f
2
1
(t)
e
γt−i[2µ+(n+n
)Ω]t
γ −i[2µ + (n +n
)Ω]
−2f
1
(t)f
2
(t)
e
γt−i(n−n
)Ωt
γ −i(n −n
)Ω
+f
2
2
(t)
e
γt+i[2µ+(n+n
)Ω]t
γ + i[2µ + (n +n
)Ω]
. (5.78)
In the limit of weak damping, γ <[µ +nΩ[ for any integer n, only the case n = n
of the second term in the brackets remains. Note that this condition is violated in
44 The parametrically driven harmonic oscillator
parameter regions where the Floquet index becomes a multiple of Ω, as is the case
along the borderlines of the regions of stability in parameter space (cf. Fig. 5.1).
For the position variance, we get
σ
xx
(t) = B
D
pp
m
2
ξ
0
1
(t)ξ
0
2
(t), (5.79)
where
B =
∞
¸
n=−∞
c
0
n
2
(5.80)
denotes a number of order unity.
In an analogous way, we ﬁnd
σ
xp
(t) = B
D
pp
2m
˙
ξ
0
1
(t)ξ
0
2
(t) + ξ
0
1
(t)
˙
ξ
0
2
(t)
, (5.81)
σ
pp
(t) = BD
pp
˙
ξ
0
1
(t)
˙
ξ
0
2
(t). (5.82)
Besides the prefactor, these variances are the same as for the master equation with
RWA in Section 5.3.
Moreover, in this limit γ →0, all diagonal elements W
nn
(x, p, t) are Floquet func
tions with the quasienergies µ
nn
= 0. However, they are diﬀerent from the Wigner
representation of the stationary solutions (5.35) of the corresponding Schr¨ odinger
equation, which are also solutions of the coherent equation of motion, Eq. (5.55)
with γ = 0. Due to the degeneracy of the Floquet indices, this is no contradiction.
The lim
γ→0
W
nn
(x, p, t) can be viewed as dissipationadapted Floquet functions.
For consistency, we check the positionmomentum uncertainty relation for the
asymptotic solution. It is satisﬁed if the variances fulﬁll the inequality
σ
xx
(t) σ
xp
(t)
σ
xp
(t) σ
pp
(t)
=
D
pp
B
m
2
≥
2
/4, (5.83)
which we have veriﬁed numerically for the case of the Mathieu oscillator.
5.5.2 The hightemperature limit
In the limit of high temperatures k
B
T ω
D
, we expect the FokkerPlanck equation
for the Wigner function to give the Kramers equation for the classical Brownian
motion [97], i.e., an equation of the form (5.55) with diﬀusion constants D
xp
= 0
and D
pp
= mk
B
T.
In the reﬁned approach (Section 5.4), the FokkerPlanck equation is already of
the required structure. With ψ(1) = −C [92], the cross diﬀusion D
xp
vanishes in
the hightemperature limit. For D
pp
, we use coth x = 1/x +O(x) and get
D
pp
= mk
B
T
¸
n
c
0
n
2
(µ
0
+nΩ). (5.84)
With the sum rule (5.8), this reduces to D
pp
= mk
B
T.
5.6 Numerical results 45
In the quasienergy spectrum approach with RWA in Section 5.3, the variances
and diﬀusion constants scale with N + 1/2. This factor, in the hightemperature
limit, reads
N +
1
2
=
¸
n
c
0
n
2
k
B
T
= B
k
B
T
. (5.85)
Therefore the diﬀusion constants D
xx
and D
xp
remain ﬁnite and the FokkerPlanck
operator (5.42) does not approach the Kramers limit for high temperatures. Never
theless the asymptotic variances in RWA coincide for high temperatures, with the
classical result in the limit γ →0.
5.6 Numerical results
In this section, we compare our approximate results to exact ones, obtained from
the pathintegral solution in Ref. [28]. Speciﬁcally, we give the numerical results for
the Mathieu oscillator, i.e., we choose
k(t) = m
ω
2
0
+ε cos Ωt
. (5.86)
By inserting (5.86) and the ansatz (5.6) into (5.4), we obtain the tridiagonal recur
rence relation
εc
n−1
+ 2
ω
2
0
−γ
2
/4 −(µ +nΩ)
2
c
n
+εc
n+1
= 0. (5.87)
From this equation, the classical Floquet index µ and the Fourier coeﬃcients c
n
are
determined numerically by continued fractions [45].
In the ﬁgures, time and driving parameters are given in the units which are
commonly used in mathematical literature [85] to obtain the scaled Mathieu equation
¨ x + (¯ ω
2
0
+ 2¯ ε cos(2
¯
t))x = 0. Variances are plotted in units of the corresponding
groundstate variance for zero driving amplitude (cf. Appendix A).
We showed in Section 5.4 that the inﬂuence of the driving on the master equa
tion results in a modiﬁcation of the momentum diﬀusion. Figure 5.2 compares the
diﬀusion coeﬃcient D
pp
, obtained from a Markov approximation with respect to the
unperturbed spectrum, to the diﬀusion coeﬃcient D
pp
, which results from the quasi
energy spectrum approach. The numerical values are given in units of the classical
momentum diﬀusion coeﬃcient mk
B
T. The parameters ω
2
0
and ε are varied along
the full line in the inset. Note that within the unstable regimes, perturbation theory
is not valid. Nevertheless, Eq. (5.52) gives a smooth interpolation. The discrepan
cies become most signiﬁcant for strong driving and large ω
2
0
. Both for low driving
amplitude ε <ω
2
0
and high temperature T ω
0
/k
B
, the diﬀerence vanishes.
The variances σ
xx
(t) and σ
pp
(t) of the Markov approximations without RWA are
compared to the exact results [28] in the Figs. 5.3a and 5.3b. The chosen driving pa
rameters ω
2
= 6.5 Ω
2
and ε = 7 Ω
2
lie inside the ﬁfth stable zone (µ = 4.53513 Ω/2).
The temperature k
B
T = 0.5 Ω is suﬃciently large, but with quantum eﬀects still
46 The parametrically driven harmonic oscillator
0 10 20 30
0
1
2
3
4
5
6
ε = ω
2
0
[Ω
2
/2]
D
p
p
[
m
k
B
T
]
k
B
T/¯ hΩ = 0.5
ε = ω
2
0
0 5 10 15
0
10
20
ε [Ω
2
/2]
ω
2 0
[
Ω
2
/
4
]
Figure 5.2: The diﬀusion constants D
pp
for the simple (dotted) and D
pp
for the im
proved (dashed) Markov approximation in units of the classical diﬀusion constant mk
B
T
for k
B
T = 0.5 Ω. The parameters ω
2
0
and ε are indicated by the full line in the inset
(cf. Fig. 5.1).
appreciable. We note that the improved Markovian treatment in Section 5.4, which
accounts for the quasienergy diﬀerences, agrees better with the exact prediction. In
the Figure we depict asymptotic times t > 100/Ω, where transient eﬀects have al
ready decayed. The asymptotic covariance elements retain the periodicity T = 2π/Ω
of the external driving. The relative error
η
xx
(t) =
σ
Markov
xx
(t) −σ
exact
xx
(t)
σ
exact
xx
(t)
(5.88)
of the position variance for these two Markov approximations is depicted in Fig. 5.4.
For the chosen parameters it is reduced by the use of the improved Markov scheme by
approximately 30%. Note that the maximal deviations do not occur in the extrema,
but happen to occur in the regions with negative slope.
As depicted in Fig. 5.5, the quality of both Markov approximations worsens with
increasing dissipation strength γ. This reﬂects the breakdown of the weakcoupling
approach.
Results for the Markovian treatment within RWA, given in Section 5.3, are de
picted for the position variance σ
xx
(t) in Fig. 5.6. The driving parameters are the
same as in Fig. 5.3. For this example, the quality of agreement to the exact result is
similar for both Markov approximations. Nevertheless, the solution without RWA
yields—up to a scale—a better overall agreement with the exact behavior over a full
driving period T.
5.6 Numerical results 47
50 51 52 53 54 55
0
2
4
6
8
10
12
14
t [Ω/2]
σ
x
x
[
¯h
/
2
m
ω
0
]
T
(a)
50 51 52 53 54 55
0
2
4
6
8
10
12
t [Ω/2]
σ
p
p
[
m
¯h
ω
0
/
2
]
(b)
Figure 5.3: The asymptotic vari
ances σ
xx
(t) (a) and σ
pp
(t) (b)
with period T = 2π/Ω for the
simple (dotted) and the improved
(dashed) Markov approximation,
compared to the exact result (full
line) for the parameters ε = 7 Ω
2
,
ω
2
0
= 6.5 Ω
2
, k
B
T = 0.5 Ω and
γ = Ω/20. The scaled driving
period T = 2π/Ω is indicated in
panel (a).
50 51 52 53 54 55
0.0
0.1
0.2
0.3
0.4
0.5
0.6
t [Ω/2]
η
x
x
Figure 5.4: Relative error η
xx
(t)
for the position variances of Fig.
5.3a.
48 The parametrically driven harmonic oscillator
0.2 0.4 0.6 0.8 1.0
0
2
4
6
8
γ [Ω/2]
σ
x
x
t
[
¯h
/
2
m
ω
0
]
Figure 5.5: The time averaged
variance σ
xx
t
for the simple (dot
ted) and the improved (dashed)
Markov approximation, compared
to the exact result (full line) for the
parameters ε = 7 Ω
2
, ω
2
0
= 6.5 Ω
2
and k
B
T = 0.5 Ω.
50 51 52 53 54 55
0
2
4
6
8
10
12
t [Ω/2]
σ
x
x
[
¯h
/
2
m
ω
0
]
Figure 5.6: Position variances ob
tained with the Markov approxi
mation with respect to the quasi
energy spectrum with (dotted) and
without (dashed) RWA, compared
to the exact result (full line) for
γ = Ω/20 and k
B
T = 0.5 Ω. The
driving parameters are ε = 7 Ω
2
and ω
2
0
= 6.5 Ω
2
.
5.7 Conclusion
The principal distinction to be made among possible Markovian approaches to the
driven dissipative dynamics, refers to the degree to which changes in dynamical
and spectral properties of the central system due to the driving are taken into
account. In the crudest treatment introduced in Section 4.1, where the dissipative
terms in the master equation are derived ignoring the explicit time dependence of
the Hamiltonian, and the driving only appears in the coherent term. An improved
master equation results from the FloquetMarkov scheme which we obtained in
Section 4.2 by coupling the central system and the driving as one whole to the heat
bath. The energydomain quantity relevant for all subsequent developments is then
the quasienergy spectrum, obtained within the Floquet formalism, instead of the
unperturbed spectrum. In the time domain, the quantities entering the dissipative
terms of the master equation, such as Heisenbergpicture operators of the central
system, gain an explicit time dependence with the periodicity of the driving.
Besides the diﬀerences in representation, the use of the improved FloquetMarkov
5.7 Conclusion 49
approximation in Section 5.4 mainly results in a modiﬁed momentum diﬀusion that
depends on the quasienergy spectrum instead of the unperturbed spectrum of the
central system. The diﬀerence becomes signiﬁcant in the limits of strong driving
amplitude and low temperature. An additive timedependent external force, applied
in addition to or instead of the parametric driving, undergoes a renormalization
which vanishes, however, in the case of an Ohmic bath.
Even within the improved Markov approach, ﬁner levels of approximation can
be distinguished. A signiﬁcant simpliﬁcation of the master equation is achieved
by a rotatingwave approximation, i.e. here, by neglecting reservoirinduced virtual
transitions between Floquet states of the central system that violate quasienergy
conservation. The resulting master equation has Lindblad form, with creation and
annihilation operators acting on Floquet states, and thus manifestly generates a
dynamical semigroup. This is not the case if the RWA is avoided. Apparently a
drawback, the lack of a Lindblad structure in the master equation without RWA
faithfully reﬂects the failure of the Markov approximation on short time scales.
An analogous situation as with the Lindblad form of the master equation arises
with its Floquet structure. If all coeﬃcients are at most periodically time dependent,
then the equation of motion for the reduced density operator complies with the
conditions for applicability of the Floquet theorem. As a consequence, the solutions
can be cast in Floquet form, i.e., can be written as eigenfunctions of a generalized
nonunitary Floquet operator that generates the evolution of the density operator
over a single period. Since all variants of the Markov approximation discussed here
truncate the memory of the central system on time scales shorter than the period of
the driving, the corresponding master equations have Floquet structure throughout.
The exact pathintegral solution, in contrast, allows for memory eﬀects of unlimited
duration and thereby generally prevents the consistent deﬁnition of a propagator
over a single period.
Additional insight is gained by discussing the dynamics in terms of phasespace
distributions, speciﬁcally in terms of the Wigner representation of the density oper
ator and its equation of motion. In this representation, the Floquet formalism is a
useful device to construct and classify solutions. Since all FokkerPlanck equations
obtained are time periodic, as are the corresponding master equations, their solu
tions may be written as eigenstates of a WignerFloquet operator (the FokkerPlanck
operator evolving the Wigner function, integrated over a single period), or Wigner
Floquet states in short. They represent the quasiprobability distributions closest to
the Floquet solutions of the corresponding classical FokkerPlanck equation.
WignerFloquet states with Floquet index zero correspond to asymptotic solu
tions. They are not literally stationary but retain the periodic time dependence of
the driving. Since we are here dealing with a linear system, the centers of gravity
of the asymptotic quasiprobability distributions follow the corresponding classical
limit cycles. In the case of parametric driving, these limit cycles are trivial and
correspond to a ﬁxed point at the origin. A time dependence arises only by the
periodic variation of the shape of the asymptotic distributions.
Concluding from a numerical comparison of certain dynamical quantities, for the
50 The parametrically driven harmonic oscillator
speciﬁc case of the Mathieu oscillator, the attributes “simple” and “improved” for
the two basic Markovian approaches prove adequate. Results for the Markov ap
proximation based on the quasienergy spectrum show consistently better agreement
with the exact pathintegral solution than those for the Markov approximation with
respect to the unperturbed spectrum. However, even in parameter regimes where
the respective approximations are expected to become problematic, the diﬀerences
in quality are not huge and the agreement with the exact solution is generally good.
Technical advantages of the Markov approximation in general and of its various
ramiﬁcations—easy analytical and numerical tractability, desirable formal proper
ties such as Floquet or Lindblad form of the master equation—can justify to tolerate
their quantitative inaccuracy.
6
The harmonically driven
doublewell potential
In this chapter we use the FloquetMarkov scheme to investigate the interplay of
chaos and dissipation in a bistable quantum system. The harmonically driven quar
tic double well will serve as our working model. In Section 6.1 we introduce its
Hamiltonian and the underlying symmetries. Moreover, we brieﬂy review coherent
driven tunneling as well as its modiﬁcation caused by the inﬂuence of classical chaos.
For moderate driving near the classical resonances, chaos already plays a signif
icant role for the classical dynamics although the motion near the bottom of the
wells is still regular. Thus, we have a mixed phase space, where the coexistence
of regular and chaotic regions leads to a variety of uncommon coherence phenom
ena. Most prominent among them is chaotic tunneling [13–17, 30–33, 99–105], the
coherent exchange of probability between symmetryrelated regular islands that are
separated by a chaotic layer, not by a static potential barrier. Chaotic tunneling
comes about by an interplay of classical nonlinear, typically bistable, dynamics and
quantum coherence. Tunneling is extremely sensitive to any disruption of coherence
as it occurs due to the unavoidable coupling to the environment: In presence of
dissipation, coherent tunneling becomes a transient that fades out on the way to an
asymptotic state [11, 12].
The quasispectrum associated with chaotic tunneling exhibits a characteristic
feature: Quasienergies of chaotic singlets intersect tunnel doublets which are sup
ported by regular tori. We study coherent and dissipative chaotic tunneling in the
vicinity of such singletdoublet crossings in Section 6.2. While in the coherent case
the dynamics is well described in a threestate approximation, the coupling to the
environment indirectly couples the three states to all other states. On the basis of
numerical results for the full driven double well with dissipation, we reveal the lim
itations of the threelevel approximation and identify additional features of the full
dynamics not covered by it. In particular, we consider the longtime asymptotics
and the phasespace structure associated with it.
Switching on friction has a dramatic consequence for the classical phase space:
A volume element contracts exponentially in time and therefore all trajectories con
verge towards a submanifold of phase space with zero volume, the socalled attrac
tor [3]. Depending on friction strength and details of the system, this attractor
may be of quite diﬀerent nature. If the dissipative dynamics is also chaotic, the
attractor has in general fractal geometry—it forms a socalled strange attractor; for
suﬃciently strong friction, the attractor typically shrinks to a limit cycle or a set of
isolated ﬁxed points. On a quantum level, the structures associated with classical at
tractors are smeared out on a scale but leave their trace in the asymptotic state of
the corresponding dissipative quantum map [106]. We study the classicalquantum
correspondence of the asymptotic state in Section 6.3.
52 The harmonically driven doublewell potential
−x
0
x
0
x
0
−E
B
V
(
x
,
t
)
ω
0
E
B
t = (n + 1/2)π/Ω
t = 2πn/Ω
Figure 6.1: Sketch of the driven dou
ble well potential described by the
timedependent Hamiltonian (6.1) at
various times.
6.1 The model
As a prototypical working model, we consider the quartic double well with a spatially
homogeneous driving force, harmonic in time. It is deﬁned by the Hamiltonian
H(t) = H
DW
+H
F
(t), (6.1)
H
DW
=
p
2
2m
−
1
4
mω
2
0
x
2
+
m
2
ω
4
0
64E
B
x
4
, (6.2)
H
F
(t) = Sxcos(Ωt). (6.3)
The potential term of the static bistable Hamiltonian H
DW
possesses two minima at
x = ±x
0
, x
0
= (8E
B
/mω
2
0
)
1/2
, separated by a barrier of height E
B
(cf. Fig. 6.1). The
parameter ω
0
denotes the (angular) frequency of small oscillations near the bottom
of a well. Apart from mere scaling, the classical phase space of H
DW
only depends
on the presence or absence, and the signs, of the x
2
and the x
4
term. Besides that, it
has no free parameter. This is obvious from the scaled form of the classical equations
of motion,
˙
¯ x = ¯ p, (6.4)
˙
¯ p =
1
2
¯ x −
1
2
¯ x
3
−F cos(
¯
Ω
¯
t), (6.5)
where the dimensionless quantities ¯ x, ¯ p and
¯
t are given by x/x
0
, p/mω
0
x
0
and ω
0
t,
respectively. The inﬂuence of the driving on the classical phasespace structure is
fully characterized by the rescaled amplitude and frequency of the driving,
F =
S
8mω
2
0
E
B
,
¯
Ω =
Ω
ω
0
. (6.6)
This implies that the classical dynamics is independent of the barrier height E
B
.
6.1 The model 53
In the quantummechanical case, however, this holds no longer true: The ﬁnite
size of Planck’s constant results in a ﬁnite number of doublets with energy below
the barrier top. It is approximately given by
D =
E
B
ω
0
, (6.7)
and distinguishes the semiclassical from the deep quantum regime. This is evident
from the classical scales for position, x
0
, and momentum, mω
0
x
0
, introduced above:
The corresponding action scale is mω
0
x
2
0
and therefore, the positionmomentum
uncertainty relation in the scaled phase space (¯ x, ¯ p) reads
∆¯ x∆¯ p ≥
eﬀ
2
(6.8)
where
eﬀ
=
mω
0
x
2
0
=
1
8D
(6.9)
denotes the eﬀective quantum of action. The classical limits hence amounts to
D →∞.
In the following, we restrict ourselves to moderate driving amplitudes, such that
the variation of the potential at the bottom of the wells is much smaller than the
barrier height. This implies that the bistable character of the potential is retained
at any time.
6.1.1 Symmetries
The model Hamiltonian (6.1) obviously is 2π/Ωperiodic in time, thus possesses
discrete timetranslational invariance. This enables a treatment within the Floquet
Markov scheme, introduced in Chapter 4. In addition, we ﬁnd two more discrete
symmetries, which allow for an improvement of numerical eﬃciency and also for a
classiﬁcation of the Floquet states as even or odd.
Timereversal symmetry
It is well known that the energy eigenfunctions of an (undriven) Hamiltonian which
obeys timereversal symmetry, can be chosen as real [8, 79]. This has, apart from
computational advantages, also direct physical consequences for the level statistics of
quantum systems with chaotic classical counterpart [8, 79]. Timereversal symmetry
is typically broken by a magnetic ﬁeld (recall that a magnetic ﬁeld is described
by an axial vector and changes sign under time reversion) or by an explicit time
dependence of the Hamiltonian. However, for the sinusoidal shape of the driving
together with the initial phase chosen above, timereversal symmetry
T : x →x, p →−p, t →−t (6.10)
is retained and the Floquet Hamiltonian obeys H(t) = H
∗
(−t) [cf. Eq. (2.11)]. If now
φ(x, t) is a Floquet state in position representation with quasienergy , then φ
∗
(x, −t)
54 The harmonically driven doublewell potential
also is a Floquet state with the same quasienergy. This means that we can always
choose the Floquet states by linear combination such that φ(x, t) = φ
∗
(x, −t), which
translates to φ(x, ω) = φ
∗
(x, ω) in the frequency regime, i.e., the Fourier coeﬃcients
of the Floquet states can be chosen real.
Generalized parity
The undriven Hamiltonian H
DW
is invariant under the parity P: x →−x, p →−p,
t → t. This symmetry is destroyed by a linerarly coupled driving ﬁeld. With
the above choice of H
F
(t), however, a more general, dynamical symmetry remains
[10, 107, 108]. It is deﬁned by the operation
P
Ω
: x →−x, p →−p, t →t +π/Ω (6.11)
and represents a generalized parity acting in the extended phase space spanned by
x, p, and phase, i.e., time t mod(2π/Ω) or in the composite Hilbert space 1 ⊗ T ,
respectively. While such a discrete symmetry is of minor importance in classical
physics, its inﬂuence on the quantum mechanical quasispectrum ¦
α
(F)¦ is more
distinct: It devides the composite Hilbert space in an even and an odd subspace,
thus allowing for a classiﬁcation of the Floquet states as even or odd. Quasiener
gies from diﬀerent symmetry classes may intersect, whereas quasienergies with the
same symmetry typically form avoided crossings [79]. However, the fact that the
generalized parity acts on the composite Hilbert space results in a particularity: If
[φ(t)` is e.g. an even Floquet state, then [φ
(1)
(t)` = exp(iΩt)[φ(t)` turns out to be
odd. Thus, two equivalent Floquet states from neighboring Brillouin zones possess
diﬀerent generalized parity. This means that a classiﬁcation of the corresponding
solutions of the Schr¨ odinger equation, [ψ(t)` = exp(−it/)[φ(t)`, as even or odd
requires a restriction to a single Brillouin zone.
The invariance of the system under the generalized parity is also of considerable
help in the numerical treatment of the Floquet matrix (2.53) [16, 100]. To obtain
a complete set of Floquet states, it is suﬃcient to compute all eigenvectors of the
Floquet Hamiltonian in the even subspace whose eigenvalues lie in the ﬁrst two
Brillouin zones. The even Floquet states are given by the eigenvectors of H
e
from
the ﬁrst Brillouin zone; the odd Floquet states are obtained by shifting the (even)
ones from the second to the ﬁrst Brillouin zone, which changes their generalized
parity. Thus, we have to diagonalize the even supermatrix
H
e
=
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
E
e
+ 2Ω X
eo
0 0 0
X
eo
E
o
+Ω X
oe
0 0
0 X
oe
E
e
X
eo
0
0 0 X
eo
E
o
−Ω X
oe
0 0 0 X
oe
E
e
−2Ω
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
, (6.12)
6.1 The model 55
which for the same number of Floquet channels has only half the dimension of the
original Floquet matrix (2.53). The matrices
E
e
=
¸
¸
¸
¸
E
0
0 0
0 E
2
0
0 0 E
4
.
.
.
.
.
.
.
.
.
.
.
.
, E
o
=
¸
¸
¸
¸
E
1
0 0
0 E
3
0
0 0 E
5
.
.
.
.
.
.
.
.
.
.
.
.
, (6.13)
X
eo
=
S
2
¸
¸
¸
¸
x
0,1
x
0,3
x
0,5
x
2,1
x
2,3
x
2,5
x
4,1
x
4,3
x
4,5
.
.
.
.
.
.
.
.
.
.
.
.
, X
oe
=
S
2
¸
¸
¸
¸
x
1,0
x
1,2
x
1,4
x
3,0
x
3,2
x
3,4
x
5,0
x
5,2
x
5,4
.
.
.
.
.
.
.
.
.
.
.
.
, (6.14)
which are part of the supermatrix H
e
, denote the undriven Hamiltonian H
DW
and
the coupling to the driving ﬁeld H
1
= Sx/2, decomposed into the even and odd
eigenstates of H
DW
.
6.1.2 Tunneling, driving, and dissipation
With the driving H
F
(t) switched oﬀ, the classical phase space generated by H
DW
exhibits the constituent features of a bistable Hamiltonian system. There is a sepa
ratrix at E = 0. It forms the border between two sets of trajectories: One set, with
E < 0, comes in symmetryrelated pairs, each partner of which oscillates in either
one of the two potential minima. The other set consists of unpaired trajectories,
with E > 0, that encircle both wells in a spatially symmetric fashion.
Due to the integrability of the undriven double well, Eq. (6.2), we can gain a
qualitative picture of its eigenstates from simple torus quantization: The unpaired
tori correspond to singlets with positive energy, whereas the symmetryrelated pairs
below the top of the barrier correspond to degenerate pairs of eigenstates. Neigh
boring pairs are separated in energy approximately by ω
0
, which reﬂects the almost
harmonic potential shape near the bottom of the wells. Exact quantization, however,
predicts that the partners of these pairs have small but ﬁnite overlap. Therefore, the
true eigenstates come in doublets, each of which consists of an even and an odd state,
[Φ
+
n
` and [Φ
−
n
`. The energies of the nth doublet are separated by a small tunnel
splitting ∆
n
. We can always choose the globals phases such that the superpositions
[Φ
R,L
n
` =
1
√
2
[Φ
+
n
` ±[Φ
−
n
`
(6.15)
are localized in the right and the left well, respectively. As time evolves, the states
[Φ
+
n
`, [Φ
−
n
` acquire a relative phase exp(−i∆
n
t/) and [Φ
R
n
`, [Φ
L
n
` are transformed
into one another after a time π/∆
n
. Thus, the particle tunnels forth and back
between the wells with a frequency ∆
n
/. This introduces an additional, purely
quantum mechanical frequencyscale, the tunnel rate ∆
0
/ of a particle which resides
in the groundstate doublet. Typically, tunnel rates are extremely small compared
56 The harmonically driven doublewell potential
to the frequencies of the classical dynamics, all the more in the semiclassical regime
we are interested in.
A driving of the form (6.3), even if its inﬂuence on the classical phase space is
minor, can entail signiﬁcant consequences for the tunnel dynamics: It may enlarge
the tunnel rate by orders of magnitude or even suppress tunneling at all. For adia
batically slow driving, Ω <∆
0
/, tunneling is governed by the timeaverage of the
instantaneous tunnel splitting, which is always larger than its unperturbed value ∆
0
and results in an enhancement of the tunneling rate [107]. If the driving is faster,
∆
0
/
<
∼
Ω < ω
0
, the opposite holds true: The relevant time scale is now given by
the inverse of the quasienergy splitting of the groundstate doublet /[
1
−
0
[. It
has been found [107,109] that in this case for ﬁnite driving amplitude [
1
−
0
[ < ∆
0
,
thus tunneling is always decelerated. It even happens that the quasienergies of the
groundstate doublet (which are of diﬀerent generalized parity) intersect as a func
tion of the driving amplitude F, thus the splitting vanishes and tunneling is brought
to a complete standstill by the purely coherent inﬂuence of the driving [10].
The small energy scales associated with make tunneling extremely sensitive to
any disruption of coherence, as it occurs due to the unavoidable coupling to the
environment. As an immediate consequence, the symmetry underlying the formation
of tunnel doublets is generally broken, and an additional energy scale is introduced,
the eﬀective ﬁnite width attained by each discrete level. Tunneling and related
coherence phenomena are thus rendered transients that occur—if at all—on the way
towards an asymptotic equilibrium state and fade out on a time scale t
decoh
. In
general, this time scale gets shorter for higher temperatures, reﬂecting the growth
of the transition rates (4.23) [53]. However, there exist counterintuitive eﬀects. For
example, for driven tunneling in the vicinity of an exact crossing of the groundstate
doublet, the coherent suppression of tunneling [10, 12, 107] can be stabilized with
higher temperatures [76–78] until levels outside the doublet start to play a role.
So far, we have considered only driving frequencies much smaller than the fre
quency scale ω
0
of the relevant classical resonances, i.e., a parameter regime where
classical motion is predominantly regular. Coherent tunneling is in this case well
described within a twostate approximation [107, 109]. In the dissipative case, how
ever, a twostate approximation of course fails for temperatures k
B
T
>
∼
ω
0
, where
thermal activation to higher doublets becomes relevant.
6.1.3 The onset of chaos
Driving with a frequency Ω ≈ ω
0
has an even stronger inﬂuence on the dynamics of
the bistable system. It enters already on the level of classical mechanics since small
oscillations near the bottom of the wells become resonant and classical chaos comes
into play. This corresponds in a quantum description to resonant multiple excitation
of interdoublet transitions until levels near the top of the barrier are signiﬁcantly
populated.
Increasing the amplitude of the driving from zero onwards has two principal
consequences for the classical dynamics: The separatrix is destroyed as a closed
6.1 The model 57
Ω
∆
0
/¯ h
ω
0
coherent destruction
of tunneling
chaotic tunneling
twolevel description multilevel description
adiabatic energies quasienergies
almost regular
chaos
Figure 6.2: Tunneling phenomena and the according appropriate levels of description
for the nondissipative driven doublewell potential, Eq. (6.1). The bars depict the corre
sponding regimes of the driving frequency Ω. See Section 6.1 for a detailed discussion.
curve and replaced by a homoclinic tangle [110] of stable and unstable manifolds.
As a whole, it forms a chaotic layer in the vicinity and with the topology of the
former separatrix (cf. Fig. 6.6). This opens the way for diﬀusive transport between
the two potential wells. Due to the nonlinearity of the potential, there is an inﬁnite
set of resonances of the driving with the unperturbed motion, both inside and outside
the wells [111,112]. Since the period of the unperturbed, closed trajectories diverges
for E → 0, the resonances accumulate towards the separatrix of the unperturbed
system. By its sheer phasespace area, the ﬁrst resonance (the one for which the
periods of the driving and of the unperturbed oscillation are in a ratio of 1:1) is
prominent among the others and soon (in terms of increasing amplitude F) exceeds
the size of the “orderzero” regular areas near the bottom of each well [16].
Both major tendencies in the evolution of the classical phase space—extension
of the chaotic layer and growth of the ﬁrst resonance—leave their speciﬁc traces
in the quasienergy spectrum. The tunnel doublets characterizing the unperturbed
spectrum for E < 0 pertain to states located on pairs of symmetryrelated quantizing
tori in the regular regions within the wells. With increasing size of the chaotic
layer, the quantizing tori successively resolve in the chaotic sea. The corresponding
doublets disappear as distinct structures in the spectrum as they attain a splitting of
the same order as the mean level separation. The gradual widening of the doublets
proceeds as a smooth function of the driving amplitude [16, 100]. This function
roughly obeys a power law [34, 113, 114]. As soon as a pair of states is no longer
supported by any toruslike manifold, including fractal [115] and vague tori [116],
the corresponding eigenvalues detach themselves from the regular ladder to which
they formerly belonged. They can then ﬂuctuate freely in the spectrum and thereby
“collide” with other chaotic singlets or regular doublets.
The appearance of a regular region, large enough to accommodate several eigen
states, around the ﬁrst resonance introduces a second ladder of doublets into the
spectrum. Size and shape of the ﬁrst resonance vary in a way diﬀerent from the
58 The harmonically driven doublewell potential
(a)
−
c
−
r
+
r
−
c
−
r
(b)
(c) (d)
Figure 6.3: Possible conﬁgura
tions of quasienergy crossings be
tween a chaotic singlet and a reg
ular doublet. Diﬀerent line types
signify diﬀerent parity. See Sec
tion 6.2.1 for the labeling of the
levels. Note that only for con
ﬁgurations (a),(b), the order of
the regular doublet is restored
in passing through the crossing.
In conﬁgurations (c),(d), it is re
versed.
main regular region. The corresponding doublet ladder therefore moves in the spec
trum independently of the doublets that pertain to the main regular region, and
of the chaotic singlets. This gives rise to additional singletdoublet and even to
doubletdoublet encounters.
6.2 Chaotic tunneling near singletdoublet crossings
Near a crossing, level separations deviate vastly, in both directions, from the typical
tunnel splitting (cf. Fig. 6.8, below). This is reﬂected in timedomain phenom
ena ranging from the suppression of tunneling to a strong increase in its rate and
to complicated quantum beats [31–33]. Singletdoublet crossings, in turn, drasti
cally change the nondissipative quasienergy scales and replace the twolevel by a
threelevel structure. As a consequence, the familiar way tunneling fades out in the
presence of dissipation is also signiﬁcantly altered. Near a crossing, the coherent
dynamics can last much longer than for the unperturbed doublet, despite the pres
ence of the same dissipation as outside the crossing, establishing “chaosinduced
coherence.” Depending on temperature, it can also be destroyed on a much shorter
time scale.
For the parameters chosen in our numerical studies, higher resonances are neg
ligible in size. Therefore, the borderline between the chaotic layer along the former
separatrix and the regular regions within and outside the wells is quite sharply de
ﬁned. The “coastal strip” formed by hierarchies of regular islands around higher
resonances remains narrow (cf. Fig. 6.6, below) on a scale of the chosen eﬀective
quantum of action. For the tunneling dynamics, the role of states located in the
border region [102, 103] is therefore not signiﬁcant in our studies.
6.2.1 Threelevel crossings
Among the various types of quasienergy crossings that occur according to the above
scenario, those involving a regular doublet and a chaotic singlet are the most com
6.2 Chaotic tunneling near singletdoublet crossings 59
mon. In order to give a quantitative account of such crossings and the associated
coherent dynamics, and for later reference in the context of the incoherent dynamics,
we shall now discuss them in terms of a simple threestate model, devised much in
the spirit of Ref. [30].
Far to the left of the crossing, we expect the following situation: There is a
doublet of Floquet states
[ψ
+
r
(t)` = e
−i
+
r
t/
[φ
+
r
(t)`, (6.16)
[ψ
−
r
(t)` = e
−i(
+
r
+∆)t/
[φ
−
r
(t)`, (6.17)
with even (superscript +) and odd (−) generalized parity, respectively, residing
on a pair of quantizing tori in one of the regular (subscript r) regions. We have
assumed that the quasienergy splitting (as opposed to the unperturbed splitting)
is
−
r
−
+
r
= ∆ > 0. The global relative phases can be chosen such that the
superpositions
[φ
R,L
(t)` =
1
√
2
[φ
+
r
(t)` ±[φ
−
r
(t)`
(6.18)
are localized in the right and the left well, respectively, and tunnel back and forth
with a frequency ∆/ given by the tunnel splitting in the presence of the driving.
As the third player, we introduce a Floquet state
[ψ
−
c
(t)` = e
−i(
+
r
+∆+∆c)t/
[φ
−
c
(t)`, (6.19)
located mainly in the chaotic (subscript c) layer, so that its timeperiodic part [φ
−
c
(t)`
contains a large number of harmonics. Without loss of generality, its generalized
parity is ﬁxed to be odd. For the quasienergy, we have assumed that
−
c
=
+
r
+∆+∆
c
,
where [∆
c
[ can be regarded as a measure of the distance from the crossing.
The structure of the classical phase space then implies that the mean energy
of the chaotic state should be close to the top of the barrier and far above that of
the doublet. We assume, like for the quasienergies, a small splitting of the mean
energies pertaining to the regular doublet, [E
−
r
−E
+
r
[ <E
−
c
−E
±
r
.
In order to model an avoided crossing between [φ
−
r
` and [φ
−
c
`, we suppose that
there is a nonvanishing ﬁxed matrix element
b ≡ ''φ
−
r
[H
DW
[φ
−
c
`` > 0. (6.20)
For the singletdoublet crossings under study, we typically ﬁnd that ∆ < b < Ω.
Neglecting the coupling with all other states, we model the system by the threestate
(subscript 3s) Floquet Hamiltonian
H
3s
=
+
r
+
¸
0 0 0
0 ∆ b
0 b ∆ + ∆
c
, (6.21)
60 The harmonically driven doublewell potential
in the threedimensional Hilbert space spanned by ¦[φ
+
r
(t)`, [φ
−
r
(t)`, [φ
−
c
(t)`¦. Its
Floquet states read
[φ
+
0
(t)` = [φ
+
r
(t)`,
[φ
−
1
(t)` =
[φ
−
r
(t)` cos β −[φ
−
c
(t)` sin β
, (6.22)
[φ
−
2
(t)` =
[φ
−
r
(t)` sin β +[φ
−
c
(t)` cos β
.
Their quasienergies are
+
0
=
+
r
,
−
1,2
=
+
r
+ ∆ +
1
2
∆
c
∓
1
2
∆
2
c
+ 4b
2
, (6.23)
and the mean energies are approximately given by
E
+
0
= E
+
r
,
E
−
1
= E
−
r
cos
2
β +E
−
c
sin
2
β, (6.24)
E
−
2
= E
−
r
sin
2
β +E
−
c
cos
2
β,
where contributions of the matrix element b have been neglected. The angle β
describes the mixing between the Floquet states [φ
−
r
` and [φ
−
c
` and is a measure of
the distance to the avoided crossing. By diagonalizing the Hamiltonian (6.21), we
obtain
2β = arctan
2b
∆
c
, 0 < β <
π
2
. (6.25)
For β → π/2, corresponding to −∆
c
b, we retain the situation far left of the
crossing, as outlined above, with [φ
−
1
` ≈ [φ
−
c
`, [φ
−
2
` ≈ [φ
−
r
`. To the far right of
the crossing, i.e., for β → 0 or ∆
c
b, the exact eigenstates [φ
−
1
` and [φ
−
2
` have
interchanged their identity with respect to the phasespace structure [31–33]. Here,
we have [φ
−
1
` ≈ [φ
−
r
` and [φ
−
2
` ≈ [φ
−
c
`. The mean energy is essentially determined
by the phasespace structure. Therefore, there is also an exchange of E
−
1
and E
−
2
in an exact crossing, cf. Eq. (6.24), while E
+
0
remains unaﬀected (Fig. 6.4b). The
quasienergies
+
0
and
−
1
must intersect close to the avoided crossing of
−
1
and
−
2
(Fig. 6.4a). Their crossing is exact, since they pertain to states with opposite parity
(cf. Fig. 6.3a,b).
In order to illustrate the above threestate model and to demonstrate its ade
quacy, we have numerically studied a singletdoublet crossing that occurs for the
doublewell potential, Eq. (6.1), with D = 4, at a driving frequency Ω = 0.982 ω
0
and amplitude F = 0.015029 (Fig. 6.5). The phasespace structure of the participat
ing Floquet states (Figs. 6.6, 6.7) meets the assumptions of our threestate theory.
A comparison of the appropriately scaled threestate theory (Fig. 6.4) with this real
singletdoublet crossing (Fig. 6.5) shows satisfactory agreement. Note that in the
real crossing, the quasienergy of the chaotic singlet decreases as a function of F, so
that the exact crossing occurs to the left of the avoided one. This numerical ex
ample also shows that the idealized threestate model is not always strictly correct.
Following the global tendency of widening of the splittings with increasing driv
ing amplitude [16, 34, 114], it may well happen that even far away from a crossing,
6.2 Chaotic tunneling near singletdoublet crossings 61
10 5 0 5 10
4b
2b
0
2b
4b
∆
c
/b
q
u
a
s
i
e
n
e
r
g
y
(a)
+
0
∆
c
−
1
−
2
∆
2b
10 5 0 5 10
D
0
∆
c
/b
m
e
a
n
e
n
e
r
g
y
(b)
E
+
0
E
−
c
E
−
r
E
−
1
E
−
2
Figure 6.4: A singletdoublet crossing, according to a threestate model (6.21) in terms
of the dependence of the quasienergies (a) and the mean energies (b) on the coupling
parameter ∆
c
/b. Unperturbed energies are marked by dotted lines, the energies for the
case with coupling by full lines for even and dashed lines for odd states.
0.014 0.015 0.016
0.001
0.0
0.001
(
α
−
+ 0
)
/
¯h
ω
0
F
(a)
−
2
−
+
0
−
1
−
+
0
a b c
0.014 0.015 0.016
3
2
1
0
1
2
E
α
/
¯h
ω
0
F
(b)
E
+
0
E
−
2
E
−
1
doublets
Figure 6.5: Singletdoublet crossing found numerically for the driven double well,
Eq. (6.1), at D = 4 and Ω = 0.982 ω
0
, in terms of the dependence of the quasiener
gies (a) and the mean energies (b) on the driving amplitude F. Values of the driving
amplitude used in Fig. 6.9 are marked by dotted vertical lines. Full and dashed lines
indicate energies of even and odd states, respectively. Bold lines give the mean energies
of the chaotic singlet and the groundstate doublet depicted in panel (a).
62 The harmonically driven doublewell potential
1.5 1.0 0.5 0.0 0.5 1.0 1.5
0.5
0.0
0.5
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
x/x
0
p
/
m
ω
0
x
0
Figure 6.6: Stroboscopic clas
sical phasespace portraits, at
t = 2πn/Ω, of the harmonical
ly driven quartic double well,
Eq. (6.1). The driving parame
ters F = 0.015, Ω = 0.982 ω
0
,
are chosen at the the center of
the singletdoublet crossing un
der study.
1.5 1.0 0.5 0.0 0.5 1.0 1.5
0.5
0.0
0.5
x/x
0
p
/
m
ω
0
x
0
(a)
1.5 1.0 0.5 0.0 0.5 1.0 1.5
0.5
0.0
0.5
x/x
0
p
/
m
ω
0
x
0
(b)
Figure 6.7: Contour plots of
the Husimi functions for the
Floquet states φ
−
1
≈ φ
−
r
(a)
and φ
−
2
≈ φ
−
c
(b) of the har
monically driven quartic dou
ble well, Eq. (6.1), at strobo
scopic times t = 2πn/Ω. The
driving parameters F = 0.014,
Ω = 0.982 ω
0
, are in suﬃcient
distance to the singletdoublet
crossing such that the mixing
between the regular and the
chaotic state is negligible. The
rectangle in the lower left cor
ner depicts the size of the eﬀec
tive quantum of action
eﬀ
.
6.2 Chaotic tunneling near singletdoublet crossings 63
0.0 0.005 0.01 0.015 0.02 0.025
10
9
10
8
10
7
10
6
10
5
10
4
10
3
10
2
F
∆
n
/
¯h
ω
0 n = 0
n = 1
n = 2
avoided
exact
Figure 6.8: Splitting of the lowest doublets for D = 4 and Ω = 0.982 ω
0
. The arrows
indicate the locations of the exact and the avoided crossing within a threelevel crossing
of the type sketched in Fig. 6.3a.
the doublet splitting does not exactly return to its value on the opposite side (see
Fig. 6.8). It is even possible that an exact crossing of
+
0
and
−
1
does not take place
at all in the vicinity of the crossing. In that case, the relation of the quasienergies
in the doublet gets reversed via the crossing (Fig. 6.3c,d). Nevertheless, the above
scenario captures the essential features.
To study the dynamics of the tunneling process, we focus on the state
[ψ(t)` =
1
√
2
e
−i
+
0
t/
[φ
+
0
(t)` + e
−i
−
1
t/
[φ
−
1
(t)` cos β + e
−i
−
2
t/
[φ
−
2
(t)` sinβ
. (6.26)
It is constructed such that at t = 0, it corresponds to the decomposition of [φ
R
`
in the basis (6.22) at ﬁnite distance from the crossing. Therefore, it is initially
localized in the regular region in the right well and follows the time evolution under
the Hamiltonian (6.21). From Eqs. (6.18), (6.22), we ﬁnd the probabilities for its
evolving into [φ
R
`, [φ
L
`, or [φ
c
`, respectively, to be
P
R
(t) = ['φ
R
(t)[ψ(t)`[
2
=
1
2
1 + cos
(
−
1
−
+
0
)t
cos
2
β + cos
(
−
2
−
+
0
)t
sin
2
β
+
¸
cos
(
−
1
−
−
2
)t
−1
cos
2
β sin
2
β
,
P
L
(t) = ['φ
L
(t)[ψ(t)`[
2
(6.27)
=
1
2
1 −cos
(
−
1
−
+
0
)t
cos
2
β −cos
(
−
2
−
+
0
)t
sin
2
β
64 The harmonically driven doublewell potential
0 10
5
2 10
5
3 10
5
0.0
0.2
0.4
0.6
0.8
1.0
t
n
= 2πn/Ω [1/ω
0
]
P
(
t
n
)
(a)
P
R
P
L
P
c
0 10
5
2 10
5
3 10
5
0.0
0.2
0.4
0.6
0.8
1.0
t
n
= 2πn/Ω [1/ω
0
]
P
(
t
n
)
(b)
0 10
5
2 10
5
3 10
5
0.0
0.2
0.4
0.6
0.8
1.0
P
(
t
n
)
t
n
= 2πn/Ω [1/ω
0
]
(c)
Figure 6.9: Stroboscopic time
evolution of a state initially lo
calized in the right well, in the
vicinity of the singletdoublet
crossing shown in Fig. 6.5, in
terms of the probabilities to be
in the right well (which here is
identical to the return proba
bility, marked by full lines), in
the reﬂected state in the left
well (dashed), or in the chaotic
state ψ
c
(dotted). Parame
ter values are as in Fig. 6.5,
and F = 0.0145 (a), 0.0149 (b),
0.015029 (c).
6.2 Chaotic tunneling near singletdoublet crossings 65
+
¸
cos
(
−
1
−
−
2
)t
−1
cos
2
β sin
2
β
,
P
c
(t) = ['φ
c
(t)[ψ(t)`[
2
=
¸
1 −cos
(
−
1
−
−
2
)t
cos
2
β sin
2
β.
We discuss the coherent dynamics of the threestate model for diﬀerent distances to
the crossing and illustrate it by numerical results for the real crossing introduced
above.
In suﬃcient distance from the crossing, there is only little mixing between the
regular and the chaotic states, i.e., sin β < 1 or cos β < 1. The tunneling process
then follows the familiar twostate dynamics involving only [φ
+
r
` and [φ
−
r
`, with
tunnel frequency ∆/ (Fig. 6.9a).
Close to the avoided crossing, cos β and sin β are of the same order of magnitude,
and [φ
−
1
`, [φ
−
2
` become very similar to one another. Both now have support in the
chaotic layer as well as in the symmetryrelated regular regions and thus are of a
hybrid nature. Here, the tunneling involves all the three states and must at least
be described by a threelevel system. The exchange of probability between the two
regular regions proceeds via a “stopover” in the chaotic region [15, 30–33]. The
three quasienergy diﬀerences that determine the time scales of this process are in
general all diﬀerent, leading to complicated beats (Fig. 6.9b).
However, for ∆
c
= −2∆, the two quasienergies
−
1
−
+
0
and
+
0
−
−
2
are de
generate. At this point, which marks the center of the crossing, the number of
diﬀerent frequencies in the threelevel dynamics reduces to two again. This restores
the familiar coherent tunneling in the sense that there is again a simple periodic
exchange of probability between the regular regions [31–33]. However, the rate is
much larger if compared to the situation far oﬀ the crossing, and the chaotic region
is now temporarily populated during each probability transfer, twice per tunneling
cycle (Fig. 6.9c).
6.2.2 Dissipative chaosassisted tunneling
The crucial eﬀect of dissipation on a quantum system is the disruption of coherence:
a coherent superposition evolves into an incoherent mixture. Thus, phenomena
based on coherence, such as tunneling, are rendered transients that fade out on
a ﬁnite time scale t
decoh
. In general, for driven tunneling in the weakly damped
regime, this time scale gets shorter for higher temperatures, reﬂecting the growth of
transition rates [53]. However, in the vicinity of an exact crossing of the groundstate
quasienergies, the coherent suppression of tunneling [10, 12, 107] can be stabilized
with higher temperatures [76–78] and increasing friction [57, 58] until levels outside
the doublet start to play a role. We have studied dissipative chaosassisted tunneling,
using again the real singletdoublet crossing introduced in Sec. 6.2.1 (see Fig. 6.5)
as our working example. The time evolution has been computed numerically by
iterating the dissipative quantum map (4.31) for the improved master equation in
66 The harmonically driven doublewell potential
0 10
5
2 10
5
3 10
5
4 10
5
5 10
5
0.0
0.2
0.4
0.6
0.8
1.0
P
(
t
n
)
t
r
ρ
2
(
t
n
)
t
n
= 2πn/Ω [1/ω
0
]
(a)
0 2 10
4
0.98
1.0
0 10
5
2 10
5
3 10
5
4 10
5
5 10
5
0.0
0.2
0.4
0.6
0.8
1.0
P
(
t
n
)
t
r
ρ
2
(
t
n
)
t
n
= 2πn/Ω [1/ω
0
]
(b)
Figure 6.10: Occupation pro
babilities as in Fig. 6.9a,c, but in
the presence of dissipation. The
dashdotted line shows the time
evolution of tr
2
. The parameter
values are D = 4, Ω = 0.982 ω
0
,
γ = 10
−6
ω
0
, k
B
T = 10
−4
ω
0
,
and F = 0.0145 (a), 0.015029
(b). The inset in (a) is a blow
up of the rectangle in the upper
left corner of that panel.
0 5 10
6
10
7
0.0
0.2
0.4
0.6
0.8
1.0
P
R
(
t
n
)
t
r
ρ
2
(
t
n
)
t
n
= 2πn/Ω [1/ω
0
]
Figure 6.11: Time evolution of
the return probability P
R
(full
line) and the coherence function
tr
2
(dashdotted) during loss
and regain of coherence. The
parameter values are as in Fig.
6.10b.
6.2 Chaotic tunneling near singletdoublet crossings 67
moderate rotatingwave approximation, Eq. (4.22). As an initial condition, we have
chosen the density operator (0) = [φ
R
`'φ
R
[, i.e. a state localized in the right well.
In the vicinity of a singletdoublet crossing, the tunnel splitting increases signiﬁ
cantly—the essence of chaosassisted tunneling. During the tunneling, the chaotic
singlet becomes populated periodically with frequency [
−
2
−
−
1
[/, cf. Eq. (6.27)
and Fig. 6.9. The high mean energy of this singlet results in an enhanced decay of
coherence at times when [φ
c
` is populated (Fig. 6.10). For the relaxation towards the
asymptotic state, also the slower transitions within doublets are relevant. Therefore,
the corresponding time scale t
relax
can be much larger than t
decoh
(Fig. 6.11).
To obtain quantitative estimates for the dissipative time scales, we approximate
t
decoh
by the decay rate of tr
2
, a measure of coherence (see Appendix B.2), averaged
over a time t
p
,
1
t
decoh
= −
1
t
p
tp
0
dt
d
dt
tr
2
(t
) (6.28)
=
1
t
p
tr
2
(0) −tr
2
(t
p
)
. (6.29)
Because of the stepwise decay of the coherence (Fig. 6.10), we have chosen the prop
agation time t
p
as an nfold multiple of the duration 2π/[
−
2
−
−
1
[ of the chaotic
beats. For this procedure to be meaningful, n should be so large that the coherence
decays substantially during the time t
p
(in our numerical studies to a value of ap
proximately 0.9). The time scale t
relax
of the approach to the asymptotic state is
given by the reciprocal of the smallest real part of the eigenvalues of the dissipative
kernel.
Outside the singletdoublet crossing we ﬁnd that the decay of coherence and the
relaxation take place on roughly the same time scale (Fig. 6.12). At F ≈ 0.013, the
chaotic singlet induces an exact crossing of the groundstate quasienergies (see Fig.
6.8), resulting in a stabilization of coherence with increasing temperature. At the
center of the avoided crossing, the decay of coherence becomes much faster and is
essentially independent of temperature. This indicates that transitions from states
with mean energy far above the ground state play a crucial role.
6.2.3 Asymptotic state
As the dynamics described by the master equation (4.3) is dissipative, it converges in
the longtime limit to an asymptotic state
∞
(t). In general, this attractor remains
time dependent but shares all the symmetries of the central system, i.e. here, period
icity and generalized parity. However, the coeﬃcients of the master equation (4.22)
for the matrix elements
αβ
, valid within the moderate rotatingwave approxima
tion, are time independent and so the asymptotic solution also is. This means that
we have eliminated the explicit time dependence of the attractor by representing it
in the Floquet basis and introducing a mild rotatingwave approximation.
To gain some qualitative insight into the asymptotic solution, we focus on the
68 The harmonically driven doublewell potential
0.01 0.012 0.014 0.016 0.018 0.02
10
6
10
7
10
8
k
B
T = 10
−4
¯ hω
0
k
B
T = 10
−3
¯ hω
0
k
B
T = 10
−2
¯ hω
0
F
t
d
e
c
o
h
[
1
/
ω
0
]
(a)
0.01 0.012 0.014 0.016 0.018 0.02
10
6
10
7
10
8
k
B
T = 10
−4
¯ hω
0
k
B
T = 10
−3
¯ hω
0
k
B
T = 10
−2
¯ hω
0
F
t
r
e
l
a
x
[
1
/
ω
0
]
(b)
Figure 6.12: Time scales of
the decay of the coherence mea
sure tr
2
(a) and of the relax
ation towards the asymptotic
solution (b) near the singlet
doublet crossing. Near the ex
act crossing (F ≈ 0.013, full
vertical line) coherence is sta
bilized, whereas at the center
of the avoided crossing (F ≈
0.015, dashed vertical line) the
decay of coherence is acceler
ated. The parameter values are
D = 4, Ω = 0.982 ω
0
, γ =
10
−6
ω
0
, temperature as given
in the legend.
diagonal elements
L
αα,α
α
= 2
¸
n
N
αα
,n
[X
αα
,n
[
2
, α = α
, (6.30)
of the dissipative kernel. They give the rates of the direct transitions from [φ
α
`
to [φ
α
`. Within the full rotatingwave approximation, given in Eqs. 4.28 and 4.29,
these are the only nonvanishing contributions to the master equation which aﬀect
the diagonal elements
αα
of the density matrix.
In the case of zero driving amplitude, the Floquet states [φ
α
` reduce to the eigen
states of the undriven Hamiltonian H
DW
. The only nonvanishing Fourier component
is then [c
α,0
`, and the quasienergies
α
reduce to the corresponding eigenenergies E
α
.
Thus L
αα,α
α
only consists of a single term proportional to N(
α
−
α
). It describes
two kinds of thermal transitions: decay to states with lower energy and, if the en
ergy diﬀerence is less than k
B
T, thermal activation to states with higher energy.
The ratio of the direct transitions forth and back then reads
L
αα,α
α
L
α
α
,αα
= exp
−
(
α
−
α
)
k
B
T
. (6.31)
6.2 Chaotic tunneling near singletdoublet crossings 69
We have detailed balance and therefore the steadystate solution
αα
(∞) ∼ e
−α/k
B
T
δ
αα
. (6.32)
In particular, the occupation probability decays monotonically with the energy of
the eigenstates. In the limit k
B
T →0, the system tends to occupy the ground state
only.
For a strong driving, each Floquet state [φ
α
` contains a large number of Fourier
components and L
αα,α
α
is given by a sum over contributions with quasienergies
α
−
α
+ nΩ. Thus a decay to states with “higher” quasienergy (recall that
quasienergies do not allow for a global ordering) becomes possible due to terms
with n < 0. Physically, they describe dissipative transitions under absorption of
drivingﬁeld quanta. Correspondingly, the system tends to occupy Floquet states
comprising many Fourier components with low index n. According to Eq. (2.42),
these states have low mean energy.
The eﬀects under study are found for a driving with a frequency of the order of
unity. Thus for a quasienergy doublet, i.e., far oﬀ the threelevel crossing, we have
[
α
−
α
[ < Ω, and L
α
α
,αα
is dominated by contributions with n < 0, where the
splitting has no signiﬁcant inﬂuence. However, as a consequence of symmetry, the
splitting is the main diﬀerence between the two partners of the quasienergy doublet.
Therefore, with respect to dissipation, both should behave similarly. In particular,
one expects an equal population of the doublets even in the limit of zero temperature
(Fig. 6.13a). This is in contrast to the undriven case.
In the vicinity of a singletdoublet crossing the situation is more subtle. Here,
the odd partner, say, of the doublet mixes with a chaotic singlet, cf. Eq. (6.22),
and thus acquires components with higher energy. Due to the high mean energy
E
−
c
of the chaotic singlet, close to the top of the barrier, the decay back to the
ground state can also proceed indirectly via other states with mean energy below
E
−
c
. Thus [φ
−
1
` and [φ
−
2
` are depleted and mainly [φ
+
0
` will be populated. However,
if the temperature is signiﬁcantly above the splitting 2b of the avoided crossing,
thermal activation from[φ
+
0
` to [φ
−
1,2
`, accompanied by depletion via the states below
E
−
c
, becomes possible. Thus asymptotically, all these states become populated in
a steady ﬂow (Fig. 6.13b,c). The longtime limit of the corresponding classical
dynamics converges to one of two limit cycles, each of which is located close to one
of the potential minima. In a stroboscopic map they correspond to two isolated
ﬁxed points. This behavior is qualitatively diﬀerent from the asymptotic limit of
the dissipative quantum dynamics near the center of the crossing and shows that
the occupation of the levels outside the singlet and the doublet at asymptotic times
is a pure quantum eﬀect.
An important global characteristic of the asymptotic state is its Shannon entropy
S = −tr(
∞
ln
∞
) or, alternatively, its coherence tr
2
∞
(see Appendix B.2). The
value of the latter gives approximately the reciprocal of the number of incoherently
occupied states. It equals unity only if the attractor is a pure state. According
to the above scenario, we expect tr
2
∞
to assume the value 1/2, in a regime with
strong driving but preserved doublet structure, reﬂecting the incoherent population
70 The harmonically driven doublewell potential
0 1 2 3 4 5 6 7 8 9 10 1112131415
10
3
10
2
10
1
10
0
k
B
T = 10
−4
¯ hω
0
k
B
T = 10
−3
¯ hω
0
k
B
T = 10
−2
¯ hω
0
Floquetstate index α
α
α
φ
+
0
φ
−
1
φ
−
2
(a)
0 1 2 3 4 5 6 7 8 9 10 1112131415
10
3
10
2
10
1
10
0
k
B
T = 10
−4
¯ hω
0
k
B
T = 10
−3
¯ hω
0
k
B
T = 10
−2
¯ hω
0
Floquetstate index α
α
α
φ
+
0
φ
−
1
φ
−
2
(b)
0 1 2 3 4 5 6 7 8 9 10 1112131415
10
3
10
2
10
1
10
0
k
B
T = 10
−4
¯ hω
0
k
B
T = 10
−3
¯ hω
0
k
B
T = 10
−2
¯ hω
0
Floquetstate index α
α
α
φ
+
0
φ
−
2
φ
−
1
(c)
Figure 6.13: Occupation prob
ability
αα
of the Floquet states
φ
α
in the longtime limit. The
parameter values are D = 4, Ω =
0.982 ω
0
, γ = 10
−6
ω
0
, and F =
0.013 (a), 0.0145 (b), 0.015029
(c), temperature as given in the
legend.
6.2 Chaotic tunneling near singletdoublet crossings 71
0.01 0.012 0.014 0.016 0.018 0.02
0.0
0.2
0.4
0.6
0.8
1.0
k
B
T = 0
k
B
T = 10
−4
¯ hω
0
k
B
T = 10
−3
¯ hω
0
k
B
T = 10
−2
¯ hω
0
F
t
r
ρ
2 ∞
(a)
0.01 0.012 0.014 0.016 0.018 0.02
0
1
2
3
k
B
T = 0
k
B
T = 10
−4
¯ hω
0
k
B
T = 10
−3
¯ hω
0
k
B
T = 10
−2
¯ hω
0
F
S
(b)
Figure 6.14: Coherence (a)
and Shannon entropy (b) of the
asymptotic state in the vicinity
of a singletdoublet crossing for
diﬀerent temperatures as given
in the legend. The other pa
rameter values are D = 4, Ω =
0.982 ω
0
, and γ = 10
−6
ω
0
.
10
6
10
5
10
4
10
3
10
2
10
1
0.0
0.5
1.0
(b)
t
r
ρ
2 ∞
0.0
0.5
1.0
(a)
T [¯ hω
0
/k
B
]
t
r
ρ
2 ∞
Figure 6.15: Coherence of the
asymptotic state in the vicin
ity of a singletdoublet cross
ing for F = 0.013 (a) and F =
0.015029 (b): exact calculation
(full line) compared to the val
ues resulting from a threelevel
description (dashed) of the dis
sipative dynamics. The other
parameter values are D = 4,
Ω = 0.982 ω
0
, and γ = 10
−6
ω
0
.
72 The harmonically driven doublewell potential
of the groundstate doublet. In the vicinity of the singletdoublet crossing where
the doublet structure is dissolved, its value should be close to unity for tempera
tures k
B
T < 2b and much less than unity for k
B
T 2b (Figs. 6.14a, 6.15). This
means that the crossing of the chaotic singlet with the regular doublet leads to an
improvement of coherence if the temperature is below the splitting of the avoided
crossing, and a loss of coherence for temperatures above the splitting. This phe
nomenon amounts to a chaosinduced coherence or incoherence, respectively. The
corresponding Shannon entropy (Fig. 6.14b), assumes approximately the value ln n
for n incoherently populated states. Thus outside the crossing, we have S ≈ ln 2
and at the center of the crossing the entropy exhibits a signiﬁcant temperature
dependence.
The crucial role of the decay via states not involved in the threelevel crossing
can be demonstrated by comparing it with the dissipative dynamics including only
these three levels (plus the bath). At the crossing, the threestate model results in
a completely diﬀerent type of asymptotic state (Fig. 6.15). The failure of the three
state model in the presence of dissipation clearly indicates that in the vicinity of the
singletdoublet crossing, it is important to take a large set of levels into account.
6.3 Signatures of chaos in the asymptotic state
In recent work it has been demonstrated that a phasespace representation of quan
tum mechanics, like the Husimi or Wigner distribution, reveals the structures of the
corresponding classical phase space [5,30,117–120]. In particular, for the case of reg
ular classical dynamics, the Husimi function of an eigenstate (or of a Floquet state if
the system is driven) is localized in phase space along the corresponding quantizing
torus; for chaotic motion, it has support in the whole chaotic layer. If the classical
dynamics is mixed, one is even able to classify quantummechanical states as reg
ular or chaotic according to their localization in phase space [120]. Moreover, the
phasespace representation of the asymptotic state of a dissipative quantum map
exhibits the structures of the corresponding classical attractor [106]. However, the
analogies have their limitations due to the Heisenberg uncertainty principle which
does not allow for arbitraryly ﬁne phasespace structures for a quantum system and
results in coarsegraining over a “phasespace unit” 2π.
The asymptotic classical dynamics of the driven dissipative doublewell potential
is for suﬃciently strong driving particularly sensitive to the friction strength: With
decreasing friction, the motion changes from regular to chaotic.
6.3.1 Classical attractor
To describe the classical dissipative dynamics of the driven double well, we add an
Ohmic friction force F
γ
= −γp to the conservative equations (6.4), (6.5) and obtain
˙ x =
1
m
p, (6.33)
6.3 Signatures of chaos in the asymptotic state 73
5 10
3
2 5 10
2
2 5 10
1
2 5 10
0
0.0
0.5
1.0
1.5
2.0
γ/ω
0
d
H
Figure 6.16: Hausdorﬀ di
mension of the classical attrac
tor for F = 0.09, Ω = 0.9 ω
0
.
˙ p = −γp −
∂V (x, t)
∂x
. (6.34)
As friction always decelerates a particle, it distinguishes between future and past,
thus destroys the timereversal symmetry (6.10) of the conservative system. Ac
cordingly, dissipation breaks the reﬂection symmetry at the xaxis of phasespace
portraits which we found for the chosen initial phase of the driving (cf. Fig. 6.6).
The lack of timereversal symmetry in presence of friction is even more evident
from the time evolution of a volume element V of phase space. It evolves by having
each point on its surface ∂V follow an orbit generated by (6.33), (6.34), which yields
by the divergence theorem [3]
dV
dt
=
V
dxdp
∂ ˙ x
∂x
+
∂ ˙ p
∂p
= −γV. (6.35)
Thus, we obtain an exponential contraction of a phasespace volume V —a con
stituent feature of dissipative ﬂows. Therefore, the dynamics is asymptotically
conﬁned to an attractor, a formation in phase space with zero volume to which
all suﬃciently close trajectories from the socalled basin of attraction converge for
long times. For periodically driven dissipative systems, the attractor is in general
also timedependent with the period of the driving and is properly rendered by its
stroboscopic map [121–123].
Depending on the values of the driving parameters and the friction strength,
an attractor consists of limit cycles or isolated ﬁxed points. For suﬃciently weak
dissipation, however, it may even happen that the dissipative dynamics is chaotic
and the attractor possess fractal geometry, forming a socalled strange attractor.
The type of geometry can be characterized as fractal or regular according to its
Hausdorﬀ dimension d
H
which is deﬁned by the scaling assumption
N ∼ l
−d
H
, l →0. (6.36)
Here, N is the number of squares with width l needed to cover the whole attractor.
74 The harmonically driven doublewell potential
1.5 1.0 0.5 0.0 0.5 1.0 1.5
0.5
0.0
0.5
.. . ... ... .. . .. .... ..
... .... ... . ... ... ..... ..... ... . .. . ... .. . ... ... . .... .. .. . ... .. . ... ... . .... ...... ... .. .. .. .. . ... .... .. .. ... .... .... ... . ...
... .. .. .... .. .. .. .. . .. ... .. .... .... .... . .. .. .... ... ... .. ... ..... .. .... .. ... .... .
.. .. .. ... ...... .... .
... .. . .. ...... .. ........ .. ... ... ... ... ..
x/x
0
p
/
m
ω
0
x
0
(a)
1.5 1.0 0.5 0.0 0.5 1.0 1.5
0.5
0.0
0.5
. .. .. ... .. .. .. . .. ...
. .. ... . .. . .. .. .. . . ..
... .. .. .. . .. .. .... . .. . .. .... .. .. .. .. .. ..
. . .. .. .. .. .. ... .. ..... . ... .. ... ... .. . . .... . .. .. .. ... .. .. . .. .
... . . ... . .. . ... .. .. .. . .. .. .. ... .. .. . .. .. . .. .. . . .. .. . ... .. ... .. . . .. ... ... . ... .. . .. ... . . .. ... .. ... .. ... .. .. .. ... . . .. ... .. . ... .. .. .. . ... .. . . ...
x/x
0
p
/
m
ω
0
x
0
(b)
1.5 1.0 0.5 0.0 0.5 1.0 1.5
0.5
0.0
0.5
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
. .
.
. .
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
. .
.
. .
.
. .
.
. .
.
. . .
. . .
. .. . .. . .. . .. . .. . .. ... ... ... .. . .. . .. . .. . .. ... ... ... ... .. .. . ... ... .. .. ... .... . . ... .. . . ... .. .. .. .. .. .. . . . .. . .. ... ... .. ... .. . . ... .. . ... ... .. ... .. . ... ..... . .. . .. . . .... .. .. . .. .. .... ... ..... .. . .. .. . .. . ... .. .. . .. .. .. .. .... .. ... . . .. .. . ... .. .. . ... ... .. .. .. .. .. ..... .. ... . . .. ... .. ..... .. ... ... . .. ... ... . .. .. ... . ... ..... .. .... ... . .. ... .... ... .. ... .. . ... .. . . .... .. . . .. ... . .. . . ... . .... .. .. .. . ... .... ... . . ... . ... ... . ... ... ... .. .. . .... . .. .... . ... . ..... . ... ... ... ... . .. .. .. .. .. ... .... .. .. .. ... . ... .. . ... . .... .. ... . ... .. ... .. .. .... .. . .. . .... .... .. .. .... .. .. . .... .. .. ... ... ... .. .. . ... .. .. .. . .... . . .. ... ... .. .. .... . .. . .. ..... ... ..... .... .. .. .... . .. ... . . .. ... .. ..... . .. . ... . .. ... . .. . ... ... .. ... . ... . .. ... .. . . .... ... ... .. ...... ... .. .. ..... .. .. .. ... .. . .. .. ... ... ..... . . . .... . .... ..... .. .. . . . . ... .. .. . ... ... ... .. .. ... .. . .. ... .. . .. ... .. .. . .. .. .. ... . .. .. . ... . .. .. .. . .. . .... .. .. .... .. . ... .. ... . .. . . . ..... ... .. .. . . ... .. ... . ... .. .. ... .. ... .. ... .... . ... .. ... .. .. .. .. .. .. . ... ... .. .. . .. .... .. .. .. ... .. ... .. ... . ... .. ... .. .. .... . . .. .. . .. . . .. ... .. . .. .. ... ... .. .. ... .... .. ... .. . .. . . .. .. .. .. .... . .. .. .. . .. . .. . ... .... .. ... ..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
. .
.
. . .
. . . .. ... ... ... .. . .. . .. . .. . .. . .. ... ... ... .. . .. . .. ... . . . . ... .. ... .... .. .. .. ... . .. .. ... .. .. . . .. .. . . ... .. ... . .... ... .. . .. . ... ... .... ... . ... . ... .. ... . ... .. . .. .. .. ... ... ...... .. .. .... .. ... .. .. .. ... .. .... . .. ... .. .. ... ... . . .. .. .... . ..... .. ... . ... ... . .. .. ... . .... ... .. . ... .. .. ... ... .. . ... .. .. . . ... . .... .. ... .. . .. .... . .. .... .... ... .. .. .. . .. .. . ..... ... . .. . .. ... . ..... .... .. ... .. . ... ...... . ... .. ... . ..... .. ... .. . . ... .. ... . .. ... .. .... . ... .. ... ... .. .. .. ... . ... ... .... .... .. .. . . .... .. ... . .. . ..... . .. ... .. . .. ... . ... .. .... .. . .. .. . ... .. ... . .. .. ... .. . .. .. .... .. .. .. ... . .... .. ... ... .. .... ... . ... .... . ... . ... ... . .. ... .. ... . .... . . .. ... .. .. . .... . . .. ... .... .. .. ... .. . .. .. .... ... .... ... . .. .. . .... .. . ... . . .. . .... ... . .. ... .. . .. .. ... ... ..... ... .. .. . ... .. .. .. . ... .. . .. . .. . .. . .. ... .. ... . . .. . ... . .. .. . .. . .. . .. .. ..... .. .. . .... .... ... .... ... .... ... .. .. ... .. .. .. .. .. . .. .. .. ... .. .. .. ... ... . .. .. .. ... . .. . .... .. . .. ... .. .. .... . .. .. .. .. ... . .. ... .... . .. ...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
..
.
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
. .
.
. . .
. ... ... ... ... ... .. . .. . .. . .. ... ... . .. .. . .. . . .... .. .. .... . .... . ... ... .. .... . .. ... .. . ... .. ... ... . .. ... . .. .. . ... .. .. .... .. ... . .. .. .. .. .. ... . ... .. .. ... ... .. . ... ... .... . ... .. ... ... .. .. .. . ... ... .. . .... .. .. .. . .. .. ...... . .. . .. ... .... . .. ... .. .... ... .. .. ... .. ... . ... .. .. .. .. . ... . ... .. .. ... .... .. ... . ... .. .. ... . ... .. . .. . ... . . .. .. .. .. ... .. ... . . .. ... .. . ... . .... .. .. .. .. .. . .. .. .. .... . .. .... . .. .... .. .. . .. ... .. .... . ... .... . ... .. ... . .. . . ... .. .... .. .. . .. .. .... .... .. .. ... .... . ... .. ... . ... . .. .. ... . .. . ... . .. .. ... ..... ... . . .. .. . . .. . ..... . ... . .. ...... ... .. ... . ... .. ... ... . .. . .. .. . ..... ..... . . .. ... ..... ... ... .. .. .. ... . ... .. . .. .. .. .. .. . . ... ... ... ... .. . ... . . .... .. .. .. .. ... . ... ... .. . .. .. . ... ... ... .. ... .. .. ... .. . ... ... .. .. .. .. ... ... .. . .. ... .. .. . .. ... .. ... . ... . . .. . .. ... ... .... .. .... ... .. ..... . ... . .... .... . . ... .. .. . .. . .. .. .... . .. .. . ... .. .. ... . .... .. .... .. .. . .. .. ... . .. . ... .. ... . . ... . ..... .. ..... .... ... .. ... ... ..... .. . .. ... . .... ... .. .. ... . .. . .. .. ... ... .. ... . ... . ... .... .. .. .. .. ... .. ... .. .. . .. ..... ... .. . . .. .. . .. ... . .. .. .. .... .. .. ... . .. .. .. ... .. . .. . .. .... . . .. .. .. .. .. .. . ... . .. .. .. .... .. ... .. .. ... .. ... .. .. .. .... .. ... .... . ... ... . ... .. .. ... .. . ... . . ... . .. .. .. .. ... ... . ... .. .. .. .... . .. ... .. ... ... .. .. . ... .. .... .. ... . .... ..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
. .
.
. .
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
. .
.
. .
.
. . . .. . .. . .. ... ... ... ... .. . .. . .. . .. . .. ... ... .. . ... .. . .. . ... . .. .. ... ... .. .. ... . . ... .. .... .. .. .. .. .. ... .. .. .. .. .. ... ... . .. ... .... . .. .. .... ... .. .. ... . ... . .. . ... ... .. ... .. .... ... . .. .. .. .. .. .... .. ... .... .. . ... .. .. . ... .. ... .. .. .... . .. ... . .. ... ... .. .. ... ... .. .. . .. . . .. . .... .. ... ... .. ... .. .. .. .. . ... ... .. ... .... . ... .. .. . ... .... . .. ... .. .. .. .. .. .. ... .. . . . ... ... . .. . .. .... . . .... .. .. .... . .. ... .. ... ... ... ... .. ..... . ... ... .. ... .. . .. .... .. . .. .. .. ... . .... ... .. ... . .. .... . .. . ... .. ... .. .... . .. ...... ... ... . . .. ... .. . .... . .. . .. . ... ... . .. ... . .. . ... .. .. .. .. ... . .. .. ... .. ... ... ... .. . .... . ... .. . .. . ... . . .... . .. .. .. .. .. . .... .. ... .. .. ... . ...... .. . ... . .. .. .. ... .. .. .. .. ..... . . . .. .. .... .. . .. .... . .. ... .... . .... .... .. .. . ... . . ... . .. .. .. . .. .. .... .. . .. .. ... . . ... . .. ... . .. .... . .... .... ... . .. ... .. .. . . .. . ... ..... ... ... ... . ... .. . . . .. .... .. ... .. .... ..... .. . .. . .. .. .. . .. ... .. . . ... ... ... . ... . ... .. .. . .. .. ..... ... .. ... . .. .. . .. .. . . . ... .. . .. .. .. . .. .. .. ... . ... .. ... ... ... .. .. .. ... .. ....... .. .. .. ... .. .. .. .. .. .. ... ... . . .. .. ... .. ... .. . .. .. .. .... .. ... . .. .. ... .. ...... . .. .. ... . ... ... ..... . ... ... .... .. . .... ... .. .. .. ...... .. . .. .. .. .. .. . ... .. .... . .. .... . . . .. .... .. .. ... ... .. .. .. .. .. ... . . ..... . ... . ... . . ... . . ... ... .... .. ... ... .... ... .. ... .... .. .. .. .. .. ... . .. .. .. .. ... . . .. .. . . .. ... .. .. .... ... .. .... . .. . .. .. ... .. .. .. ... .... .... ... . .....
x/x
0
p
/
m
ω
0
x
0
(c)
Figure 6.17: Stroboscopic clas
sical phasespace portrait at t =
2πn/Ω, of the dissipative harmon
ically driven quartic double well,
Eqs. (6.33), (6.34), for the driv
ing amplitude F = 0.09 and fre
quency Ω = 0.9 ω
0
. The fric
tion strength is γ = 0.3 ω
0
(a),
0.2 ω
0
(b), 0.03 ω
0
(c). In panels
(a) and (b) the stroboscopic por
trait is marked by a full dot and
the broken lines show the corres
ponding limit cycles.
6.3 Signatures of chaos in the asymptotic state 75
It is computed numerically by box counting. Consequently, continuous formations
correspond to integer values of d
H
. For d
H
< 2 the attractor has zero volume.
The Hausdorﬀ dimension of the classical attractor for the parameter values
F = 0.09 and Ω = 0.9 ω
0
for diﬀerent friction strength γ is depicted in Fig. 6.16.
Although the attractor of the driven dissipative double well is periodically time
dependent with the period of the driving, its Hausdorﬀ dimension d
H
has no sig
niﬁcant timedependence [121]. Near γ ≈ 0.06 ω
0
, the classical dynamics undergoes
with decreasing γ a transition from regular motion (Fig. 6.17a, 6.17b) to chaos,
manifest by a strange attractor (Fig. 6.17c). For this driving amplitude and fre
quency, the regular islands near the bottom of the wells (cf. Fig. 6.6) are in absence
of dissipation already completely resolved in the chaotic sea.
6.3.2 Quantum attractor
In the quantum case, the selfsimilar ﬁne structures of a strange attractor are in
contradiction to the positionmomentum uncertainty relation, thus they are smeared
out in the Husimi representation of the asymptotic state (Figs. 6.18, 6.19). These
“quantum attractors” clearly reﬂect the structures of the corresponding classical
asymptotic state as well as their qualitative change from isolated ﬁxed points to a
strange attractor. This transition is, however, in the quantum case not as sharp as in
the classical case: Although the asymptotic state for γ = 0.2 ω
0
(Figs. 6.18b, 6.19b)
is still mainly located near the ﬁxed points of the classical stroboscopic map, it covers
a broader phasespace area that already indicates the shape of the strange attractor.
The underlying classical structures in the Husimi functions become more distinct for
smaller values of the eﬀective quantum of action
eﬀ
= 1/8D, as expected. Like the
phasespace portrait of the dissipative classical dynamics (Fig. 6.17), its quantum
mechanical counterparts obey no reﬂection symmetry at the xaxis. This feature is
in contrast to the Husimi representation of the Floquet states in absence of dissi
pation (cf. Fig. 6.7) and is caused by ﬁnite oﬀdiagonal elements of the asymptotic
density matrix in Floquet representation, since diagonal representations share the
symmetries of the basis. Thus, oﬀdiagonal matrix elements play a signiﬁcant role
for the asymptotic state. This demonstrates that a description within a full rotating
wave approximation is insuﬃcient, since it would result in a diagonal asymptotic
state (see Section 4.3.2).
Because the selfsimilar structures at an arbitrary small length scale of the classi
cal attractor are washed out in the quantum case, we cannot characterize the quan
tum attractor by a Hausdorﬀ dimension. A more suitable measure for the qualitative
shape of the quantum attractor is the Wehrl entropy S
Q
of its Husimi representa
tion [120, 124] (see Appendix A.3.2). Its exponential, exp(S
Q
), gives approximately
the number of minimum uncertainty states covered by the Husimi function. Thus,
the occupied phasespace area is 2π exp(S
Q
). The Wehrl entropy of the asymptotic
state for our numerical example for diﬀerent values of the eﬀective quantum of action
is depicted in Fig. 6.20. It becomes larger with decreasing friction γ, reﬂecting the
increasing dispersion of the Husimi functions. In the semiclassical regime, i.e., for
76 The harmonically driven doublewell potential
1.5 1.0 0.5 0.0 0.5 1.0 1.5
0.5
0.0
0.5
.. . ... ... .. . .. .... ..
... .... ... . ... ... ..... ..... ... . .. . ... .. . ... ... . .... .. .. . ... .. . ... ... . .... ...... ... .. .. .. .. . ... .... .. .. ... .... .... ... . ...
... .. .. .... .. .. .. .. . .. ... .. .... .... .... . .. .. .... ... ... .. ... ..... .. .... .. ... .... .
.. .. .. ... ...... .... .
... .. . .. ...... .. ........ .. ... ... ... ... ...
x/x
0
p
/
m
ω
0
x
0
(a)
1.5 1.0 0.5 0.0 0.5 1.0 1.5
0.5
0.0
0.5
. .. .. ... .. .. .. . .. ...
. .. ... . .. . .. .. .. . . ..
... .. .. .. . .. .. .... . .. . .. .... .. .. .. .. .. ..
. . .. .. .. .. .. ... .. ..... . ... .. ... ... .. . . .... . .. .. .. ... .. .. . .. .
... . . ... . .. . ... .. .. .. . .. .. .. ... .. .. . .. .. . .. .. . . .. .. . ... .. ... .. . . .. ... ... . ... .. . .. ... . . .. ... .. ... .. ... .. .. .. ... . . .. ... .. . ... .. .. .. . ... .. . . ....
x/x
0
p
/
m
ω
0
x
0
(b)
1.5 1.0 0.5 0.0 0.5 1.0 1.5
0.5
0.0
0.5
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
. .
.
. .
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
. .
.
. .
.
. .
.
. .
.
. . .
. . .
. .. . .. . .. . .. . .. . .. ... ... ... .. . .. . .. . .. . .. ... ... ... ... .. .. . ... ... .. .. ... .... . . ... .. . . ... .. .. .. .. .. .. . . . .. . .. ... ... .. ... .. . . ... .. . ... ... .. ... .. . ... ..... . .. . .. . . .... .. .. . .. .. .... ... ..... .. . .. .. . .. . ... .. .. . .. .. .. .. .... .. ... . . .. .. . ... .. .. . ... ... .. .. .. .. .. ..... .. ... . . .. ... .. ..... .. ... ... . .. ... ... . .. .. ... . ... ..... .. .... ... . .. ... .... ... .. ... .. . ... .. . . .... .. . . .. ... . .. . . ... . .... .. .. .. . ... .... ... . . ... . ... ... . ... ... ... .. .. . .... . .. .... . ... . ..... . ... ... ... ... . .. .. .. .. .. ... .... .. .. .. ... . ... .. . ... . .... .. ... . ... .. ... .. .. .... .. . .. . .... .... .. .. .... .. .. . .... .. .. ... ... ... .. .. . ... .. .. .. . .... . . .. ... ... .. .. .... . .. . .. ..... ... ..... .... .. .. .... . .. ... . . .. ... .. ..... . .. . ... . .. ... . .. . ... ... .. ... . ... . .. ... .. . . .... ... ... .. ...... ... .. .. ..... .. .. .. ... .. . .. .. ... ... ..... . . . .... . .... ..... .. .. . . . . ... .. .. . ... ... ... .. .. ... .. . .. ... .. . .. ... .. .. . .. .. .. ... . .. .. . ... . .. .. .. . .. . .... .. .. .... .. . ... .. ... . .. . . . ..... ... .. .. . . ... .. ... . ... .. .. ... .. ... .. ... .... . ... .. ... .. .. .. .. .. .. . ... ... .. .. . .. .... .. .. .. ... .. ... .. ... . ... .. ... .. .. .... . . .. .. . .. . . .. ... .. . .. .. ... ... .. .. ... .... .. ... .. . .. . . .. .. .. .. .... . .. .. .. . .. . .. . ... .... .. ... ..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
. .
.
. . .
. . . .. ... ... ... .. . .. . .. . .. . .. . .. ... ... ... .. . .. . .. ... . . . . ... .. ... .... .. .. .. ... . .. .. ... .. .. . . .. .. . . ... .. ... . .... ... .. . .. . ... ... .... ... . ... . ... .. ... . ... .. . .. .. .. ... ... ...... .. .. .... .. ... .. .. .. ... .. .... . .. ... .. .. ... ... . . .. .. .... . ..... .. ... . ... ... . .. .. ... . .... ... .. . ... .. .. ... ... .. . ... .. .. . . ... . .... .. ... .. . .. .... . .. .... .... ... .. .. .. . .. .. . ..... ... . .. . .. ... . ..... .... .. ... .. . ... ...... . ... .. ... . ..... .. ... .. . . ... .. ... . .. ... .. .... . ... .. ... ... .. .. .. ... . ... ... .... .... .. .. . . .... .. ... . .. . ..... . .. ... .. . .. ... . ... .. .... .. . .. .. . ... .. ... . .. .. ... .. . .. .. .... .. .. .. ... . .... .. ... ... .. .... ... . ... .... . ... . ... ... . .. ... .. ... . .... . . .. ... .. .. . .... . . .. ... .... .. .. ... .. . .. .. .... ... .... ... . .. .. . .... .. . ... . . .. . .... ... . .. ... .. . .. .. ... ... ..... ... .. .. . ... .. .. .. . ... .. . .. . .. . .. . .. ... .. ... . . .. . ... . .. .. . .. . .. . .. .. ..... .. .. . .... .... ... .... ... .... ... .. .. ... .. .. .. .. .. . .. .. .. ... .. .. .. ... ... . .. .. .. ... . .. . .... .. . .. ... .. .. .... . .. .. .. .. ... . .. ... .... . .. ...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
..
.
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
. .
.
. . .
. ... ... ... ... ... .. . .. . .. . .. ... ... . .. .. . .. . . .... .. .. .... . .... . ... ... .. .... . .. ... .. . ... .. ... ... . .. ... . .. .. . ... .. .. .... .. ... . .. .. .. .. .. ... . ... .. .. ... ... .. . ... ... .... . ... .. ... ... .. .. .. . ... ... .. . .... .. .. .. . .. .. ...... . .. . .. ... .... . .. ... .. .... ... .. .. ... .. ... . ... .. .. .. .. . ... . ... .. .. ... .... .. ... . ... .. .. ... . ... .. . .. . ... . . .. .. .. .. ... .. ... . . .. ... .. . ... . .... .. .. .. .. .. . .. .. .. .... . .. .... . .. .... .. .. . .. ... .. .... . ... .... . ... .. ... . .. . . ... .. .... .. .. . .. .. .... .... .. .. ... .... . ... .. ... . ... . .. .. ... . .. . ... . .. .. ... ..... ... . . .. .. . . .. . ..... . ... . .. ...... ... .. ... . ... .. ... ... . .. . .. .. . ..... ..... . . .. ... ..... ... ... .. .. .. ... . ... .. . .. .. .. .. .. . . ... ... ... ... .. . ... . . .... .. .. .. .. ... . ... ... .. . .. .. . ... ... ... .. ... .. .. ... .. . ... ... .. .. .. .. ... ... .. . .. ... .. .. . .. ... .. ... . ... . . .. . .. ... ... .... .. .... ... .. ..... . ... . .... .... . . ... .. .. . .. . .. .. .... . .. .. . ... .. .. ... . .... .. .... .. .. . .. .. ... . .. . ... .. ... . . ... . ..... .. ..... .... ... .. ... ... ..... .. . .. ... . .... ... .. .. ... . .. . .. .. ... ... .. ... . ... . ... .... .. .. .. .. ... .. ... .. .. . .. ..... ... .. . . .. .. . .. ... . .. .. .. .... .. .. ... . .. .. .. ... .. . .. . .. .... . . .. .. .. .. .. .. . ... . .. .. .. .... .. ... .. .. ... .. ... .. .. .. .... .. ... .... . ... ... . ... .. .. ... .. . ... . . ... . .. .. .. .. ... ... . ... .. .. .. .... . .. ... .. ... ... .. .. . ... .. .... .. ... . .... ..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
. .
.
. .
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
. .
.
. .
.
. . . .. . .. . .. ... ... ... ... .. . .. . .. . .. . .. ... ... .. . ... .. . .. . ... . .. .. ... ... .. .. ... . . ... .. .... .. .. .. .. .. ... .. .. .. .. .. ... ... . .. ... .... . .. .. .... ... .. .. ... . ... . .. . ... ... .. ... .. .... ... . .. .. .. .. .. .... .. ... .... .. . ... .. .. . ... .. ... .. .. .... . .. ... . .. ... ... .. .. ... ... .. .. . .. . . .. . .... .. ... ... .. ... .. .. .. .. . ... ... .. ... .... . ... .. .. . ... .... . .. ... .. .. .. .. .. .. ... .. . . . ... ... . .. . .. .... . . .... .. .. .... . .. ... .. ... ... ... ... .. ..... . ... ... .. ... .. . .. .... .. . .. .. .. ... . .... ... .. ... . .. .... . .. . ... .. ... .. .... . .. ...... ... ... . . .. ... .. . .... . .. . .. . ... ... . .. ... . .. . ... .. .. .. .. ... . .. .. ... .. ... ... ... .. . .... . ... .. . .. . ... . . .... . .. .. .. .. .. . .... .. ... .. .. ... . ...... .. . ... . .. .. .. ... .. .. .. .. ..... . . . .. .. .... .. . .. .... . .. ... .... . .... .... .. .. . ... . . ... . .. .. .. . .. .. .... .. . .. .. ... . . ... . .. ... . .. .... . .... .... ... . .. ... .. .. . . .. . ... ..... ... ... ... . ... .. . . . .. .... .. ... .. .... ..... .. . .. . .. .. .. . .. ... .. . . ... ... ... . ... . ... .. .. . .. .. ..... ... .. ... . .. .. . .. .. . . . ... .. . .. .. .. . .. .. .. ... . ... .. ... ... ... .. .. .. ... .. ....... .. .. .. ... .. .. .. .. .. .. ... ... . . .. .. ... .. ... .. . .. .. .. .... .. ... . .. .. ... .. ...... . .. .. ... . ... ... ..... . ... ... .... .. . .... ... .. .. .. ...... .. . .. .. .. .. .. . ... .. .... . .. .... . . . .. .... .. .. ... ... .. .. .. .. .. ... . . ..... . ... . ... . . ... . . ... ... .... .. ... ... .... ... .. ... .... .. .. .. .. .. ... . .. .. .. .. ... . . .. .. . . .. ... .. .. .... ... .. .... . .. . .. .. ... .. .. .. ... .... .... ... . .....
x/x
0
p
/
m
ω
0
x
0
(c)
Figure 6.18: Contour plot of
the Husimi function of the quan
tum attractor (full lines) at t =
2πn/Ω, n → ∞, superposed on
the corresponding classical phase
space portrait, Fig. 6.17. The pa
rameter values F = 0.09, Ω =
0.9 ω
0
, γ = 0.3 ω
0
(a), 0.2 ω
0
(b),
0.03 ω
0
(c) are as in Fig. 6.17. The
eﬀective action is D = 6. The
rectangle in the lower left corner
depicts the size of the eﬀective
quantum of action
eﬀ
= 1/8D.
6.3 Signatures of chaos in the asymptotic state 77
1.5 1.0 0.5 0.0 0.5 1.0 1.5
0.5
0.0
0.5
.. . ... ... .. . .. .... ..
... .... ... . ... ... ..... ..... ... . .. . ... .. . ... ... . .... .. .. . ... .. . ... ... . .... ...... ... .. .. .. .. . ... .... .. .. ... .... .... ... . ...
... .. .. .... .. .. .. .. . .. ... .. .... .... .... . .. .. .... ... ... .. ... ..... .. .... .. ... .... .
.. .. .. ... ...... .... .
... .. . .. ...... .. ........ .. ... ... ... ... ...
x/x
0
p
/
m
ω
0
x
0
(a)
1.5 1.0 0.5 0.0 0.5 1.0 1.5
0.5
0.0
0.5
. .. .. ... .. .. .. . .. ...
. .. ... . .. . .. .. .. . . ..
... .. .. .. . .. .. .... . .. . .. .... .. .. .. .. .. ..
. . .. .. .. .. .. ... .. ..... . ... .. ... ... .. . . .... . .. .. .. ... .. .. . .. .
... . . ... . .. . ... .. .. .. . .. .. .. ... .. .. . .. .. . .. .. . . .. .. . ... .. ... .. . . .. ... ... . ... .. . .. ... . . .. ... .. ... .. ... .. .. .. ... . . .. ... .. . ... .. .. .. . ... .. . . ....
x/x
0
p
/
m
ω
0
x
0
(b)
1.5 1.0 0.5 0.0 0.5 1.0 1.5
0.5
0.0
0.5
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
. .
.
. .
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
. .
.
. .
.
. .
.
. .
.
. . .
. . .
. .. . .. . .. . .. . .. . .. ... ... ... .. . .. . .. . .. . .. ... ... ... ... .. .. . ... ... .. .. ... .... . . ... .. . . ... .. .. .. .. .. .. . . . .. . .. ... ... .. ... .. . . ... .. . ... ... .. ... .. . ... ..... . .. . .. . . .... .. .. . .. .. .... ... ..... .. . .. .. . .. . ... .. .. . .. .. .. .. .... .. ... . . .. .. . ... .. .. . ... ... .. .. .. .. .. ..... .. ... . . .. ... .. ..... .. ... ... . .. ... ... . .. .. ... . ... ..... .. .... ... . .. ... .... ... .. ... .. . ... .. . . .... .. . . .. ... . .. . . ... . .... .. .. .. . ... .... ... . . ... . ... ... . ... ... ... .. .. . .... . .. .... . ... . ..... . ... ... ... ... . .. .. .. .. .. ... .... .. .. .. ... . ... .. . ... . .... .. ... . ... .. ... .. .. .... .. . .. . .... .... .. .. .... .. .. . .... .. .. ... ... ... .. .. . ... .. .. .. . .... . . .. ... ... .. .. .... . .. . .. ..... ... ..... .... .. .. .... . .. ... . . .. ... .. ..... . .. . ... . .. ... . .. . ... ... .. ... . ... . .. ... .. . . .... ... ... .. ...... ... .. .. ..... .. .. .. ... .. . .. .. ... ... ..... . . . .... . .... ..... .. .. . . . . ... .. .. . ... ... ... .. .. ... .. . .. ... .. . .. ... .. .. . .. .. .. ... . .. .. . ... . .. .. .. . .. . .... .. .. .... .. . ... .. ... . .. . . . ..... ... .. .. . . ... .. ... . ... .. .. ... .. ... .. ... .... . ... .. ... .. .. .. .. .. .. . ... ... .. .. . .. .... .. .. .. ... .. ... .. ... . ... .. ... .. .. .... . . .. .. . .. . . .. ... .. . .. .. ... ... .. .. ... .... .. ... .. . .. . . .. .. .. .. .... . .. .. .. . .. . .. . ... .... .. ... ..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
. .
.
. . .
. . . .. ... ... ... .. . .. . .. . .. . .. . .. ... ... ... .. . .. . .. ... . . . . ... .. ... .... .. .. .. ... . .. .. ... .. .. . . .. .. . . ... .. ... . .... ... .. . .. . ... ... .... ... . ... . ... .. ... . ... .. . .. .. .. ... ... ...... .. .. .... .. ... .. .. .. ... .. .... . .. ... .. .. ... ... . . .. .. .... . ..... .. ... . ... ... . .. .. ... . .... ... .. . ... .. .. ... ... .. . ... .. .. . . ... . .... .. ... .. . .. .... . .. .... .... ... .. .. .. . .. .. . ..... ... . .. . .. ... . ..... .... .. ... .. . ... ...... . ... .. ... . ..... .. ... .. . . ... .. ... . .. ... .. .... . ... .. ... ... .. .. .. ... . ... ... .... .... .. .. . . .... .. ... . .. . ..... . .. ... .. . .. ... . ... .. .... .. . .. .. . ... .. ... . .. .. ... .. . .. .. .... .. .. .. ... . .... .. ... ... .. .... ... . ... .... . ... . ... ... . .. ... .. ... . .... . . .. ... .. .. . .... . . .. ... .... .. .. ... .. . .. .. .... ... .... ... . .. .. . .... .. . ... . . .. . .... ... . .. ... .. . .. .. ... ... ..... ... .. .. . ... .. .. .. . ... .. . .. . .. . .. . .. ... .. ... . . .. . ... . .. .. . .. . .. . .. .. ..... .. .. . .... .... ... .... ... .... ... .. .. ... .. .. .. .. .. . .. .. .. ... .. .. .. ... ... . .. .. .. ... . .. . .... .. . .. ... .. .. .... . .. .. .. .. ... . .. ... .... . .. ...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
..
.
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
. .
.
. . .
. ... ... ... ... ... .. . .. . .. . .. ... ... . .. .. . .. . . .... .. .. .... . .... . ... ... .. .... . .. ... .. . ... .. ... ... . .. ... . .. .. . ... .. .. .... .. ... . .. .. .. .. .. ... . ... .. .. ... ... .. . ... ... .... . ... .. ... ... .. .. .. . ... ... .. . .... .. .. .. . .. .. ...... . .. . .. ... .... . .. ... .. .... ... .. .. ... .. ... . ... .. .. .. .. . ... . ... .. .. ... .... .. ... . ... .. .. ... . ... .. . .. . ... . . .. .. .. .. ... .. ... . . .. ... .. . ... . .... .. .. .. .. .. . .. .. .. .... . .. .... . .. .... .. .. . .. ... .. .... . ... .... . ... .. ... . .. . . ... .. .... .. .. . .. .. .... .... .. .. ... .... . ... .. ... . ... . .. .. ... . .. . ... . .. .. ... ..... ... . . .. .. . . .. . ..... . ... . .. ...... ... .. ... . ... .. ... ... . .. . .. .. . ..... ..... . . .. ... ..... ... ... .. .. .. ... . ... .. . .. .. .. .. .. . . ... ... ... ... .. . ... . . .... .. .. .. .. ... . ... ... .. . .. .. . ... ... ... .. ... .. .. ... .. . ... ... .. .. .. .. ... ... .. . .. ... .. .. . .. ... .. ... . ... . . .. . .. ... ... .... .. .... ... .. ..... . ... . .... .... . . ... .. .. . .. . .. .. .... . .. .. . ... .. .. ... . .... .. .... .. .. . .. .. ... . .. . ... .. ... . . ... . ..... .. ..... .... ... .. ... ... ..... .. . .. ... . .... ... .. .. ... . .. . .. .. ... ... .. ... . ... . ... .... .. .. .. .. ... .. ... .. .. . .. ..... ... .. . . .. .. . .. ... . .. .. .. .... .. .. ... . .. .. .. ... .. . .. . .. .... . . .. .. .. .. .. .. . ... . .. .. .. .... .. ... .. .. ... .. ... .. .. .. .... .. ... .... . ... ... . ... .. .. ... .. . ... . . ... . .. .. .. .. ... ... . ... .. .. .. .... . .. ... .. ... ... .. .. . ... .. .... .. ... . .... ..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
. .
.
. .
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. .
.
. .
.
. .
.
. . . .. . .. . .. ... ... ... ... .. . .. . .. . .. . .. ... ... .. . ... .. . .. . ... . .. .. ... ... .. .. ... . . ... .. .... .. .. .. .. .. ... .. .. .. .. .. ... ... . .. ... .... . .. .. .... ... .. .. ... . ... . .. . ... ... .. ... .. .... ... . .. .. .. .. .. .... .. ... .... .. . ... .. .. . ... .. ... .. .. .... . .. ... . .. ... ... .. .. ... ... .. .. . .. . . .. . .... .. ... ... .. ... .. .. .. .. . ... ... .. ... .... . ... .. .. . ... .... . .. ... .. .. .. .. .. .. ... .. . . . ... ... . .. . .. .... . . .... .. .. .... . .. ... .. ... ... ... ... .. ..... . ... ... .. ... .. . .. .... .. . .. .. .. ... . .... ... .. ... . .. .... . .. . ... .. ... .. .... . .. ...... ... ... . . .. ... .. . .... . .. . .. . ... ... . .. ... . .. . ... .. .. .. .. ... . .. .. ... .. ... ... ... .. . .... . ... .. . .. . ... . . .... . .. .. .. .. .. . .... .. ... .. .. ... . ...... .. . ... . .. .. .. ... .. .. .. .. ..... . . . .. .. .... .. . .. .... . .. ... .... . .... .... .. .. . ... . . ... . .. .. .. . .. .. .... .. . .. .. ... . . ... . .. ... . .. .... . .... .... ... . .. ... .. .. . . .. . ... ..... ... ... ... . ... .. . . . .. .... .. ... .. .... ..... .. . .. . .. .. .. . .. ... .. . . ... ... ... . ... . ... .. .. . .. .. ..... ... .. ... . .. .. . .. .. . . . ... .. . .. .. .. . .. .. .. ... . ... .. ... ... ... .. .. .. ... .. ....... .. .. .. ... .. .. .. .. .. .. ... ... . . .. .. ... .. ... .. . .. .. .. .... .. ... . .. .. ... .. ...... . .. .. ... . ... ... ..... . ... ... .... .. . .... ... .. .. .. ...... .. . .. .. .. .. .. . ... .. .... . .. .... . . . .. .... .. .. ... ... .. .. .. .. .. ... . . ..... . ... . ... . . ... . . ... ... .... .. ... ... .... ... .. ... .... .. .. .. .. .. ... . .. .. .. .. ... . . .. .. . . .. ... .. .. .... ... .. .... . .. . .. .. ... .. .. .. ... .... .... ... . .....
x/x
0
p
/
m
ω
0
x
0
(c)
Figure 6.19: Same as Fig. 6.18
for the eﬀective action D = 12.
78 The harmonically driven doublewell potential
10
3
2 5 10
2
2 5 10
1
2 5 10
0
2
3
4
D = 12
D = 6
D = 3
γ/ω
0
S
Q
Figure 6.20: Wehrl entropy of
the asymptotic state of the dissi
pative quantum map for diﬀerent
values of the eﬀective quantum of
action
eﬀ
= 1/8D. Other pa
rameters like in Fig. 6.16.
a suﬃciently large value of the eﬀective action D, we observe a kink of the entropy
near γ ≈ 0.06 ω
0
, where the classical attractor undergoes a transition from a set of
isolated ﬁxed points to a strange attractor.
Note that for γ
>
∼
0.1 ω
0
, the Markov approximation becomes inaccurate, since
γ is of the order of the mean level spacing and the condition (3.33) is violated for
at least some of the transitions between Floquet states. Nevertheless, we obtain the
qualitative behavior which we expected from classical considerations.
7 Summary and outlook
In this thesis, we put focus on a special class of system: a particle which moves in a
onedimensional potential under the inﬂuence of a heat bath and of an external ﬁeld
which is periodic in time. A Markovian approach to quantum dissipation, based
on the Floquet solutions of the coherent dynamics, has proven welladapted to the
description of such systems. We have derived this FloquetMarkov approach from
an exact pathintegral expression and have applied it to the parametrically driven
harmonic oscillator and the driven doublewell potential.
The study of the parametrically driven harmonic oscillator has been devoted
mainly to a thorough understanding of the diﬀerent approximation schemes. It
turned out that the dissipative part of the Markovian master equation depends
quantitatively on whether the driving is included in its derivation or not: Consider
ing the driving mainly results in a modiﬁed momentum diﬀusion that depends on
the quasienergy spectrum instead of the unperturbed spectrum of the central system
without the driving. The diﬀerence becomes signiﬁcant in the limits of strong driv
ing and low temperature. An additional additive timedependent force undergoes a
renormalization which, however, vanishes for strictly Ohmic damping. Concluding
from numerical results for the case of a Mathieu oscillator, the attributes “simple”
and “improved” for the two basic Markovian approaches prove adequate. To solve
the master equation, we have transformed it to Wigner representation, thus obtained
a partial diﬀerential equation for the Wigner function that corresponds to the den
sity operator, and derived an analytical expression for the Floquet solutions of the
resulting FokkerPlancklike equation. In doing so, we have incidentally obtained
the Floquet solutions of the FokkerPlanck equation for the corresponding classical
Brownian motion.
A quantum system with more complex dynamics is the quartic doublewell po
tential under the inﬂuence of a driving with frequency near resonance. Here, classical
chaos plays a signiﬁcant role for the coherent dynamics. Even for arbitrarily small
driving amplitude, the separatrix is replaced by a chaotic layer, but the motion near
the bottom of the wells remains regular. Nevertheless, the inﬂuence of states located
in the chaotic region alters the splittings of the regular doublets and thus the tunnel
rates, which is the essence of chaotic tunneling. We have studied chaotic tunneling
in the vicinity of crossings of chaotic singlets with tunnel doublets under the inﬂu
ence of an environment. As a simple intuitive model to compare against, we have
constructed a threestate system which in the case of vanishing dissipation, provides
a faithful description of an isolated singletdoublet crossing. Dissipation introduces
new time scales to the system: one for the loss of coherence and a second one for
the relaxation to an asymptotic state. Well outside the crossing, both timescales
are of the same order, reﬂecting an eﬀective twostate behavior. The center of the
80 Summary and outlook
crossing is characterized by a strong mixing of the chaotic state with one state of
the tunnel doublet. The high mean energy of the chaotic state introduces additional
decay channels to states outside the threestate system. Thus, decoherence becomes
far more eﬀective and, accordingly, tunneling fades out much faster.
The study of the asymptotic state, the quantum attractor, demonstrates clearly
that a threestate model of the singletdoublet crossing is insuﬃcient once dissipation
is eﬀective. This is so because the coupling to the heat bath enables processes of
decay and thermal activation that connect the states in the crossing with other,
“external” states of the central system. In the presence of driving, the asymptotic
state is no longer literally a state of equilibrium. Rather, incoherent processes create
a steady ﬂow of probability involving states within as well as outside the crossing.
As a result, the composition of the asymptotic state, expressed for example by its
coherence tr
2
∞
, are markedly diﬀerent at the center of the crossing as compared to
the asymptotic state far away from the crossing, even if that is barely visible in the
corresponding classical phasespace structure.
With increasing driving amplitude, the dynamics near the bottom of the wells,
in absence of dissipation, becomes fully chaotic. This has striking consequences
for the dissipative classical dynamics: For suﬃciently small dissipation, it remains
chaotic, but for strong friction it becomes regular. Accordingly, the geometry of the
classical attractor is fractal or regular, respectively. We have observed the signatures
of this qualitative diﬀerence in the asymptotic state of the corresponding quantum
dynamics. However, in contrast to the sudden change of the classical behavior, the
quantum attractor undergoes a smooth transition: The structure of the strange
attractor is already felt by the Husimi function for parameter values where the
classical attractor consists only of two isolated ﬁxed points. For the observation of
these semiclassical structures, oﬀdiagonal matrix elements of the asymptotic state
in Floquet basis proved crucial. This clearly reﬂects the failure a full rotatingwave
approximation.
Many more phenomena at the overlap of chaos, tunneling, and dissipation await
being unraveled. They include fourstate crossings formed when two doublets inter
sect, chaotic Bloch tunneling along extended potentials with a large number of unit
cells instead of just two, and the inﬂuence of decoherence on a multistep mechanism
of chaotic tunneling. These phenomena are typically observed in the far semiclas
sical regime, which requires to take very many levels into account. A semiclassical
description of the dissipative quantum system may circumvent this problem.
A The harmonic oscillator
In many ﬁelds of physics, the harmonic oscillator plays an important role as an ex
actly solvable model as well as an approximation to a smooth potential minimum.
In this work, we use its eigenfunctions as a basis set for numerical computations.
Moreover, the ground state of a harmonic oscillator, displaced in phase space (co
herent state), forms the initial state for the propagation of the density matrix in
Chapter 6. In this appendix, we give a synopsis of basic properties of the harmonic
oscillator, described by the Hamiltonian
H
HO
=
1
2m
p
2
+
mω
2
HO
2
x
2
(A.1)
= ω
HO
a
+
a +
1
2
, (A.2)
and of the closely related coherent states and quasiprobabilities.
The form (A.2) of the Hamiltonian is achieved by the transformation
a =
mω
HO
2
x + i
1
2mω
HO
p, (A.3)
a
+
=
mω
HO
2
x −i
1
2mω
HO
p, (A.4)
x =
2mω
HO
(a
+
+a), (A.5)
p = i
mω
HO
2
(a
+
−a). (A.6)
From [x, p] = i results the bosonic commutation relation
[a, a
+
] = 1, (A.7)
which yields for the energy eigenstates [n` the relations [125]
a[n` =
√
n[n −1`, (A.8)
a
+
[n` =
√
n + 1 [n + 1`. (A.9)
These justify the denotation creation and destruction operator (of a quantum) or
shift operators (between eigenstates) for a
+
and a. By recursion of (A.9), the so
called number states
[n` =
(a
+
)
n
√
n!
[0` (A.10)
82 The harmonic oscillator
are constructed from the ground state [0`, which is deﬁned by a[0` = 0.
The state [n` in a semiclassical interpretation [126,127] is a quantized torus with
action (n + 1/2). Therefore, it is restricted to phasespace areas which obey
p
2
2m
+
1
2
mω
2
HO
x
2
<
∼
nω
HO
, (A.11)
thus
[p[
<
∼
p
n
=
2nω
HO
m, (A.12)
[x[
<
∼
x
n
=
2n
mω
HO
. (A.13)
A.1 Number states as a basis set
For numerical computations, wave functions and operators are decomposed into a
complete set of basis functions. Dealing with polynomial potentials, the eigenfunc
tions of the harmonic oscillator form a wellsuited basis set, as matrix elements of
powers of the position operator for these states obey a simple analytical expression
resulting from (A.5)–(A.9).
In numerical calculations, one uses N number states (A.10) as a (incomplete)
basis set, thus formally approximates inﬁnite matrices by ﬁnite ones. Thus, we
eﬀectively diagonalize—instead of the Hamiltonian H—the truncated Hamiltonian
{
N
H{
N
, where {
N
projects on the subspace spanned by the ﬁrst N basis functions
¦[n`¦
n=0...N
. This subspace, according to (A.12), (A.13), corresponds to a ﬁnite
region of phase space. Consequently, a state with energy E can be approximated
reasonably by a linear combination of the ﬁrst N number states only if its corre
sponding classical torus is contained in this region of phase space. This results in
the conditions
E <
p
2
N
2m
= Nω
HO
, (A.14)
E < V (x
N
) = −
Nω
2
0
2ω
HO
+
N
2
2
ω
4
0
16E
B
ω
2
HO
. (A.15)
To visualize the inﬂuence of a ﬁnite basis set, we have depicted some eigenvalues
of the truncated Hamiltonian {
N
H
DW
{
N
for N = 100 over the scaling parameter
ω
HO
/ω
0
of the basis functions in Fig. A.1. Outside the limits (A.14) and (A.15), the
energies depend on the scaling parameter, thus their value is a numerical artefact
caused by using a ﬁnite basis set. The numerical computations in Chapter 6 were
performed using number states with an oscillator frequency
ω
HO
= ω
0
Nω
0
16E
B
1/3
, (A.16)
and N was chosen according to the required numerical precision.
A.2 Coherent states 83
10
1
2 5 10
0
2 5
0
20
40
60
80
100
ω
HO
/ω
0
E
/
¯h
ω
0
Figure A.1: Some eigenval
ues of the truncated Hamil
tonian P
N
H
DW
P
N
for N =
100 and D = 4 (full lines).
The broken lines give the lim
its of convergence according
to (A.14) and (A.15).
A.2 Coherent states
Due to the Heisenberg uncertainty principle
∆x∆p ≥
2
(A.17)
a quantummechanical state cannot be localized in phase space with arbitrary pre
cision, as would be possible in classical mechanics. The coherent states (or Glauber
states) [128, 129]
[z` = e
za
+
−z
∗
a
[0`, z ∈ C (A.18)
obey
'z[x[z` =
2
mω
HO
Re z, 'z[∆x
2
[z` =
2mω
HO
, (A.19)
'z[p[z` =
2mω
HO
Imz, 'z[∆p
2
[z` =
mω
HO
2
. (A.20)
Thus according to (A.17), they have minimal uncertainty and approximate a point
in phase space at best.
A.3 Quasiprobabilities
The unique representation of a density operator as a phasespace function is closely
related to the question on quasiclassical states. The most prominent example from
a variety of possibilities [130–134] is the sparameterized quasiprobability or Cahill
Glauber distribution [135]
W
s
(x, p) =
1
2π
2
dξ
dξ
e
zξ
∗
−z
∗
ξ
χ
s
(ξ), (A.21)
84 The harmonic oscillator
χ
s
(ξ) = tr
e
ξa
+
−ξ
∗
a+sξ
∗
ξ/2
¸
, s ∈ [−1, 1], (A.22)
z = x
mω
HO
2
+ ip
1
2mω
HO
. (A.23)
It includes the Wigner and the Husimi function as limiting cases. The integration
in (A.21) runs over real and imaginary part of ξ. In general, W
s
may also assume
negative values and for positive s may even be singular—thus a strict probabilistic
interpretation is not possible. Quasiprobabilities are used for the calculation of
expectation values alike classical phasespace distributions. Thereby the operator
ordering is ﬁxed by the parameter s as the sordered product
¸
(a
+
)
n
a
m
¸
s
=
∂
∂z
n
−
∂
∂z
∗
m
exp
za
+
−z
∗
a +
s
2
z
∗
z
z
∗
=z=0
, (A.24)
which gives an interpolation between normal ordering (a
+
)
n
a
m
= ¦(a
+
)
n
a
m
¦
1
and
antinormal ordering a
m
(a
+
)
n
= ¦(a
+
)
n
a
m
¦
−1
of creation and annihilation operators
[136].
For each operator acting on the density matrix , there exists a corresponding
diﬀerential operator acting on W
s
(x, p) [137]. From Eq. (A.21) with Eq. (A.22) we
obtain the relations
x ←→
x +
i
2
∂
p
−
s
2mω
HO
∂
x
W
s
(x, p), (A.25)
p ←→
p −
i
2
∂
x
−
smω
HO
2
∂
p
W
s
(x, p), (A.26)
x ←→
x −
i
2
∂
p
−
s
2mω
HO
∂
x
W
s
(x, p), (A.27)
p ←→
p +
i
2
∂
x
−
smω
HO
2
∂
p
W
s
(x, p). (A.28)
For powers of x and p they hold iteratively. It is obvious from these operator cor
respondences that, except for the case s = 0, the sparameterized quasiprobability
depends on the choice of the oscillator frequency ω
HO
.
A.3.1 Wigner function
For s = 0, W
s
results in the Wigner function [130, 133]
W(x, p) =
1
2π
dx
e
ipx
/
'x +x
/2[[x −x
/2` = W
0
(x, p). (A.29)
It is independent of the oscillator frequency ω
HO
, thus basis independent. In nu
merical computations of Wigner functions or their reconstruction from experimental
data, a negative s with small absolute value is often used to ensure numerical con
vergence.
A.3 Quasiprobabilities 85
A.3.2 Husimi function and Wehrl entropy
The Husimi function is deﬁned as the expectation value of the density operator with
coherent states [131] and coincides with the quasiprobability W
−1
,
Q(x, p) =
1
2π
'z[[z` = W
−1
(x, p), (A.30)
where z(x, p) is given by (A.23). It is nonnegative, due to the positivity of the
density operator [133]. The fact that already the diagonal matrix elements hold the
full information on the quantum state reﬂects the overcompleteness of the coherent
states [128].
In a semiclassical limit, the Husimi function of a state is localized in phase space
along the corresponding Lagrangian manifolds. Thus, in case of regular classical
dynamics, the Husimi function of an eigenstate is located on the corresponding
quantizing torus; for the case of irregular classical dynamics, it is smeared out over
the whole chaotic layer [5]. This allows for a classiﬁcation of single eigenstates
as regular or chaotic if the classical dynamics is mixed. For driven systems, the
respective assignment of Floquet states to regions in classical phase space holds
true [120].
For a classiﬁcation of quantum mechanical states according to their phasespace
structure, it is desirable to have a direct measure for localization properties. One
possibility is provided by the Wehrl entropy S
Q
of the state which is deﬁned as the
entropy of the corresponding Husimi function [120, 124],
S
Q
= −
dxdp Q(x, p) ln[2πQ(x, p)]. (A.31)
The number of minimum uncertainty states occupied by the Husimi function is
approximately given by exp(S
Q
), thus the occupied phasespace area is 2π exp(S
Q
).
Consequently, for a coherent state the Wehrl entropy assumes its minimum value
S
Q
min
= 1.
86
B The density operator
An observer, who is not fully aware of the state of a system, can at best describe it
by a density operator [138]. Its eigenvalues p
i
give the probability for the system
to reside in the corresponding eigenstate. Therefore, the eigenvalues of a proper
density operator have to suﬃce the intrinsic restrictions of probabilities,
0 ≤ p
i
≤ 1, (B.1)
¸
i
p
i
= tr = 1, (B.2)
i.e. positivity and a total probability which equals unity. In the limit of a pure state,
where the full quantummechanical information (i.e. the wavefunction) is known, one
of the probabilities p
i
equals unity, all the others vanish.
B.1 Lindblad form
The conditions on a physically meaningful density operator, Eqs. (B.1) and (B.2),
as well as its Hermitecity, of course, have to be conserved during time evolution.
Lindblad proved [93] that a Markovian master equation with constant coeﬃcients
meets this requirement, thus generates a socalled completely positive dynamical
semigroup, if and only if it is of the form
˙ = −
i
[H, ] +
¸
i
γ
i
2Q
i
Q
†
i
−Q
†
i
Q
i
−Q
†
i
Q
i
. (B.3)
The operators Q
i
, which are introduced phenomenologically, induce dissipative tran
sitions of the system.
It turned out, however, that many Markovian master equations occurring in the
literature [11, 70, 71, 139, 140], including our master equation (4.3), are not of this
socalled Lindblad form, thus they do not ensure positivity of an arbitrary density
operator at any future time. This apparent contradiction was resolved only recently:
A master equation of the form (4.3) violates positivity only for initial conditions
that do not meet the requirements under which it has been derived. Namely, if the
system is prepared with a position variance ∆x smaller than the thermal de Broglie
wavelength,
∆x < λ
dB
= /
4mk
B
T, (B.4)
positivity will be violated until ∆x becomes larger than λ
dB
[68–71, 141]. Thus,
dissipative eﬀects on a length scale l < λ
dB
cannot be described selfconsistently
within a Markov approximation.
88 The density operator
B.2 Coherence and entropy
The lack of information inherent in a density operator can be measured by the
Shannon entropy
S = −
¸
i
p
i
ln p
i
= −tr( ln ). (B.5)
Consequently, for a pure state S = 0. This deﬁnition agrees, besides a factor k
B
,
with the entropy known from statistical thermodynamics. The entropy also gives a
proper measure for the coherence of a system, thus for the ability to observe interfer
ence eﬀects. However, it has the disadvantage that its direct numerical computation
requires diagonalization of the density operator. A numerically less expensive, re
lated quantity is the “linearized entropy”
S
lin
= tr (1 −) = 1 −tr
2
, (B.6)
introduced by Zurek et al. [142]. It arises formally by Taylor expansion of (B.5) if
describes an almost pure state. In the case of many incoherently populated states,
all p
i
<1 and both entropies diﬀer drastically. Nevertheless, the related quantity
C = tr
2
= 1 −S
lin
(B.7)
is a proper measure for the coherence of a density operator. Its value approximately
gives the reciprocal of the number of incoherently populated states and equals unity
if the system resides in a pure state.
C
Solution of the
FokkerPlanck equation
In this appendix, we solve the equation of motion (5.55) for the Wigner function by
the method of characteristics. We write W(x, p, t) as
W(x, p, t) =
dXdP e
ixX+ipP
e
S(X,P,t)
. (C.1)
By this ansatz, equation (5.55) is transformed to the quasilinear partial diﬀerential
equation
T(X, S
X
, P, S
P
, t, S
t
) = 0 (C.2)
for S(X, P, t), where T is given by
T = S
t
−XS
P
+γPS
P
+ω
2
(t)PS
X
+γD
pp
P
2
+γD
xp
XP. (C.3)
We denote the partial derivatives of S(X, P, t) with respect to X, P, and t by S
X
,
S
P
, and S
t
, respectively.
The characteristic equations [95] of (C.2) are given by
˙
t =
∂T
∂S
t
= 1, (C.4)
˙
X =
∂T
∂S
X
= ω
2
(t)P, (C.5)
˙
P =
∂T
∂S
P
= γP −X, (C.6)
˙
S
X
= −
∂T
∂X
= S
P
−γD
xp
P, (C.7)
˙
S
P
= −
∂T
∂P
= −γS
P
−ω
2
(t)S
X
−2γD
pp
P −γD
xp
X, (C.8)
˙
S
t
= −
∂T
∂t
= −
dω
2
(t)
dt
PS
X
, (C.9)
whose solutions give the characteristics of the partial diﬀerential equation (C.2).
Equation (C.4) signiﬁes that the characteristics can be parameterized by the
time t. Instead of equation (C.9), we will use (C.2) to get an expression for S
t
. So
we only have to solve (C.5)–(C.8). The solutions of these equations can be traced
back to the fundamental solutions f
i
(t) of the classical equation of motion (5.3).
From (C.5) and (C.6), we ﬁnd
¨
P −γ
˙
P +ω
2
(t)P = 0. (C.10)
90 Solution of the FokkerPlanck equation
This is simply the classical equation of motion with a negative damping constant.
Therefore the solutions for X and P read
P(t) = −c
1+
e
γt
f
2
(t) +c
2+
e
γt
f
1
(t), (C.11)
X(t) = c
1+
e
γt
˙
f
2
(t) −c
2+
e
γt
˙
f
1
(t), (C.12)
where c
i+
denote integration constants.
From (C.7) and (C.8) we ﬁnd for S
X
¨
S
X
+γ
˙
S
X
+ω
2
(t)S
X
= −2γDP, (C.13)
which is the classical equation of motion with an inhomogeneity. The eﬀective
diﬀusion constant D is given by
D = D
pp
+γD
xp
. (C.14)
With the integration constants c
i−
, we integrate (C.13) with the Green function
(5.10) to
S
X
(t) = c
1−
f
1
(t) +c
2−
f
2
(t) −2γD
t
t
0
dt
G(t, t
)P(t
), (C.15)
and get by use of (C.7)
S
P
(t) = c
1−
˙
f
1
(t) +c
2−
˙
f
2
(t) −2γD
t
t
0
dt
∂G(t, t
)
∂t
P(t
) +γD
xp
P(t). (C.16)
By inserting
P(t
) = G(t, t
)X(t) +
∂G(t, t
)
∂t
P(t), (C.17)
obtained from Eqs. (C.11) and (C.12), we get a result for S
X
and S
P
that only
depends on the endpoints of the characteristics. Now together with Eq. (C.2), we
have an expression for grad S(X, P, t) = (S
X
, S
P
, S
t
), which can be integrated to
S(X, P, t) =
c
1−
f
1
(t) +c
2−
f
2
(t)
X +
c
1−
˙
f
1
(t) +c
2−
˙
f
2
(t)
P
−
1
2
σ
xx
(t, t
0
)X
2
−σ
xp
(t, t
0
)XP −
1
2
σ
pp
(t, t
0
)P
2
, (C.18)
with
σ
xx
(t, t
0
) = 2γD
t
t
0
dt
[G(t, t
)]
2
, (C.19)
σ
xp
(t, t
0
) = 2γD
t
t
0
dt
G(t, t
)
∂
∂t
G(t, t
), (C.20)
σ
pp
(t, t
0
) = −γD
xp
+ 2γD
t
t
0
dt
¸
∂
∂t
G(t, t
)
2
. (C.21)
By inserting S(X, P, t) into (C.1), we ﬁnd a timedependent solution for the Wigner
function W(x, p, t).
Solution of the FokkerPlanck equation 91
The integration constants c
i±
are constant along the characteristics by construc
tion. Thus, the Poisson brackets between the expressions c
i±
(X, S
X
, P, S
P
, t) and
T(X, S
X
, P, S
P
, t, S
t
) vanish [95]. By transforming back from Fourier space to real
space, one ﬁnds that the operators ˆ c
i±
≡ c
i±
(−i∂
x
, −ix, −i∂
p
, −ip, t) commute with
the operator ∂
t
− L(t), whose nullspace is the solution of the equation of motion.
Therefore, the ˆ c
i±
are shift operators in the subspace of solutions, i.e., if W(x, p, t)
is a solution of (5.55), then ˆ c
i±
W(x, p, t) is also a solution. For the ˆ c
i±
we ﬁnd
ˆ c
1+
=
1
2
f
1
(t)∂
x
+
˙
f
1
(t)∂
p
, (C.22)
ˆ c
2+
=
1
2
f
2
(t)∂
x
+
˙
f
2
(t)∂
p
, (C.23)
ˆ c
1−
= i
˙
f
2
(t)
x +σ
xx
(t, t
0
)∂
x
+σ
xp
(t, t
0
)∂
p
−if
2
(t)
p +σ
xp
(t, t
0
)∂
x
+σ
pp
(t, t
0
)∂
p
, (C.24)
ˆ c
2−
= −i
˙
f
1
(t)
x +σ
xx
(t, t
0
)∂
x
+σ
xp
(t, t
0
)∂
p
+ if
1
(t)
p +σ
xp
(t, t
0
)∂
x
+σ
pp
(t, t
0
)∂
p
. (C.25)
Note that because of the linear structure of the characteristic equations, there is
no ambiguity concerning the ordering of operators. The operators Q
i+
(t), used in
Section 5.4.2 to construct the Floquet solutions of the FokkerPlanck equation, are
proportional to the ˆ c
i+
.
92
References
[1] G. Casati, B. V. Chirikov, F. M. Izrailev, and J. Ford, in Stochastic Behavior
in Classical and Quantum Hamiltonian Systems, Vol. 93 of Lecture Notes in
Physics, edited by G. Casati and J. Ford (Springer, Berlin, 1979), p. 334.
[2] T. Dittrich and R. Graham, Long Time Behavior in the Quantized Standard
Map with Dissipation, Ann. Phys. (N.Y.) 200, 363 (1990).
[3] H.G. Schuster, Deterministic chaos: an introduction, 2nd ed. (VCH, Wein
heim, 1989).
[4] E. Heller, BoundState Eigenfunctions of Classically Chaotic Hamiltonian Sys
tems: Scars of Periodic Orbits, Phys. Rev. Lett. 53, 1515 (1984).
[5] K. Takahashi and N. Saitˆ o, Chaos and Husimi Distribution Function in Quan
tum Mechanics, Phys. Rev. Lett. 55, 645 (1985).
[6] F. J. Dyson, The Threefold Way: Algebraic Structure of Symmetry Groups
and Ensembles in Quantum Mechanics, J. Math. Phys. 3, 1199 (1962).
[7] M. V. Berry and M. Robnik, Semiclassical level spacings when regular and
chaotic orbits coexist, J. Phys. A 17, 2413 (1984).
[8] M. L. Mehta, Random matrices and the statistical theory of energy levels (Aca
demic Press, New York, 1967).
[9] F. Hund, Zur Deutung der Molekelspektren III: Bemerkungen ¨ uber das
Schwingungs und Rotationsspektrum bei Molekeln mit mehr als zwei Kernen,
Z. Phys. 43, 805 (1927).
[10] F. Grossmann, T. Dittrich, P. Jung, and P. H¨ anggi, Coherent Destruction of
Tunneling, Phys. Rev. Lett. 67, 516 (1991).
[11] A. O. Caldeira and A. L. Leggett, Quantum Tunnelling in a Dissipative Sys
tem, Ann. Phys. (N.Y.) 149, 374 (1983).
[12] M. Grifoni and P. H¨ anggi, Driven Quantum Tunneling, Phys. Rep. 304, 219
(1998).
[13] O. Bohigas, S. Tomsovic, and D. Ullmo, Dynamical quasidegeneracies and
separation of regular and irregular quantum levels, Phys. Rev. Lett. 64, 1479
(1990).
[14] O. Bohigas, S. Tomsovic, and D. Ullmo, Classical transport eﬀects on chaotic
levels, Phys. Rev. Lett. 65, 5 (1990).
[15] S. Tomsovic and D. Ullmo, Chaosassisted tunneling, Phys. Rev. E 50, 145
(1994).
[16] R. Utermann, T. Dittrich, and P. H¨ anggi, Tunneling and the Onset of Chaos
in a Driven Bistable System, Phys. Rev. E 49, 273 (1994).
[17] W. A. Lin and L. E. Ballentine, Quantum tunneling and chaos in a driven
anharmonic oscillator, Phys. Rev. Lett. 65, 2927 (1990).
94 References
[18] J. Plata and J. M. Gomez Llorente, Classicalquantum correspondence for
barrier crossing in a driven bistable potential, J. Phys. A 25, L303 (1992).
[19] V. B. Magalinski˘ı, Dynamical model in the theory of the Brownian motion,
Zh. Eksp. Teor. Fiz. 36, 1942 (1959), [Sov. Phys. JETP 9, 1381 (1959)].
[20] R. Zwanzig, Ensemble method in the theory of irreversibility, J. Chem. Phys.
33, 1338 (1960).
[21] R. P. Feynman and F. L. Vernon, The theory of a general quantum system
interacting with a linear dissipative system, Ann. Phys. (N.Y.) 24, 118 (1963).
[22] R. Bl¨ umel, R. Graham, L. Sirko, U. Smilansky, H. Walther, and K. Yamada,
Microwave excitation of Rydberg atoms in presence of noise, Phys. Rev. Lett.
62, 341 (1989).
[23] A. G. Fainshtein, N. L. Manakov, and L. P. Rapoport, Some general proper
ties of quasienergetic spectra of quantum systems in classical monochromatic
ﬁelds, J. Phys. B 11, 2561 (1978).
[24] N. L. Manakov, V. D. Ovsiannikov, and L. P. Rapoport, Atoms in a laser
ﬁeld, Phys. Rep. 141, 319 (1986).
[25] T. Dittrich, P. H¨ anggi, G.L. Ingold, B. Kramer, G. Sch¨ on, and W. Zwerger,
Quantum Transport and Dissipation (WileyVCH, Weinheim, 1998).
[26] R. Graham and R. H¨ ubner, Generalized QuasiEnergies and Floquet States for
a Dissipative System, Ann. Phys. (N.Y.) 234, 300 (1994).
[27] S. Kohler, T. Dittrich, and P. H¨ anggi, FloquetMarkovian description of the
parametrically driven, dissipative harmonic quantum oscillator, Phys. Rev. E
55, 300 (1997).
[28] C. Zerbe and P. H¨ anggi, Brownian parametric quantum oscillators with dissi
pation, Phys. Rev. E 52, 1533 (1995).
[29] J. I. Cirac and P. Zoller, Quantum Computations with Cold Trapped Ions,
Phys. Rev. Lett. 74, 4091 (1995).
[30] O. Bohigas, S. Tomsovic, and D. Ullmo, Manifestations of classical phase space
structures in quantum mechanics, Phys. Rep. 223, 43 (1993).
[31] M. Latka, P. Grigolini, and B. J. West, Chaos and avoided level crossing, Phys.
Rev. E 50, 596 (1994).
[32] M. Latka, P. Grigolini, and B. J. West, Chaosinduced avoided level crossing
and tunneling, Phys. Rev. A 50, 1071 (1994).
[33] M. Latka, P. Grigolini, and B. J. West, Control of dynamical tunneling in a
bichromatically driven pendulum, Phys. Rev. E 50, R3299 (1994).
[34] S. Kohler, R. Utermann, P. H¨ anggi, and T. Dittrich, Coherent and incoherent
chaotic tunneling near singletdoublet crossings, Phys. Rev. E 58, 7219 (1998).
[35] G. Floquet, Ann. de l’Ecole Norm. Sup. 12, 47 (1883).
References 95
[36] G. Casati and L. Molinari, “Quantum Chaos” with TimePeriodic Hamiltoni
ans, Prog. Theor. Phys. Suppl. 98, 287 (1989).
[37] S.I. Chu, Generalized Floquet theoretical approach to intenseﬁeld multiphoton
and nonlinear optical processes, Adv. Chem. Phys. 73, 739 (1989).
[38] J. S. Howland, Stationary Scattering Theory for Timedependent Hamiltoni
ans, Math. Ann. 207, 315 (1974).
[39] J. H. Shirley, Solution of the Schr¨ odinger Equation with a Hamiltonian Peri
odic in Time, Phys. Rev. 138, B979 (1965).
[40] H. Sambe, Steady States and Quasienergies of a QuantumMechanical System
in an Oscillating Field, Phys. Rev. A 7, 2203 (1973).
[41] D. J. Moore, Time dependence in quantum mechanics—Floquet theory and the
Berry phase, Helv. Phys. Acta 66, 3 (1993).
[42] W. Magnus and S. Winkler, Hill’s Equation (Dover, New York, 1979).
[43] J. von Neumann, Mathematical foundations of quantum mechanics (Princeton
Univ. Press, Princeton, 1955).
[44] C. CohenTannoudji, J. DupontRoc, and G. Grynberg, Atom photon interac
tion: basic processes and applications (Wiley, New York, 1992).
[45] H. Risken, The FokkerPlanck Equation, Vol. 18 of Springer Series in Syner
getics (Springer, Berlin, 1984).
[46] N. Moiseyev, Timeindependent scattering theory for general timedependent
Hamiltonians, Comments At. Mol. Phys. 31, 87 (1995).
[47] U. Peskin and N. Moiseyev, The solution of the timedependent Schr¨ odinger
equation by the (t, t
) method: Theory, computational algorithm and applica
tions, J. Chem. Phys. 99, 4590 (1993).
[48] H. Goldstein, Classical Mechanics, 2nd ed. (AddisonWesley, Reading, 1980).
[49] U. Weiss, Quantum Dissipative Systems, Vol. 2 of Series in Modern Condensed
Matter Physics (World Scientiﬁc, Singapore, 1993).
[50] W. H. Louisell, Quantum Statistical Properties of Radiation (Wiley & Sons,
New York, 1973).
[51] F. Haake, in Quantum Statistics in Optics and SolidState Physics, Vol. 66
of Springer Tracts in Modern Physics, edited by G. H¨ ohler (Springer, Berlin,
1973).
[52] H. Grabert, in Projection Operator Techniques in Nonequilibrium Statistical
Mechanics, Vol. 95 of Springer Tracts in Modern Physics, edited by G. H¨ ohler
(Springer, Berlin, 1982).
[53] P. H¨ anggi, P. Talkner, and M. Borkovec, Reactionrate theory: ﬁfty years after
Kramers, Rev. Mod. Phys. 62, 251 (1990).
96 References
[54] H. Grabert, U. Weiss, and P. Talkner, Quantum Theory of the Damped Har
monic Oscillator, Z. Phys. B 55, 87 (1984).
[55] P. Riseborough, P. H¨ anggi, and U. Weiss, Exact Results for a Damped Quan
tum Mechanical Harmonic Oscillator, Phys. Rev. A 31, 471 (1985).
[56] G. A. Voth, Feynman Path Integral of Quantum Mechanical TransitionState
Theory, J. Phys. Chem. 97, 8365 (1993).
[57] D. E. Makarov and N. Makri, Control of dissipative tunnelling dynamics by
continuous wave electromagnetic ﬁelds: Localization and largeamplitude co
herent motion, Phys. Rev. E 52, 5863 (1995).
[58] N. Makri, Stabilization of localized states in dissipative tunneling systems in
teracting with monochromatic ﬁelds, J. Chem. Phys. 106, 2286 (1997).
[59] R. Zwanzig, Nonlinear Generalized Langevin Equations, J. Stat. Phys. 9, 215
(1973).
[60] H. Grabert, P. Schramm, and G.L. Ingold, Quantum Brownian Motion: The
Funtional Integral Approach, Phys. Rep. 168, 115 (1988).
[61] R. Benguria and M. Kac, Quantum Langevin Equation, Phys. Rev. Lett. 46,
1 (1981).
[62] A. Schmid, On a Quasiclassical Langevin Equation, J. Low Temp. Phys. 49,
609 (1982).
[63] G. W. Ford and M. Kac, On the Quantum Langevin Equation, J. Stat. Phys.
46, 803 (1987).
[64] P. H¨ anggi, Generalized Langevin Equations: A Useful Tool for the Perplexed
Modeller of Nonequilibrium Fluctuations?, in Stochastic Dynamics, Vol. 484
of Lecture Notes in Physics, edited by L. SchimanskyGeier and T. P¨ oschel
(Springer, Berlin, 1997), p. 15.
[65] R. P. Feynman and A. R. Hibbs, Quantum Mechanics and Path Integrals
(McGrawHill, New York, 1965).
[66] L. S. Schulman, Techniques and Applications of Path Integrals (Wiley & Sons,
New York, 1981).
[67] N. Makri and D. E. Makarov, Tensor propagator for iterative quantum evolu
tion of reduced density matrices. I. Theory, J. Chem. Phys. 102, 4600 (1995).
[68] P. Pechukas, in Proc. NATO ASI “Largescale molecular systems”, edited by
W. Gans (Plenum Press, New York, 1991), Vol. B258, p. 123.
[69] P. Pechukas, Reduced Dynamics Need Not Be Completely Positive, Phys. Rev.
Lett. 73, 1060 (1994).
[70] L. Di´ osi, CaldeiraLeggett master equation and medium temperatures, Physica
A 199, 517 (1993).
[71] L. Di´ osi, On HighTemperature Markovian Equation for Quantum Brownian
Motion, Europhys. Lett. 22, 1 (1993).
References 97
[72] M. Grifoni, M. Sassetti, J. Stockburger, and U. Weiss, Nonlinear response of
a periodically driven damped twostate system, Phys. Rev. E 48, 3497 (1993).
[73] M. Grifoni, M. Sassetti, P. H¨ anggi, and U. Weiss, Cooperative eﬀects in the
nonlinearly driven spinboson system, Phys. Rev. E 52, 3596 (1995).
[74] T.S. Ho, K. Wang, and S.I. Chu, FloquetLiouville supermatrix approach:
Time development of densitymatrix operator and multiphoton resonance spec
tra in intense laser ﬁleds, Phys. Rev. A 33, 1798 (1986).
[75] R. Bl¨ umel, A. Buchleitner, R. Graham, L. Sirko, U. Smilansky, and H.
Walther, Dynamical localization in the microwave interaction of Rydberg
atoms: The inﬂuence of noise, Phys. Rev. A 44, 4521 (1991).
[76] T. Dittrich, B. Oelschl¨ agel, and P. H¨ anggi, Driven Dissipative Tunneling, Eu
rophys. Lett. 22, 5 (1993).
[77] B. Oelschl¨ agel, T. Dittrich, and P. H¨ anggi, Damped periodically driven quan
tum transport in bistable systems, Acta Physica Polonica B 24, 845 (1993).
[78] T. Dittrich, P. H¨ anggi, B. Oelschl¨ agel, and R. Utermann, Driven Tunneling:
New Possibilities for Coherent and Incoherent Quantum Transport, in 25 Years
of NonEquilibrium Statistical Mechanics, Vol. 445 of Lecture Notes in Physics,
edited by J. J. Brey (Springer, Berlin, 1995), p. 269.
[79] F. Haake, Quantum Signatures of Chaos, Vol. 54 of Springer Series in Syner
getics (Springer, Berlin, 1991).
[80] R. Graham, Global and Local Dissipation in a Quantum Map, Z. Phys. B 59,
75 (1985).
[81] V. S. Popov and A. M. Perelomov, Parametric exitation of a quantum oscil
lator, Zh. Eksp. Teor. Fiz. 55, 589 (1968), [Sov. Phys. JETP 29, 719 (1969)].
[82] V. S. Popov and A. M. Perelomov, Parametric exitation of a quantum os
cillator II, Zh. Eksp. Teor. Fiz. 57, 1684 (1969), [Sov. Phys. JETP 30, 910
(1970)].
[83] V. S. Perelomov, A. M. Popov, Grouptheoretical aspects of the variable fre
quency oscillator problem, Teor. Mat. Fiz. 1, 360 (1969).
[84] W. Paul, Electromagnetic traps for charged and neutral particles, Rev. Mod.
Phys. 62, 531 (1990).
[85] N. W. McLachlan, Theory and Applications of Mathieu Functions (Dover Pub
lications Inc., New York, 1964).
[86] H. R. Lewis, Jr., Classical and Quantum Systems with TimeDependent
HarmonicOscillatorType Hamiltonians, Phys. Rev. Lett. 18, 510, 636 (1967).
[87] H. R. Lewis, Jr. and W. B. Riesenfeld, An Exact Quantum Theory of the
TimeDependent Harmonic Oscillator and of a Charged Particle in a Time
Dependent Electromagnetic Field, J. Math. Phys. 10, 1458 (1969).
[88] L. S. Brown, Quantum Motion in a Paul Trap, Phys. Rev. Lett. 66, 527 (1991).
98 References
[89] G. Schrade, V. I. Man’ko, W. P. Schleich, and R. J. Glauber, Wigner Functions
in the Paul Trap, Quantum Semiclass. Opt. 7, 307 (1995).
[90] J. G. Hartley and J. R. Ray, Coherent States for the timedependent harmonic
oscillator, Phys. Rev. D 25, 382 (1982).
[91] D. B. Monteoliva, B. Mirbach, and H.J. Korsch, Global and local dynamical
invariants and quasienergy states of timeperiodic Hamiltonians, Phys. Rev.
A 57, 746 (1998).
[92] I. M. Gradshteyn, I. S. Ryzhik, Table of Integrals, Series, and Products, 5th
ed. (Academic Press, San Diego, 1994).
[93] G. Lindblad, On the Generators of Quantum Dynamical Semigroups, Com
mun. Math. Phys. 48, 119 (1976).
[94] P. H¨ anggi and H. Thomas, Stochastic Processes: Time Evolution, Symmetries
and Linear Response, Phys. Rep. 88, 206 (1982).
[95] E. Kamke, Diﬀerentialgleichungen, Vol. II: Partielle Diﬀerentialgleichungen,
6th ed. (Teubner, Stuttgart, 1979).
[96] L. H’walisz, P. Jung, P. H¨ anggi, P. Talkner, and L. SchimanskyGeier, Colored
noise driven systems with inertia, Z. Phys. B 77, 471 (1989).
[97] C. Zerbe, P. Jung, and P. H¨ anggi, Brownian parametric oscillators, Phys. Rev.
E 49, 3626 (1994).
[98] U. M. Titulaer, A systematic solution procedure for the FokkerPlanck equation
of a Brownian particle in the high friction case, Physica A 91, 321 (1978).
[99] W. A. Lin and L. E. Ballentine, Quantum tunneling and regular and irregular
quantum dynamics of a driven doublewell oscillator, Phys. Rev. A 45, 3637
(1992).
[100] P. H¨ anggi, R. Utermann, and T. Dittrich, Tunnel Splittings and Chaotic Trans
port in Periodically Driven Bistable Systems, Physica B 194196, 1013 (1994).
[101] E. M. Zanardi, J. Guti´errez, and J. M. Gomez Llorente, Mixed dynamics and
tunneling, Phys. Rev. E 52, 4736 (1995).
[102] E. Doron and S. D. Frischat, Semiclassical description of tunneling in mixed
systems: Case of the annular billiard, Phys. Rev. Lett 75, 3661 (1995).
[103] S. D. Frischat and E. Doron, Dynamical tunneling in mixed systems, Phys.
Rev. E 57, 1421 (1998).
[104] F. Leyvraz and D. Ullmo, The level splitting distribution in chaosassisted
tunneling, J. Phys. A 29, 2529 (1996).
[105] R. Roncaglia, L. Bonci, F. M. Izrailev, B. J. West, and P. Grigolini, Tunneling
versus chaos in the kicked Harper model, Phys. Rev. Lett. 73, 802 (1994).
[106] T. Dittrich and R. Graham, Quantum Eﬀects in the Steady State of the Dis
sipative Standard Map, Europhys. Lett. 4, 263 (1987).
References 99
[107] F. Grossmann, P. Jung, T. Dittrich, and P. H¨ anggi, Tunneling in a Periodically
Driven Bistable System, Z. Phys. B 84, 315 (1991).
[108] A. Peres, Dynamical quasidegeneracies and quantum tunneling, Phys. Rev.
Lett. 67, 158 (1991).
[109] F. Großmann and P. H¨ anggi, Localization in a Driven TwoLevel Dynamics,
Europhys. Lett. 18, 571 (1992).
[110] A. J. Lichtenberg and M. A. Liebermann, Regular and Stochastic Motion,
Vol. 38 of Applied Mathical Sciences (Springer, New York, 1983).
[111] D. F. Escande, Stochasticity in classical Hamiltonian systems: universal as
pects, Phys. Rep. 121, 165 (1985).
[112] L. E. Reichl and W. M. Zheng, in Directions in Chaos, edited by H. B. Lin
(World Scientiﬁc, Singapore, 1987), Vol. 1, p. 17.
[113] M. Wilkinson, Tunnelling between tori in phase space, Physica D 21, 341
(1986).
[114] M. Wilkinson, Narrowly avoided crossings, J. Phys. A 20, 635 (1987).
[115] L. E. Reichl, The Transition to Chaos: In Conservative and Classical Systems:
Quantum Manifestations (Springer, New York, 1992).
[116] R. B. Shirts and W. P. Reinhardt, Approximate constants of motion for clas
sically chaotic vibrational dynamics: Vague tori, semiclassical quantization,
and classical intramolecular energy ﬂow, J. Chem. Phys. 77, 5204 (1982).
[117] S.J. Chang and K.J. Shi, Time Evolution and Eigenstates of a Quantum
Iterative System, Phys. Rev. Lett. 55, 269 (1985).
[118] S.J. Chang and K.J. Shi, Evolution and exact eigenstates of a resonant quan
tum system, Phys. Rev. A 34, 7 (1986).
[119] B. Mirbach and H. J. Korsch, Semiclassical quantization of KAM resonances
in timeperiodic systems, J. Phys. A 27, 6579 (1994).
[120] T. Gorin, H. J. Korsch, and B. Mirbach, Phasespace localization and level
spacing distributions for a driven rotor with mixed regular/chaotic dynamics,
Chem. Phys. 217, 145 (1997).
[121] F. C. Moon and G.X. Li, The fractal dimension of the twowell potential
strange attractor, Physica D 17, 99 (1985).
[122] F. C. Moon and G.X. Li, Fractal Basin Boundaries and Homoclinic Orbits for
Periodic Motion in a TwoWell Potential, Phys. Rev. Lett. 55, 1439 (1985).
[123] W. SzemplinskaStupnicka, CrossWell Chaos and Escape Phenomena in
Driven Oscillators, Nonlinear Dynamics 3, 225 (1992).
[124] A. Wehrl, On the relation between classical and quantummechanical entropy,
Reps. Math. Phys. 16, 353 (1979).
100 References
[125] A. Messiah, Quantum Mechanics, 3rd ed. (Wiley & Sons, New York, 1965),
Vol. I.
[126] M. V. Berry, Semiclassical mechanics in phase space: a study of Wigner’s
function, Proc. R. Soc. A 287, 237 (1977).
[127] M. Brack and R. K. Bhaduri, Semiclassical Physics, Vol. 96 of Frontiers in
Physics (AddisonWesley, New York, 1997).
[128] R. J. Glauber, Coherent and Incoherent States of a Radiation Field, Phys.
Rev. 131, 2766 (1963).
[129] E. C. G. Sudarshan, Equivalence of semiclassical and quantum mechanical
description of statistical light beams, Phys. Rev. Lett. 10, 277 (1963).
[130] E. P. Wigner, On the Quantum Correction for Thermodynamic Equilibrium,
Phys. Rev. 40, 749 (1932).
[131] K. Husimi, Proc. Phys. Math. Soc. Japan 22, 264 (1940).
[132] P. J. Drummond and C. W. Gardiner, Generalised Prepresentations in quan
tum optics, J. Phys. A 13, 2353 (1980).
[133] M. Hillery, R. F. O’Connell, M. Scully, and E. P. Wigner, Distribution Func
tions in Physics: Fundamentals, Phys. Rep. 106, 121 (1984).
[134] C. W. Gardiner, Handbook of Stochastic Methods, Vol. 13 of Springer Series
in Synergetics, 2nd ed. (Springer, Berlin, 1985).
[135] K. E. Cahill and R. J. Glauber, Density Operators and Quasiprobability Dis
tributions, Phys. Rev. 177, 1883 (1969).
[136] K. E. Cahill and R. J. Glauber, Ordered Expansions in Boson Amplitude Op
erators, Phys. Rev. 177, 1857 (1969).
[137] H. Weyl, Quantenmechanik und Gruppentheorie, Z. Phys. 46, 1 (1927).
[138] E. Fick, Einf¨ uhrung in die Grundlagen der Quantentheorie, 6th ed. (Aula,
Wiesbaden, 1988).
[139] R. Alicki and K. Lendi, in Quantum Dynamical Semigroups and Applications,
Vol. 286 of Lecture Notes in Physics, edited by W. Beiglb¨ ock (Springer, Berlin,
1987).
[140] P. Talkner, The Failure of the Quantum Regression Hypothesis, Ann. Phys.
(N.Y.) 167, 390 (1986), see Appendix C therein.
[141] V. Ambegaokar, Quantum Brownian Motion and its Classical Limit, Berichte
der Bunsengesellschaft 95, 400 (1991).
[142] W. H. Zurek, S. Habib, and J. P. Paz, Coherent states via decoherence, Phys.
Rev. Lett. 70, 1187 (1993).
Acknowledgment
First, I would like to thank Prof. Dr. Peter H¨ anggi and Prof. Dr. Thomas Dittrich
for accepting me as a Doktorand and for giving me the opportunity to work on an
intriguing project. I gained a lot from their experience. I’m grateful to Thomas also
for collaborating with me, even while staying at several remote places all over the
world.
Christine Zerbe provided the numerical code for the exact solution of the dissipative,
parametrically driven harmonic oscillator.
During the time I spent in Augsburg, I enjoyed many discussions on dissipative quan
tum mechanics and driven quantum systems with Milena Grifoni, Ludwig Hartmann,
GertLudwig Ingold, Michael Thorwart, Ralf Utermann, and Dietmar Weinmann.
Especially Gert has always been a competent and interested partner for discussions
and questions during his Teerunde.
Ralf Utermann not only built up a great computer environment, but also kept it
(mostly :) well tuned. With him, Peter Schmitteckert, and Andr´e Wobst, I had
lots of fruitful discussions about eﬃcient computing and objectoriented program
ming.
The members of the groups Theoretische Physik I and Theoretische Physik II—
present and former ones—provided a stimulating and pleasant working atmosphere.
Thomas Dittrich, GertLudwig Ingold, and Sonja Thunnessen were of indispensible
help in proofreading and improving the English of this thesis.
Last, but not least, I’m grateful to the DFGSchwerpunkt “Zeitabh¨ angige Ph¨ ano
mene und Methoden in Quantensystemen der Physik und Chemie” for founding my
position at the Universit¨ at Augsburg from September ’95 to February ’99 under
grant no. Di 511/1 and Di 511/2 as well as for the possibility to participate in
conferences in Freiburg, Berlin, Dresden, W¨ urzburg, and Haifa.
Erster Berichter: Prof. Dr. Peter H¨nggi a Zweiter Berichter: Prof. Dr. Thomas Dittrich Tag der m¨ ndlichen Pr¨ fung: 5. M¨rz 1999 u u a
Contents
1 Introduction 2 Driven quantum systems and Floquet theory 2.1 Discrete timetranslation and Floquet ansatz . 2.2 Composite Hilbert space . . . . . . . . . . . . 2.3 Properties of Floquet states . . . . . . . . . . 2.4 The propagator . . . . . . . . . . . . . . . . . 2.5 Numerical computation of Floquet states . . . 2.5.1 Floquetmatrix methods . . . . . . . . 2.5.2 Propagator methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 5 5 7 9 11 11 12 12 15 15 16 18 20 23 23 24 25 25 27 28 31 31 34 36 38 39 40 41 43 43
3 Quantum dissipation and Markov approximation 3.1 The systembath model . . . . . . . . . . . . . . . . 3.2 Quantum Langevin equation . . . . . . . . . . . . . 3.3 Inﬂuence functional . . . . . . . . . . . . . . . . . . 3.4 Markovian master equation . . . . . . . . . . . . .
4 Driving and dissipation: FloquetMarkov theory 4.1 Simple inclusion of the driving . . . . . . . . . . . . . . . . . . . 4.2 An improved Markovian master equation . . . . . . . . . . . . . 4.3 Decomposition into Floquet basis . . . . . . . . . . . . . . . . . 4.3.1 Matrix elements . . . . . . . . . . . . . . . . . . . . . . . 4.3.2 Rotatingwave approximation . . . . . . . . . . . . . . . 4.4 The dissipative quantum map and its numerical implementation 5 The 5.1 5.2 5.3 5.4 parametrically driven harmonic oscillator The model and its classical dynamics . . . . . . . . . . . . Floquet states in stable regimes . . . . . . . . . . . . . . . FloquetMarkov description in full RWA . . . . . . . . . . Basisindependent description beyond RWA . . . . . . . . 5.4.1 Wigner representation and FokkerPlanck equation 5.4.2 WignerFloquet solutions . . . . . . . . . . . . . . . 5.4.3 Inﬂuence of the driving on the master equation . . 5.5 Asymptotics . . . . . . . . . . . . . . . . . . . . . . . . . . 5.5.1 The conservative limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . .2 Tunneling.1 Classical attractor . . . . 6. . .3 harmonic oscillator Number states as a basis set . .1 Lindblad form . . . .3 Asymptotic state . 48 . . . . . .1 Threelevel crossings . . . . . . . . . . . . . . . . . . . . . . . . .ii 5. . . . . . . 6. . 6. . . . . . . . . . . . . . .6 5. . . . . . . 6. . . . . . . . . . .3. . . . .7 Contents 5. . . . . . . . . . . . . . . . . . . .2 Coherence and entropy . . . . . Coherent states . . . . . . . . . . . . 6. . . . . . . . . 51 52 53 55 56 58 58 65 67 72 72 75 79 . . 6. . . . . . driving. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Husimi function and Wehrl entropy B The density operator 87 B. . . . . . . . . . . . . . . .2 A. . 6. . . . . . . . . . . . . . . . . . A. . . . . .3 The onset of chaos . 6. . . . . . . . . . . . . . . . 81 82 83 83 84 85 6 The harmonically driven doublewell potential 6. . . . . 7 Summary and outlook A The A. . . . . . .2. . . . . . . . . . . . . . 44 Numerical results . . . . . . . . . . . . .3. . . .3 Signatures of chaos in the asymptotic state . . . . . . . . . .3. . . . . . . . . . .3. . . . . . . .2 Chaotic tunneling near singletdoublet crossings 6. . . . . . . . . . . . . . . . . . . .1. . . . . . . . . . . . . A. . . . . Quasiprobabilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Wigner function . . . . .2 Dissipative chaosassisted tunneling . . . . . . . . . . . . . . . . . . . . . .2 Quantum attractor . . . . . . . . . . and dissipation . .1 The model . 45 Conclusion . . . .5. . . . . . . . .1 Symmetries . . 6. . . . . . . . . . . . 87 B. . . . . . . . . . . . . . . 88 C Solution of the FokkerPlanck equation References Acknowledgment 89 93 101 . .2. . .1. . . . .2. . . . .1 A. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1. . . . . .2 The hightemperature limit . .
g. scars along unstable periodic orbits in the wave functions [4]. after which the quantum dynamics becomes quasiperiodic and thus. the transition from regular motion to chaos is most clearly visible in the change of the phasespace structure: With increasing nonlinearity. the classical dynamics leaves in the corresponding quantum system. A timedependent external ﬁeld acting on . A reason may be the fact that by including dissipation. the positionmomentum uncertainty does not allow for the arbitrarily ﬁne classical phasespace structures and results in coarsegraining over an area which is given by Planck’s quantum of action. restores the characteristics of classical features at least to some extent [2]. chaotic dynamics is characterized by a sensitive dependence on the initial conditions: Neighboring phasespace points start to diverge exponentially in time and a completely deterministic system evolves in a practically diﬀusive manner on a chaotic sea [3]. In classical Hamiltonian systems. its signatures like. since one has to deal with density matrices instead of wave functions. the coherent transport through a potential barrier. Another characteristic quantum feature is the discreteness of the energy levels in bounded systems. any disruption of coherence. In complex systems. While the motion along regular tori is stable and predictable for long times. While the mutual inﬂuence of quantum coherence and classical chaos has been an extensive ﬁeld of research since many years. In the fully chaotic case. the additional eﬀects caused by coupling the chaotic system to an environment. A generic setting for the observation of tunneling is a symmetric bistable potential whose wells are separated by a static energy barrier.. On a quantum level. the eigenenergies are anticorrelated and the inverse of their mean spacing deﬁnes a time scale.1 Introduction The interplay of classical chaos and dissipation in a quantum system bears interesting eﬀects at the border between classical and quantum mechanics like. the suppression of classical chaos by quantum interference [1] or its restauration by dissipation [2]. Thus. eigenenergies are eﬀectively random numbers whose statistical properties depend on the integrability of the corresponding classical dynamics [6–8]. e. One of the most intriguing quantum eﬀects is tunneling. e. at most.. the socalled break time. or the centering of Husimi functions on classical manifolds [5].g. classical chaos is suppressed [1]. the computational eﬀort grows drastically. It was originally proposed by Hund [9] to explain the ammonium spectrum and studied since then in various modiﬁcations. have been studied only rarely. like it occurs due to the coupling to an environment. regular tori start to dissolve in a chaotic layer which grows in size until it covers the whole phase space. Therefore. namely dissipation and decoherence. This suppression of chaos relies on the perfect coherence of a superposition which remains for arbitrary long times.
for the description of strongly driven. the simpliﬁcation brought about by the Markovian description is achieved only at the expense of accuracy. 21]. Probably the ﬁrst proof that such a systembath scheme results in dissipative quantum mechanics was given by Magalinski˘ [19] for a harmonic oscillator. The small but ﬁnite overlap of the tunnel doublets with the chaotic states. Using a perturbative approach. an external driving can modify the tunnel rate by orders of magnitude or even bring tunneling to a complete standstill [10]. quantum optics. do not commute. nonlinear systems subject to weak dissipation. 12]. Master equations of this kind have been applied to various problems in solid state physics. Thus. As soon as the chaotic layer grows in size and attains a signiﬁcant overlap with the tunnel doublets. like harmonic potentials or twolevel systems—the investigation of dissipative systems with complex dynamics requires to fall back to the weakcoupling regime. 27]. even if its eﬀect is barely visible in the classical phase space. Tunneling is particularly sensitive to any disruption of coherence—in presence of dissipation it becomes a transient eﬀect that fades out on a ﬁnite time scale [11. and chemistry. While the Floquet formalism is exact and essentially amounts to using an optimal representation for the treatment of timeperiodic problems [23–25]. Here.. it is desirable to combine a Markovian approach to quantum dissipation.2 Introduction such a system may entail dramatic consequences for the quantum dynamics. consistent with the fundamental laws of quantum mechanics. Zwanzig ı [20] derived from this model a Markovian master equation for a general classical system subject to weak dissipation. They are apparent already in the classical phase space since chaos comes into play and the separatrix which encloses the wells is replaced by a chaotic layer. However. the tunnel rates—the essence of chaosassisted tunneling [13–15].e. rather than by a static barrier. leading to a master equation for the density operator. we will implement a Markovian approach to quan . consequently. which enabled studying dissipative quantum systems. with states which are localized in the chaotic layer. typically increases the tunnel splittings and. the tunnel splittings become of the order of the mean level spacing [16] and tunneling is replaced by chaotic diﬀusion [16–18]. Depending on the driving amplitude and frequency. The most successful approach to dissipation in quantum mechanics. Later. Within the present work. a subtle technical diﬃculty lies in the fact that the truncation of the longtime memory introduced by the bath. beyond a weakcoupling limit. Driving the doublewell potential with a frequency near the classical resonances results in even more signiﬁcant consequences. we therefore observe chaotic tunneling—coherent transport between regular islands which are separated by a chaotic layer. In the corresponding quantum system. and the inclusion of the driving. Caldeira and Leggett eliminated the bath exactly [11. i. This implies that the result of the Markov approximation depends on whether the driving is considered in its derivation or not [26. even a partially analytical solution of the resulting pathintegral expression is only feasible for the simplest systems. with the Floquet formalism that allows to treat timeperiodic forces of arbitrary strength and frequency [22]. is based on the coupling of the conservative system to external degrees of freedom.
A number of merely technical issues is deferred to the appendix. will be used as a working model for the investigation of chaotic tunneling in presence of dissipation. Recent studies of nondissipative chaotic tunneling suggest that tunneling is accelerated by the inﬂuence of chaotic states. These traps have gained new interest very recently. Within this FloquetMarkov approach. we expect to observe a novel dissipative tunnel scenario which is on the one hand richer than the conservative dynamics and on the other hand substantially diﬀerent from the familiar twostate tunneling. Parts of this thesis have already been published in Refs. the loss of coherence once the computation has started. besides the preparation of the ground state. This thesis is organized as follows: In Chapter 2 we give an introduction to Floquet theory for quantum systems with periodic timedependence. Thereby the main obstacle is. all approximative steps can be reliably checked since an exact solution is at hand [28]. for each of which we have. respectively. replacing a doublet structure by a threelevel dynamics [30–33]. in turn. Besides being an exactly solvable model with yet nontrivial dynamics. couples these states indirectly to all other states of the system and. since it describes the motion of an ion in a Paul trap. A brief review of the systembath model for quantum dissipation and a derivation of a Markovian master equation is provided in Chapter 3 and combined with Floquet theory in Chapter 4 to obtain a Markovian description of periodically driven quantum systems subject to weak dissipation. this system is interesting in its own right. will serve predominantly to test diﬀerent approximation schemes for the FloquetMarkov master equation and to study the modiﬁcation of its dissipative part brought about by the driving. For this linear system. thus. since they form the central system in a scheme for a quantum computer [29] whose experimental realization is currently attempted. 34]. . The bath. one central question in mind: The parametrically driven harmonic oscillator . Chapter 7 serves to summarize the main results. besides other interesting aspects. a system which exhibits complex nonlinear dynamics. we investigate the dynamics of the parametrically driven harmonic oscillator and the driven doublewell potential in Chapters 5 and 6. [27. The harmonically driven quartic doublewell potential .Introduction 3 tum dissipation based on the Floquet formalism to the investigation of two diﬀerent systems.
4 .
also eigenfunctions o of the symmetry operator [38]. Onedimensional driven systems also play an important role as models for (quantum) chaos: Their “one and a half degrees of freedom” represent the minimal requirement for nonintegrable dynamics [36]. In this chapter we give an introduction to Floquet theory for quantum systems with periodic time dependence [12. H(t) = H(t + T ). i. the ﬁeld is in a coherent state with large mean photon number and. This implies that an explicit time dependence of the Hamiltonian serves as a substitute for a canonical degree of freedom and raised interest in a theory for quantum systems with explicit periodic time dependence. thus. can be described adequately by its expectation value. 37]. commutes with the operator H(t) − i ∂t . Thus. thus an extension of Floquet theory [35] from classical to quantum mechanics. In quantum mechanics.2) the related symmetry operation is a discrete time translation by one period of the driving. given by a function harmonic in time.e. Thus. but the backaction of the system on the ﬁeld is negligible. the inﬂuence of the ﬁeld on the system is typically so strong that a treatment beyond perturbation theory becomes necessary. For a Hamiltonian with T periodic time dependence. the solutions of the Schr¨dinger equation are.2 Driven quantum systems and Floquet theory Interactions of quantum systems with strong laser ﬁelds are characterized by two properties of the ﬁeld: On the one hand. where we put strong focus on the properties of Floquet states and numerical methods which we use in subsequent chapters.1 Discrete timetranslation and Floquet ansatz To reduce the complexity of a physical system. T = 2π . (2. Ω (2. its symmetries are analyzed to obtain a proper ansatz for the symmetryreduced solutions. 2.. besides a timedependent phase factor.3) . On the other hand. 23–25. 36. they exemplify the simplest quantum systems with chaotic classical counterpart. ST : t → t + T. symmetry is expressed by an operator S which leaves the Schr¨dinger equation o H(t) − i ∂ ∂t ψ(t) = 0 (2.1) invariant.
the eigenvalues of S are pure phase factors and we may assume for an eigenfunction ψ(t) the eigenvalue exp(−iθ). whereas the longtime dynamics is governed by the phase factors exp(−i α t/ ). We emphasize that the T periodic timedependence of the Floquet states is only relevant for the dynamics within a period of the driving. ∂t (2.. a general solution of the Schr¨dinger equation (2. Inserting (2. The Floquet states φα (t) are.10) is one of the main tasks in dealing with periodically timedependent systems. there exists a complete set {ψα (t) } of solutions of the Schr¨dinger equation which have Floquet structure.11) (2. φα (t) = φα (t + T ) . By inserting this eigenvalue equation into the ansatz ψ(t) = e−i t/ φ(t) .4) However.1) is given by a supero position of many Floquet states.8) = θ/T. . (2. they are of the form ψα (t) = e−i α t/ φα (t) .6) which means that φ(t) is periodic in time. the determination of the Floquet states from (2. The α have the dimension o energy and in periodically driven systems play a role analogous to the eigenenergies in timeindependent systems.e.5).7) (2.5) (2. 40] H(t)φ(t) = φ(t) with the Hermitian Floquet Hamiltonian [40] H(t) = H(t) − i ∂ . they are called quasienergies. (2. alike the Hamiltonian. Thus for a system which obeys discrete timetranslational symmetry. in contrast to the ψα (t) .6 Driven quantum systems and Floquet theory As symmetry operations have to conserve the norm of any wavefunction.10) Technically. ST ψ(t) = ψ(t + T ) = e−iθ ψ(t) . 39. θ ∈ R. we obtain the condition φ(t) = φ(t + T ) . (2. not solutions of the Schr¨dinger equation. o i. (2. ψ(t) = uα e−i α t/ φα (t) .9) α and is in general not of the form (2. In analogy to the quasimomentum of electrons in spatially periodic systems.5) into the Schr¨dinger equation yields the eigenvalue equation for o the Floquet states [23.
13) by (2. R is the space of squareintegrable functions [43].16) ϕ∗ (t) ϕn (t ) = δT (t − t ). In these cases. According to (2. which describes the system’s degrees of freedom. thus the Floquet indices may be complex. we deﬁne the vectors n ϕn (t) = tn T . ϕn ) = δn. For a bounded particle moving in a potential.2 Composite Hilbert space 7 From a grouptheoretical point of view. as well as the Floquet states φα (t) . The inner product (2. the Floquet states are elements of the space of T periodic functions. are elements of a Hilbert space R.2. Solutions of Floquet structure are found for dynamical systems that can be described by diﬀerential equations with periodically timedependent coeﬃcients [35. φφ = 1 T T dt φ(t)φ (t) . In many cases. The basis set {ϕn } is orthonormalized and complete [43].10) is in general nonHermitian.15) (2.8). 2. (2. We also use this fact for the solution of classical equations of motion and for the solution of FokkerPlanck equations in subsequent chapters. To avoid confusion with elements of conﬁguration space R.12) (f. 1 T n (2.14) For a basis independent notation. each Floquet state φα (t) belongs to an irreducible representation of an Abelian group.2 Composite Hilbert space The state ψ(t) of a system. This exponent allows for an interpretation as a Berry phase [41].17) . (ϕn . We combine the periodic time dependence of the Floquet states with their spatial degrees of freedom and interpret them as elements of a composite Hilbert space R ⊗ T . Ω= 2π . R can be approximated by a Hilbert space with ﬁnite dimension. the eigenvalue equation which corresponds to (2. 42]. It is possible to describe the time dependence of the Floquet states within the framework of a Hilbert space theory. we mark these vectors by an index T . characterized by the Floquet exponent θα = α T / [40].12) is extended accordingly. T (2.n . however. 0 (2. denoted by T [40]. g) = T 0 and a set of orthonormalized basis functions reads [43] ϕn (t) = e−inΩt . T n ∈ Z. n where δT denotes the T periodic delta function. An inner product on T is deﬁned by 1 T dt f ∗ (t) g(t).
The Fourier modes in this context are also called Floquet channels.24) . 40].n + µx n0 (δn.20) n 1 = T dt einΩt φα (t) .23) (2. We replace the operators a and a+ by their expectation values (see Appendix A) and obtain a driven system with a timedependent Hamiltonian.n . √ z =  n0 exp(iΩt) . can be applied accordingly [39.18) By this introduction of a Hilbert space structure for the time dependence.n +1 + δn.13) is equivalent to its representation as a Fourier series.8 Driven quantum systems and Floquet theory The elements of this composite Hilbert space. φα (t) = cα.22) Under this condition. Semiclassical interpretation of the Floquet states A timedependent Hamiltonian is usually obtained from a timeindependent theory by substituting a part of the system by its classical limit [25]. like e.21) We assume in the semiclassical limit that the state of the laser ﬁeld is a coherent one (see Appendix A) and that it possesses a very high mean photon number.n −1 ) + Ω(n0 − n)δn. The corresponding Floquet Hamiltonian reads √ H = HS + 2µx n0 cos(Ωt) + Ωn0 − i ∂t . We restrict ourselves to the case of a linearly coupled driving ﬁeld with cosine shape. (2. tφ ≡ φ(t) = φ(t + T ) . the description of the system can be simpliﬁed in two ways: 1.g. The methods known for the computation of energy eigenstates of a timeindependent Hamiltonian. T 0 (2. (2. This allows for a semiclassical interpretation of the vectors n T and the Floquet states [40]. which couples via dipole interaction to a singlemode laser with frequency Ω.n . A system S.” are T periodic states.19) (2. decomposed into the basis set {n T }. n0 1.. (2. (2. perturbation theory.n = HS δn. we formally traced back the computation of Floquet states to the computation of eigenstates of a timeindependent Hamiltonian with an additional degree of freedom. √ Hn. can be described by the Hamiltonian [44] H = HS + µx(a + a+ ) + Ωa+ a. written in “time representation. The decomposition of a state φ(t) into the set of basis functions (2.n e−inΩt cα.
We decompose the Hamiltonian H.e. H(t) φ(t) = φ(t) .. In the following. (2.25) If the state of the laser ﬁeld is the highly exited coherent state (2.3 Properties of Floquet states Equivalent representations Assuming that φ(t) is an eigenvector of H(t) with eigenvalue . the state obeys φ(n) (t) = einΩt φ(t) H(t) φ(n) (t) = (H(t) + ∂t ) einΩt φ(t) = ( + n Ω) einΩt φ(t) . (2.28) (2.30) The respective solutions of the Schr¨dinger equation. The prefactors n and n + 1 √ −1/2 in this limit become n0 + O(n0 ). They all describe the same physical state. whose quasienergies lie within a single Brillouin zone ωBZ ≤ < (ωBZ + Ω). into the number states (A.n −1 + Ωn δn.26) (2.31) (2.24) agrees—besides a shift in the index—with the Hamiltonian (2. Therefore. The Floquet Hamiltonian (2.√ get we √ relevant contributions only for n ≈ n0 1. there exists a class of equivalent Floquet states whose quasienergies diﬀer only by integer multiples of Ω. but with eigenvalue (n) = + n Ω.22).α .17). it is suﬃcient to take only those Floquet states into account. Thus. Therefore the basis states n T allow for an interpretation as the semiclassical limit of the number states of the laser ﬁeld and the Floquet states as the semiclassical limit of the dressed states.3 Properties of Floquet states 9 2.27) (2.32) are identical. whose eigenfunctions are the socalled dressed states. (2. i. (2. o ψ (n) (t) = e−i( +n = ψ(t) Ω)t/ φ(n) (t) (2. They are orthonormalized with respect to the inner product (2. φα φα = δα.2. we denote by {φα (t) } a complete set of Floquet states with corresponding quasienergies { α }.n + µx n + 1 δn.n = HS δn.n +1 + n δn.25).33) . a Floquet state.10) of the laser mode to obtain √ √ Hn. 2.29) This means that φ(n) (t) is also an eigenvector of the Floquet Hamiltonian H(t).n .
since they vary with time.42) Thus the nth Floquet channel gives a contribution α +n Ω. Mean energy Due to the Brillouinzone structure (2. Thus.39) do not either.40) (2.n cα.36) φα φα (n) = δα. quasienergies do not allow for global ordering.20). caution is appropriate: The orthonormalization on R is in general only valid for equal times and is in particular not valid for the Fourier components (2. 23–25] Eα = = 1 T α T dt Eα (t) 0 (2.34) The Fourier coeﬃcients read 1 κn = T = T dt einΩt φα (t )φα (t ) 0 (2. weighted by the squared modulus cα. .0 . (2. (2. A quantity that is deﬁned on the full real axis and therefore does allow for a complete ordering is the mean energy [12. ∂t which results from averaging over one period of the driving.10 Driven quantum systems and Floquet theory Orthonormalization on R The inner product of two nonequivalent Floquet states on R obeys the T periodicity of the Floquet states and can be written as a Fourier series.n cα.35) (2. Here however.41) +i φα  ∂ φα .17). By use of the Fourier representation (2. we get φα (t)φα (t) = δα.n of the corresponding Fourier coeﬃcient. where the time integration has been expressed by the inner product (2.n (2.α δn.α . The instantaneous energies Eα (t) = ψα (t)H(t)ψα (t) = φα (t)H(t)φα (t) (2.17) on R ⊗ T we obtain orthonormalization with respect to the inner product on R at equal times.30).19) we obtain Eα = n ( α + n Ω) cα.37) This means that from the orthonormalization of the Floquet states with respect to the inner product (2.38) (2. φα (t)φα (t) = n κn e−inΩt .
25]. we used the T periodicity of the Floquet states and their completeness and orthogonality at equal times. e−in α T / φα (0) φα (0) (2.46) as this expression obviously solves the Schr¨dinger equation and the initial condition o (2. t ).50) as can easily be seen by inserting the Floquetstate representation (2. which is a solution of the Schr¨dinger equation.43) (2. t)φα (t) = e−i α T / φα (t) . o i ∂ U (t. 0) is indispensable for the investigation of the longtime dynamics of driven quantum systems [23. 0) deﬁnes a quantum map for the propagation over a full period of the driving. t ).44) A formal integration yields U (t.49) U (nT. The propagator U (T. (2. To obtain the last line. 2. t). Expressed in terms of the Floquet states. The propagator U (T. 0) = α e−i α T / φα (0) φα (0). (2. ∂t U (t. Due to the time dependence of the Hamiltonian. 0) = α = [U (T.4 The propagator 11 2.5 Numerical computation of Floquet states Among the methods for the computation of Floquet states of bounded systems. t ) = T exp − i t t dt H(t ) . (2. U (T. The Floquet states at time t are instantaneous eigenstates of the oneperiod propagator U (t + T.44) is ensured by the completeness of the Floquet states. (2.47) (2.45) where T denotes time ordering. U (t.48) (2. not only on their diﬀerence. 0)]n . t) = 1. the propagator reads U (t.2. U (t + T. t ) = α e−i α (t−t )/ φα (t) φα (t ). t ) depends explicitly on both times t and t .46) of the propagator. we essentially discern two classes [37]: The ﬁrst class consists of methods based directly . t ) = H(t) U (t.4 The propagator The time evolution of a quantum system can be written by use of a unitary operator U (t.
.5. A further eﬃcient method for the computation of eigenvectors of the tridiagonal matrix (2. As a basis set for the Hilbert space R. 0 0 ··· 0 0 ··· H1 0 ··· H0 − Ω H1 ··· H1 H0 − 2 Ω · · · . . · · · H0 + 2 Ω H1 ··· H1 H0 + Ω 0 H1 H = ··· ··· 0 0 ··· 0 0 .52) The eigenvectors of (2. Accordingly. Due to the Brillouinzone like structure. .45].51) where we have introduced a factor 2 for ease of notation. 0). however.10) of the Floquet Hamiltonian. .51) decomposed into the basis {n T } reads Hn. . They can be generalized straightforwardly.12 Driven quantum systems and Floquet theory on the solution of the eigenvalue equation (2. (2. which itself has been decomposed into the eigenfunctions of the harmonic oscillator (see Appendix A). .50) of the unitary operator U (T. one commonly uses M eigenstates of the undriven Hamiltonian H0 .50) for the propagator.5. it is advantageous to diagonalize the Hermitian operator 1 + U (T.n −1 ). . (2. In numerical calculations. . . . it is suﬃcient to compute all eigenvectors whose eigenvalues lie in an interval of size Ω. . . . (2. . 0 H1 H0 H1 0 .1 Floquetmatrix methods The Floquet Hamiltonian for (2. . . . followed by the solution of the eigenvalue equation (2. . . we treat systems subject to a cosineshaped driving. for N Floquet channels the dimension of the Floquet matrix is N M and the computational eﬀort for the matrix diagonalization is proportional to (N M )3 .53) 2. or in matrix notation.2 Propagator methods The quasienergies and the Floquet states at time t = 0 can be extracted from the oneperiod propagator by use of the eigenvalue equation (2. . we elucidate the numerical methods for the case of a Hamiltonian of the structure H(t) = H0 + 2H1 cos(Ωt).54) 1 − U (T. as the decomposition into {n T } corresponds to Fourier representation. . We shall not apply this method. . 0) V =i . 0) . (2. . .n of the Floquet states.n = (H0 + n Ω)δn. . . A second class of methods starts with the computation of the Floquet propagator U (T. . 2.. In the present work. .n +1 + δn.n + H1 (δn.53) are the Fourier components cα. . Thus. .53) are matrix continued fractions [25. 0).. . .
the Schr¨dinger is extended by a second time coordinate to read o i ∂ ψ(t. obeys i ∂ ∂ ∂ ψ(t.1).58) of the extended Schr¨dinger equation (2. They are obtained by propagating the φα (0) over one period of the driving. The (t. it is necessary to know the Floquet states’ Fourier coeﬃcients cα.51) is derived from the (t. t) . t0 ) = T 0U (t − t0 )t T t =t (2.55) The time t is treated formally like an additional canonical coordinate of a timeindependent problem. The propagation can be performed in various ways. (2. It is straightforward to show that the corresponding eigenvalues read cot( α T /2 ). The cα. t ) ∂t ∂t ∂t = H(t)ψ(t. which yields ψα (t) = exp(−i α t/ )φα (t) .13). T ]. In the following. V possesses the same eigenvectors as U (T. o It reads U (t. 0).55) one can extract the “true” propagator.56) t =t (2. according to their deﬁnition (2. we sketch the methods implemented in this work. namely the Floquet states φα (0) . t ) = ∂t H(t ) − i ∂ ∂t ψ(t. 46.20).55). t) is a solution of the “true” Schr¨dinger equation (2. t) = i + ψ(t. In an analogous o way. 47]. There. We postulate T periodic boundary conditions in t . 0). where the o initial condition is the unit matrix. for t in the range [0.n result from Fourier decomposition. t )formalism [38. t ) . t )formalism A very eﬃcient numerical method for the computation of the propagator for a Hamiltonian of the form (2. An extension of this method to other shapes of driving is rather easy. Direct integration of the Schr¨dinger equation o The most simple method for the computation of the propagator is the direct integration of the Schr¨dinger equation by use of a RungeKutta routine.5 Numerical computation of Floquet states 13 Being a function of U (T. It emerges that the numerical eﬀort for the propagation is proportional to N M 3 . (2. which enables decomposition into the basis set (2.n . Being a solution of (2. Therefore.57) Thus ψ(t. ψ(t.59) .2. however with a much larger prefactor compared to the diagonalization of the Floquet matrix. computing the Floquet states by direct integration is wellsuited if a large number of Floquet channels is required. t ) . on the cut t = t where ∂t /∂t = 1. For the computation of the mean energies and to determine the coeﬃcients of the master equation for the dissipative dynamics (see next chapter). from the propagator U (t − t0 ) = e−iH(t−t0 )/ (2.
45). already a few Floquet channels are suﬃcient to obtain numerical convergence [47]. the ﬁrst term of the Taylor expansion of the timeordered exponential (2. (2. t0 ) = i + ∂t ∂t ∂t = H(t)U (t. (ν) 1 = ν! − iτ 0Hν n T . For larger N .n (τ ).n (τ ) U0. In the special case N = 1 we obtain U (t + τ.66) For a suﬃciently small time step τ . t) = 1 − iH(t + τ )τ / . (2. T 0U (t − t0 )t T (2.63) Here the time ordering. (2. the time ordering results in a more complicated expression. t0 ) = n T 01R⊗T n T einΩt0 = 1R . is intrinsic. Typically. By Taylor expansion of the extended propagator U one obtains for the time step from t to t + τ U (t + τ.65) = with (ν) U0.45). in the sum over n all terms with n > N vanish. which we have to consider explicitly in (2. t0 ).61) and on the other hand solves the Schr¨dinger equation.64) (2. it is possible to truncate the sum over ν after N + 1 terms.62) t =t (2.60) This is so because on the one hand it fulﬁlls the initial condition U (t0 . t) = n ∞ einΩ(t+τ ) T 0U (τ )n e n inΩ(t+τ ) ν=0 ν T T (2.14 Driven quantum systems and Floquet theory = n T 0U (t − t0 )n T einΩt . . o i ∂ ∂ ∂ U (t. Due to the tridiagonal structure of H.
Caldeira and Leggett rediscovered the systembath model in the context of dissipative tunneling [11] and. quantization results in unphysical properties. we introduce the systembath model and derive a Markovian master equation for the reduced density operator for the case of a static central system. is based on the coupling of the conservative system to external degrees of freedom. we couple the system bilinearly to a bath of noninteracting harmonic oscillators with masses mν . Later.g. Zwanzig generalized this concept within ı the framework of classical stochastic processes to arbitrary potentials and derived a Markovian master equation for the dynamics of the dissipative system by the socalled projector formalism [20]. in a pathintegral formulation. Strong systembath correlations result in interesting eﬀects. beyond a weakcoupling limit. 19. it is desirable to to treat the inﬂuence of the bath in perturbation theory.1 The systembath model To achieve a microscopic model of dissipation. eliminated the bath exactly [21. the pathintegral approach requires to resort to extensive and sophisticated numerics. master equations for quantum systems [50–52] were derived and applied in laser physics [50] and to nuclear magnetic resonance and electronspin resonance. frequencies ων . with the coupling strength cν [11. 59]. or doesn’t handle the uncertainty relation properly [49].53]. consistent with the fundamental laws of quantum mechanics.. Thus. This enabled the investigation of dissipative quantum systems. among them most prominently the algebraic decay of correlation functions at zero temperature [54. such as MonteCarlo calculations [56–58]. Although an extension of the Lagrange formalism to this model of dissipation is possible [48]. for the description of nonlinear systems subject to weak dissipation. 55]. leading to a Markovian master equation for the density operator [50–52]. The total Hamiltonian of system and bath is then given by H = HS + HSB + HB . a timedependent mass. and coordinates xν . By similar approaches. as soon as nonlinear forces come into play. momenta pν . The most successful approach to dissipation in quantum mechanics. However. dissipation can be introduced phenomenologically just by adding a velocityproportional friction force. 3. Probably the ﬁrst proof that such a systembath scheme results in dissipative quantum mechanics was given by Magalinski˘ [19] for a harmonic oscillator.3 Quantum dissipation and Markov approximation Within the framework of classical mechanics. with their own shortcomings. e. (3.1) . In this chapter.
For the time evolution we choose an initial condition of the FeynmanVernon type: at t = t0 . Although this choice is somewhat artiﬁcial. one can eliminate the bath variables to get an exact. i. the canonical ensemble of the whole system including the coupling [60]. which depends only on the position x of the system.e. 3. it is favorable due to its technical simplicity. like e. The second term in HSB . This results in a dissipative diﬀerential equation for the Heisenberg position operator of the system.16 Quantum dissipation and Markov approximation where HS denotes the Hamiltonian of the central system and HB = ν 2 pν 1 2 + m ν ω ν x2 .1) we derive the Heisenberg equations of motion for the system and the bath operators and solve the latter formally. Other initial conditions. below we will deal with driven systems where specifying a more sophisticated preparation is not meaningful without specifying an onset of the driving. it oﬀers a possibility for interpretations. are more realistic.6) .4) W (t0 ) = (t0 ) ⊗ tr e−HB /kB T where is the density operator of the system and kB T denotes Boltzmann’s constant times temperature. 53. (3.g. ν 2mν 2 (3. The elimination can be performed in two ways. The Heisenberg equations of motion for the position operators of the system and of the bath oscillators read ¨ x+ 1 1 cν V (x) = c ν xν − x .2 Quantum Langevin equation From the systembath Hamiltonian (3. this quantum Langevin equation cannot be solved exactly and thus is of limited practical use. Due to the bilinearity of the bath and its coupling to the system. which are the subjects of the following sections.3) HSB = −x c ν xν + x 2 ν ν c2 ν . However. the bath is not correlated to the system and canonically distributed with respect to the free bath Hamiltonian. 2 m m ν mν ω ν cν 2 x. x ν + ω ν xν = ¨ mν (3.5) (3. subject to dissipation. 59]. which is driven by an operatorvalued stochastic force. closed integrodiﬀerential equation for the dynamics of the central system. the density operator W of system plus bath reads e−HB /kB T . 2 2mν ων describe the heat bath and its coupling to the system. serves to cancel a renormalization of the potential due to the coupling [49.2) (3.. Although in general.
In the limit of zero temperature. 2 c2 ων ν = cos ων (t − t ). inserting into (3. (3. the Gaussian property holds. . coth 2mν ων 2kB T ν (3.. we have made use of the equilibrium expectation values ων 1 1 2 p ν pν = m ν ω ν x ν xν = coth 2 2mν 4 ων 2kB T δνν (3.14) which also marks the time scale below which correlations between system and bath are relevant.2 Quantum Langevin equation Equation (3.1) is bilinear in the bath coordinates xν . The inﬂuence of the ﬂuctuating force on the system is fully characterized by its symmetric autocorrelation function.8) gives rise to an initial slip due to the sudden coupling of the system and the bath at time t0 [19.10) The last term in (3.11) (3. τB diverges and these correlations play a dominant role [54. i. 55].5) results in the socalled quantum Langevin equation [61–64] t x(t) + ¨ t0 dt γ(t − t )x(t ) + ˙ 1 1 V (x(t)) = ξ(t) − γ(t)x(t0 ) m m (3.6) is easily integrated to yield the formal solution xν (t) = xν (t0 ) cos ω(t − t0 ) + + cν mν ω ν t t0 17 pν (t0 ) sin ων (t − t0 ) mν ω ν (3.3. The correlation function K(τ ) decays within a time τB = /kB T. we can express moments and correlations of higher order by products of K’s.8) with the damping kernel γ(t) = 1 m c2 ν cos ων (t − t0 ) 2 mν ω ν (3. mν ω ν (3.12) To obtain the last line.e. the noise kernel K(t − t ) = 1 ξ(t)ξ(t ) + ξ(t )ξ(t) . As the systembath Hamiltonian (3. After integration by parts.7) dt sin ων (t − t ) x(t ). It will be omitted in the following as we will not study preparation eﬀects within this framework. 64].9) ν and the operatorvalued ﬂuctuating force ξ(t) = ν cν xν (t0 ) cos ων (t − t0 ) + pν (t0 ) sin ων (t − t0 ) .13) for the bath operators in the canonical ensemble. 61.
8) drops to zero. 3. where trB denotes the trace over the bath variables.15) 2mν ων ν In a continuum limit for the heat bath we assume I(ω) to be a smooth function. (3.19) which deﬁnes the Drude model. The assumption of an increasing spectral density for arbitrarily high frequencies. 63]. 2 ω 2 + ωD (3. The damping and the noise kernel can be expressed by the spectral function to read ∞ I(ω) 2 dω cos ωt πm 0 ω ω 1 ∞ dω I(ω) coth K(t) = π 0 2kB T γ(t) = (3. which results in an equation of motion for the reduced density operator = trB W of the central system subject to dissipation. This corresponds to the Ohmic spectral density I(ω) = mγω. (3. Eq. which in Fourier representation reads 1 ω K(ω) = m ωγ(ω) coth 2 2kB T .16) cos ωt. As a prototypical model for damping. is not only somewhat artiﬁcial. we use the Ohmic friction kernel γ(t) = 2γδ(t). We regularize them. (3. but also results in divergent integrals. 62. An Ohmic spectral density is often used as an approximation to a more complicated one and therefore in literature sometimes appears as “ﬁrst Markov approximation” [50]. The cutoﬀ frequency ωD introduces a short but ﬁnite memory τD = 1/ωD for the friction.18 Quantum dissipation and Markov approximation At this point it is convenient to introduce the spectral density of the systembath coupling c2 ν I(ω) = π δ(ω − ων ). (3.3 Inﬂuence functional Despite the fact that the quantum Langevin equation (3. For an exact elimination of the .18) reads K(ω) = mγ(ω)kB T and the quantum Langevin equation becomes in the longtime limit formally equivalent to the corresponding classical Langevin equation [49.18) In the classical limit kB T ω. A more useful approach is the elimination of the heat bath in the equation of motion for the full density operator W . by a cutoﬀ in the spectral density. its practical use is limited to the very rare cases where it can be integrated directly. where the memory of the friction in (3. however.8) appears quite simple.17) Both are not independent of each other since they obey the socalled second ﬂuctuationdissipation relation [49]. I(ω) = mγω 2 ωD . if required.
We start with the time evolution of the full density matrix. t t Re φFV [x. t0 ) describes the timeevolution of the dissipative system.25) . t. xf . x0 . 49. 65] x(t)=xf x(t)=xf Dx x(t0 )=x0 Dx exp U (xf . xf . x ] . x0 . x(t0 )=x0 (3. x] .24) The variable x is a shorthand for all bath coordinates xν and Dx denotes path integration over all of them. We insert the initial condition (3. (3. t0 ). t0 )W (x0 . t. xf = dx0 dx0 dx0 dx0 U (xf . x0 . t. t0 ). t0 ) ×U ∗ (xf . x0 . t.27) Dx x (t0 )=x0 Dx exp i i S[x] − S[x ] (3. ˙ 2 dt 1 cν mν 2 xν (t )2 − mν ων xν (t ) − ˙ x(t ) 2 2 2 mν ω ν 2 (3. xf . x(t)=xf x(t0 )=x0 x (t)=xf (3. the pathintegral formulation of quantum mechanics has proved to be more convenient than operator notation [11. (3.23) (3. x0 . t0 ) of the system plus the bath is given by the path integral expression [11. respectively. xf . t) = J(xf . t. x0 . x0 . 53]. x ]. x0 . we obtain [11. t0 ) = dx0 dx0 J(xf . After tracing out the bath variables by integrating over all the bath coordinates xf . t) ≡ xf . x0 . W (t) = e−iH(t−t0 )/ W (t0 )eiH(t−t0 )/ . x ] = t0 dt t0 dt x(t ) − x (t ) K(t − t ) x(t ) − x (t ) .29) . x0 .22) (3. 21. xf . xf . xf . x0 .20) The propagator U (x.3.4) and evaluate the path integral over the bath variables. xf W (t)xf . The entire inﬂuence of the bath is subsumed in the socalled inﬂuence functional [21] φFV [x. which in position representation reads W (xf . t0 ) = i i S[x] + SB [x. x0 . t0 ) (x0 . x0 . The propagator J(xf . (3. The actions t S[x] = t0 dt t m 2 x (t ) − V (x(t )) .26) SB [x.21) (3. t. x0 . x0 . xf . x0 . t. x0 . x. x] = ν t0 correspond to the Hamiltonian HS of the central system and HB + HSB for the bath plus systembath coupling. xf . 49] (xf . x0 .3 Inﬂuence functional 19 heat bath.28) 1 × exp − φFV [x. xf .
t. 66]. (3. t0 ). one depending only on x. Thus.20 Quantum dissipation and Markov approximation Im φFV [x. It can be separated into two parts.1). The last term of the imaginary part gives the initial slip. The steps to introduce this Markov approximation are usually performed in operator notation. xf . In standard perturbation theory for path integrals [65. x ] 1 ≈ 1 − φFV [x. starting from the full systembath Hamiltonian (3. whereas the imaginary part gives rise to friction [49]. ∆/ .31) The small parameter in this approximation is the eﬀective coupling strength γ. They are easily identiﬁed as the propagator for the Schr¨dinger equation of the pure system and its complex conjugate. Here.28).8). x0 .30).30). without memory. one can here distinguish more clearly between the inﬂuence of friction and noise.30) dt x(t ) − x (t ) γ(t ) x(t0 ) + x (t0 ) .28) [49]. i. First. t. x0 .34) . but has some advantages. x ]. The present derivation requires essentially the same approximations as the standard projection technique approach. A second beneﬁt is the exact cancellation of the potential renormalization in (3. we give a derivation from the pathintegral expression (3. the friction part in the Markovian master equation becomes exact. the exponent of the inﬂuence functional is approximated by a Taylor series. because each of them is easily identiﬁed in the path integral expression (3. we have integrated Im φFV by parts. canceling the potential renormalization in (3.4 Markovian master equation By perturbation theory for the propagator (3. The propagator for the density matrix is at order zero in the perturbation given by the ﬁrst line of (3. which means that γ has to be the smallest frequency scale in the problem.28) up to lowest nontrivial order in the systembath coupling. known from Eq.32) (3. And last but not least. the other only on x . t. The real part of the inﬂuence functional describes the noise. γ γ 1/τB = kB T / .28).26).33) where τB is the correlation time of the bath and ∆ denotes any energy diﬀerence in the spectrum of the conservative problem. t0 ) U0 (xf . t0 ) = U0 (xf . 3. x0 . (3.e. and is omitted in the following. (3. one can show that for the case of an Ohmic spectral density. 1 exp − φFV [x. (3. x0 . x ] = − m 2 m − 2 t t dt t0 t t0 t0 dt x(t ) − x (t ) γ(t − t ) x(t ) + x (t ) ˙ ˙ (3. To obtain (3. we derive a master equation of Markovian type. o ∗ J0 (xf ..
t ). t. x2 . t − τ .27) and (3. x2 .36) We insert into (3. x0 . t) = − dx0 dx0 J0 (xf . t ) γ(t − t ) (x2 + x2 ) (x2 . x1 . xf . we will solve the master equation in energy basis. t − τ ) ∗ ×(x2 + x2 ) U0 (x2 . x2 . x2 . t − τ . t . x. where we have assumed that path integration commutes with the integrals over t and t . t) + im 2 t dτ γ(τ ) t0 ∗ dx1 dx1 dx2 dx2 (xf − xf ) U0 (xf . x.30) only yield contributions at times t and t . t) U0 (x2 . t ) (x1 − x1 ) (3.. x2 . or in Wigner representation.29) and (3. (x2 . t) ∗ dx1 dx1 dx2 dx2 (xf − xf ) U0 (xf . By use of (3. the inﬂuence functionals (3. t − τ ) dτ K(τ ) t0 ∗ ×(x2 − x2 ) U0 (x2 . in Floquet basis. x0 . x2 . t )K(t − t ) (x2 − x2 ) (x2 . t . an operator notation is more convenient than a position representation. t. t ) = dx0 dx0 J0 (x2 . t) i = − H(xf ) − H(xf ) − 1 t (xf . 66] to get (xf . x1 . x .35). t. t). In all these cases. i. This master equation is Markovian since ˙(t) depends only on (t). Deriving from (3. t0 ) (x0 . x2 . xf . not on the history of . t. x0 . x2 . t ) (x1 − x1 ) ×J0 (x1 . xf . t) (x. t ) im + 2 t t dt t0 t0 dt dx1 dx1 dx2 dx2 J0 (xf . t − τ . x2 . t. x2 . xf .e. diﬀerentiate with respect to t.4 Markovian master equation 21 In ﬁrst order of perturbation (which is already second order in the coupling constants cν ). t − τ )U0 (xf . x . and obtain the master equation ˙ (xf . x2 . t) (x. x .37) where the free propagator J0 for the density matrix has been substituted by the propagator U0 of the Schr¨dinger equation and the integration variable t by τ = o t − t . x0 . x1 . xf .37) .3. at equal times. x0 . t − τ . x1 . t) U0 (x2 . In the following chapters. t) (x0 . t. xf . x1 . Thus we can dissect the path integral into an explicit integration over x1 = x(t ) and x2 = x(t ) and free time evolution [65. (3. x1 . ˙ ˙ (3. respectively. t0 ) 1 t t dt t0 t0 dt dx1 dx1 dx2 dx2 J0 (xf . t. x0 . t). x . t − τ )U0 (xf . we can express (t ) in zeroth order of the perturbation by (t).35) ˙ ˙ ×J0 (x1 .34). x2 . t .
22
Quantum dissipation and Markov approximation
the representationfree form is straightforward and yields d i 1 ∞ = − [HS , ] − dτ K(τ ) [x, [xH (t − τ, t), ]] dt 0 ∞ i dτ γ(τ ) [x, [pH (t − τ, t), ]+ ], + 2 0
(3.38)
with the anticommutator [A, B]+ = AB + BA. The Heisenberg position and momentum operators xH and pH are deﬁned according to
† OH (t, t ) = U0 (t, t ) O U0 (t, t ),
(3.39)
where U0 (t, t ) = exp(−iHS (t − t )/ ) denotes the propagator of the conservative system. We have assumed further that the integration kernel K(τ ) is practically zero for τ > τB [67] and extended the upper integration limit in (3.38) to inﬁnity. This implicitly moved the preparation time t0 → −∞, thus the master equation (3.38) describes only the system dynamics suﬃciently close to equilibrium. For an Ohmic spectral density γ(τ ) = 2γδ(τ ), the integration in the second line of (3.38) can be evaluated and we obtain the Markovian master equation d i = − [HS , ] + Lfriction + L0 noise dt . (3.40)
The commutator in (3.40) gives the coherent dynamics, whereas the superoperators Lfriction = − L0 noise iγ [x, [p, ]+ ], 2 1 = − [x, [Q, ]], (3.41) (3.42)
describe the inﬂuence of the the bath: friction and noise. The operator
∞
Q=
0
dτ K(τ ) xH (t − τ, t),
(3.43)
is qualitatively the Heisenberg position operator xH of the system in Fourier representation. Therefore L0 depends on the conservative dynamics (superscript 0 ), noise thus on the energy spectrum of the central system. Note that Q is time independent, since for a static Hamiltonian xH (t − τ, t) = xH (−τ ). The Markovian master equation (3.40) together with (3.41) and (3.42), does not exhibit Lindblad form (B.3), thus the positivity of the density operator is not guaranteed for all possible initial states. The violation of positivity due to a master equation in this case, however, is a transient eﬀect which only arises for preparations far from equilibrium [68–71], where the conditions under which the master equation has been derived, are not fulﬁlled. (See Appendix B.1 for a more detailed discussion).
4
Driving and dissipation: FloquetMarkov theory
For a dissipative quantum system subject to external driving, even a partially analytical solution within the pathintegral approach is feasible only for the very simplest systems, in particular, for the periodically driven, damped harmonic oscillator [28], or for driven dissipative twolevel systems [72, 73]. Thus, for the description of strongly driven systems subject to weak dissipation, it is desirable to combine a Markovian approach to quantum dissipation, leading to a master equation for the density operator, with the Floquet formalism that allows to treat timeperiodic forces of arbitrary strength and frequency. While the Floquet formalism amounts essentially to using an optimal representation and is exact [23], the simpliﬁcation brought about by the Markovian description is achieved only at the expense of accuracy. Here, a subtle technical diﬃculty lies in the fact that the truncation of the longtime memory introduced by the bath and the inclusion of the driving do not commute: As pointed out in Refs. [26, 27], the result of the Markov approximation depends on whether it is made with respect to the eigenenergy spectrum of the central system without the driving, or with respect to the quasienergy spectrum obtained from the Floquet solution of the driven system. In the second case it cannot be treated as a system with proper eigenstates and eigenenergies. Figure 4.1 depicts the two diﬀerent possibilities for including driving and dissipation to the description of a quantum system. Both approaches yield a Markovian master equation, but diﬀer quantitatively. We will investigate this diﬀerence in detail for the case of a parametrically driven harmonic oscillator in Chapter 5. A Floquet theory for dissipative driven systems based on the energy spectrum has been worked out and applied to intenseﬁeld excitations of atoms in Refs. [37,74]; a quasienergy spectrum approach has been implemented in recent work on driven Rydberg atoms [22, 75] and coherent destruction of tunneling [76–78].
4.1
Simple inclusion of the driving
A simple Markovian approach to dissipative driven quantum systems results directly from the master equation for the undriven system: We replace in (3.40) the static Hamiltonian HS by the timedependent Hamiltonian HS (t) = H0 + HF (t) which yields i d = − [HS (t), ] + Lfriction + L0 noise dt . (4.2) (4.1)
24
Driving and dissipation: FloquetMarkov theory
S Markov (energy spectrum)
Floquet theory −− − − − − − − − − − − − − − − − − −→ for Schr¨dinger equation o
S+D Markov (quasienergy spectrum)
S+B
Floquet theory −− − − − − − − − − − − − − −→ for master equation
S+D+B
Figure 4.1: Successive inclusion of the driving (D) and the inﬂuence of a heat bath (B) to the description of a quantum system (S). The horizontal arrows denote exact Floquet treatment, whereas the vertical arrows mark an approximate step, namely the truncation of the longtime memory. The result depends on the route taken.
Here, the driving enters only the coherent part of the master equation, whereas L0 noise has been derived from the undriven Hamiltonian H0 . Thus, we refer to this approach as the Markovian approach with respect to the unperturbed spectrum. For a periodically timedependent driving, HF (t) = HF (t + T ), the master equation (4.2) allows for a Floquet treatment [37].
4.2
An improved Markovian master equation
We pointed out in Section 3.4, that the coherent dynamics of the central system plays an important role in the derivation of the Markovian master equation (3.40). This means that for a driven system the Markovian master equation depends on whether the driving is considered in its derivation or not. To obtain an improved master equation whose dissipative kernel accounts for the inﬂuence of the driving, we start anew from the full systembath Hamiltonian including the driving. Performing the same steps as in the preceeding chapter, but for an explicitly timedependent Hamiltonian HS (t), we obtain the Markovian master equation i d = − [HS (t), ] + (Lfriction + Lnoise ) . (4.3) dt Friction and noise are described by the superoperators Lfriction = − Lnoise iγ [x, [p, ]+ ], 2 1 = − [x, [Q(t), ]]. (4.4) (4.5)
Whereas Lfriction is the same for both Markovian approaches, Lnoise has acquired a time dependence which stems from the operator
∞
Q(t) =
0
dτ K(τ ) xH (t − τ, t),
(4.6)
.4.3).11) = Qαβ (t) ≡ φα (t)Q(t)φβ (t) = einΩt Qαβ. Therefore Lnoise is time dependent and does—in contrast to L0 —not depend on the energy spectrum of the undriven noise system. we are able to make use of the Floquet theorem and expand the reduced density operator into the timeperiodic Floquet states φα (t) of the isolated driven system. einΩt Pαβ.13) φα (t)xφβ (−n) (t) .3 Decomposition into Floquet basis So far. but on the quasienergy spectrum of the driven system. t ) x U (t. The inﬂuence of a driving force on Lnoise will be studied in detail for the case of a parametrically driven harmonic oscillator in Chapter 5.10) (4. They all are T periodic and can be expressed as a Fourier series.8) αβ = φα (t) φβ (t) is derived from the basisindependent improved master equation (4. t ) = U † (t.n .n = = 1 T T dt e−inΩt φα (t)xφβ (t) 0 (4.n and Qαβ. we will express the Fourier coeﬃcients Pαβ. A master equation for the matrix elements (4. Xαβ (t) ≡ Pαβ (t) ≡ φα (t)xφβ (t) φα (t)pφβ (t) = n einΩt Xαβ. not only on their diﬀerence. t ) 25 (4. Next. we did not specify the time dependence of the system Hamiltonian in the derivation of the master equation. Since the role of the eigenenergies is now taken over by the quasienergies. we need to know the matrix elements of the operators x. 4.7) is the Heisenberg position operator of the driven system which depends explicitly on both times. They form a welladapted basis for the case of weak dissipation. we refer to this master equation as the Markovian approach with respect to the quasienergy spectrum.n in terms of Xαβ. n The Fourier coeﬃcients of the position matrix elements read Xαβ.3).3.n .n .3 Decomposition into Floquet basis where xH (t.9) (4.12) (4. t and t . By assuming a T periodic Hamiltonian. p and Q(t) in the Floquet basis.1 Matrix elements To decompose the master equation (4. n (4. 4.n .
n . By use of the Fourier representations (4.−n )Xαα .17) Xαβ. i i where H = H − i ∂/∂t denotes the Floquet Hamiltonian.14). 27. dτ mγ π ∞ dω ω coth 0 where we have inserted the spectral representation (3. the socalled Lamb shifts [50. (4. x] = [H. and have been neglected. with the thermal occupation number nth ( ) = For 1 e /kB T kB T .n .n = mγ ( 2 α − β + n Ω) coth α − +n Ω 2kB T β Xαβ.n − Nα β .18) The contributions of the principal part result in quasienergy shifts of the order γ. the momentum operator can be expressed by a commutator. Note that the coeﬃcients of this diﬀerential equation are periodic in time with the period of the driving. (4.14) p = [H.n = N ( α − α + n Ω). 75] ˙ αβ (t) = d φα (t) (t)φβ (t) dt i = − ( α − β ) αβ (t) + α β nn ei(n+n )Ωt (Nαα . We end up with Qαβ. The Fourier coeﬃcients of the timedependent matrix element Qαβ (t) read Qαβ.n (4.n Xβ α .n Xα β.n Xαβ .n αβ αβ Xβ β.26 Driving and dissipation: FloquetMarkov theory For a Hamiltonian of the form H = p2 /2m + V (x.n = φα pφβ (4.20) Nαβ. The τ integration is evaluated by using 0 dτ exp(iωτ ) = πδ(ω) + iP(1/ω).17) of the noise kernel and ∞ made use of (4. N ( ) approaches zero.n = = 0 1 T T ∞ dt e−inΩt 0 ∞ 0 dτ K(τ ) φα (t)xH (t − τ.21) . t). Thus we get m (−n) Pαβ. m m (4. where P denotes Cauchy’s principal part.12).n are given by mγ (4. we made use of the eigenvalue equation (2.9)–(4. i To obtain the last line.19) − Nβ α . 51].16) = ( α − β + n Ω)Xαβ. N ( ) = 2 nth ( ). x].3) the FloquetMarkov master equation [22.15) i m (4.29) for the Floquet states after inserting (4.11) we obtain from Eq.n .n + Nββ . −1 = 1 coth 2 2kB T −1 . t)φβ (t) ω 2kB T cos(ωτ )e−i( α − β +n Ω)τ / (4. The Nαβ.−n αβ Xβ α . (4.n .
25) by their time average. we average the likewise 2π/Ωperiodic coeﬃcients of the master equation (4. 75].−n Xβ α .n Xαβ .3. (4. as detailed in the next paragraph. Thus only the Lαβ.n + Nββ .n − δαα The timeindependence of its coeﬃcients reﬂects that the inﬂuence of the driving has been formally absorbed by decomposing into the Floquet basis. Full rotatingwave approximation In some cases one can even go one step further. we explore the conditions under which these coeﬃcients can be replaced by their time average. (4.22) by the ansatz αβ (t) = e−i( α − β )t/ σαβ (t).−n n (4.25) If dissipative eﬀects are only relevant on a time scale much longer than all ﬁnite times 2π /( α − β − α + β ). we are allowed to replace the coeﬃcients in (4.n . the coeﬃcients of the dissipative part are still time dependent and complicate the solution of the master equation.α β = (Nαα .n α . Moderate rotatingwave approximation Assuming that dissipative eﬀects are relevant only on a time scale much larger than the period 2π/Ω of the driving.2 Rotatingwave approximation We used the Floquet basis to formally eliminate a driving force of arbitrary strength from the coherent part of the master equation. Note that diagonal and oﬀdiagonal elements of the density matrix are not decoupled. It has also to be stressed that the rotatingwave approximation introduced here is less restrictive than the one in Refs.22) with the dissipative transition rates Lαβ. Here. We solve the coherent part of the master equation (4. This step eﬀectively amounts to a rotatingwave approximation (RWA). [22.−n Xα β.26) .4.3 Decomposition into Floquet basis 27 4.n Xβ α .α β which fulﬁll the fullRWA condition α − β = α − β .α β αβ (t).22) yields σαβ (t) = ˙ αβ ei( α − β − α + β )t/ Lαβ. Inserting into (4.n Xβ β.24) a transformation to the Heisenberg picture of the central system plus the driving. However. (4.n ) Xαα . − δββ Nβ β .19) over one period of the driving and end up with the equation of motion i ˙ αβ (t) = − ( α − β ) αβ (t) + αβ Lαβ.23) Nα α n β .α β σα β (t) (4.
σαα (t) = ˙ α Lαα.26) results in [22. since here. (4. Moreover. it converges in the longtime limit to an asymptotic state ∞ . This condition is.3) meets the conditions for a Floquet treatment. α = β.4 The dissipative quantum map and its numerical implementation The master equation (4.. 75] α=α. Therefore. ˙ The second equation results in an exponential decay of the oﬀdiagonal matrix elements. ﬁnd in Section 6. 0) in the conservative case—which describes the stroboscopic dissipative time evolution of the density operator. it is possible to deﬁne a dissipative quantum map G(T ) [25.29) σαβ (t) = Lαβ.22). Eq.28 Driving and dissipation: FloquetMarkov theory remain in (4. the oﬀdiagonal matrix elements play an important role for the asymptotic state.α α σα α (t). harmonic potentials with their equidistant (quasi) energy levels. Then the fullRWA condition (4. (4. Decomposing into the Floquet basis {φα (t) } yields the onecycle propagation of the density matrix elements αβ ((n + 1)T ) = αβ Gαβ..28) (4. however.25) yields two decoupled sets of equations for the diagonal and the oﬀdiagonal matrix elements. the “quantum attractor” which is the ﬁxed point of the dissipative quantum map G(T ). a full RWA seems to be appropriate. the quasienergy differences have no degeneracy at all. 74.3) generates a dynamical semigroup for the time evolution of the density operator.e. Therefore in the asymptotic limit. (4.30) is dissipative.31) . we have averaged over a longer time scale. much more restrictive than the one in the previous paragraph.αβ σαβ (t). β =β or α = β. the density matrix becomes diagonal in the Floquet basis.3 that even in a case where the dynamics of the system is fully chaotic and thus. (4. (4. α = β .g. We will.α β (T ) αβ (nT ). (nT ) = [G(T )]n (0). 4. Its coeﬃcients share the T periodicity of the driven system Hamiltonian HS (t). 80]— the analogue of the onecycle propagator U (T.27) Inserting into (4. i. 79]. Therefore the applicability of a full RWA is limited to very rare cases like.25) or (4. one can assume that for the case of a completely irregular spectrum where all quasienergies are eﬀectively random numbers [8. respectively. e.30) As the dynamics generated by (4. however.
The time evolution of the density operator results from iteration according to (4.31).32) over one period of the driving T to obtain the dissipative map G(T ).α β (t) (4.22).α β (t) = − ( α 29 − β )Gαβ. .4.32) follows straightforwardly from (4. i ˙ Gαβ.α β (t) δαα δββ + α β Lαβ. This form enables a numerical treatment of the master equation: We integrate (4.α β Gα β .4 The dissipative quantum map and its numerical implementation An equation of motion for the dissipative map.
30 .
Here. their quality can be reliably checked since in this system. is given in Section 5. interesting in its own right.4. Forming a convenient “laboratory animal” due to its simplicity and linearity. Therefore.7. which allows for a detailed analysis of the inﬂuence of the driving on the dissipative terms of the master equation. We give a brief review of the model. A reﬁned investigation within a basisindependent description.1 and 5.1) . and a comparison with the known quantum pathintegral solution [28] is possible.5 is devoted to a discussion of the asymptotics of the quantal solutions. A merely technical issue. Since this relation is particularly close in the case of linear systems. and the comparison to the pathintegral solution.5 The parametrically driven harmonic oscillator In this chapter we investigate the properties and the quality of the diﬀerent Markovian approaches to damped periodically driven quantum dynamics for a linear system where an exact pathintegral solution is still available: The parametrically driven. 5. the quasienergy spectrum is suﬃciently diﬀerent from the unperturbed energy spectrum [81. with their interrelations.82] (this feature is in contrast to the additively driven harmonic oscillator where the diﬀerence of two quasienergies does not depend on the driving parameters [83]).2.6 contains numerical results for a number of characteristic dynamical quantities as obtained for the alternative Markovian approaches. by switching to a phasespace representation such as the Wigner function. damped harmonic oscillator allows for a very transparent and wellcontrolled investigation of the diﬀerent approximation schemes introduced in Chapters 3 and 4. the Hamiltonian is given by HS (t) = 1 p2 + k(t)x2 . A summary of the various representations and levels of Markovian description. its classical dynamics. the solution of a FokkerPlanck equation by the method of characteristics. it is possible to elucidate the relationship of the quantal results to the corresponding classical FokkerPlanck dynamics. this provides an additional consistency check. Section 5. Section 5. is deferred to Appendix C. and its coherent quantum dynamics in Sections 5. such as the conservative and the hightemperature limits. a strong emphasis of this chapter is on the testing and thorough understanding of the available methods.1 The model and its classical dynamics For a particle with mass m moving in a harmonic potential with timedependent frequency. Moreover. In Section 5. 2m 2 (5. the parametrically driven harmonic oscillator still shows nontrivial behavior. is given in Section 5.3 we present the solution of the dissipative dynamics in FloquetMarkov description.
ϕ(t + T ) = ϕ(t).1 shows the zones of stable and 2 unstable motion. for the Mathieu oscillator. resulting in the sum rule ∞ (5.e. Ω = 2π/T.8) . the functions ϕ(t). 5. m (5. An initial phase of the driving can be taken into account by a proper time translation. Fig.1). we can formally remove the damping to get an undamped equation with a modiﬁed potential ¨ ξ + ω 2 (t) − γ 2 /4 ξ = 0. is given by ˙ ˙ W = ξ1 (t)ξ2 (t) − ξ1 (t)ξ2 (t) = 2i. we can apply the Floquet theorem for secondorder diﬀerential equations with timeperiodic coeﬃcients. The equation of motion for a classical particle with Ohmic (i. In stable regions µ is real. ∞ ϕ(t) = n=−∞ cn einΩt . µ by ϕ0 (t).. For given k(t). velocityproportional) dissipation in the potential given in (5. It asserts [42. Being periodic in time.5) The solution ξ2 (t) is related to ξ1 (t) by the fact that the coeﬃcients in the diﬀerential equation (5.1) reads x + γx + ˙ ¨ 1 k(t)x = 0. Figure 5. in the ω0 ε plane. Depending on its frequency and amplitude. We denote the limit γ → 0 of ϕ(t). On the border between a stable and an unstable region. 85] that Eq. The normalization of the cn is chosen such that the Wronskian W. There exist driving functions for which µ is complex so that one of the solutions ξi (t) becomes unstable (cf.4) are real. (5. ∗ ξ2 (t) = ξ1 (t). (5. which is a constant of the motion. µ becomes a multiple of Ω/2 and the solutions ξ1 (t) and ξ2 (t) are not linearly independent. A special case is the Mathieu oscillator.4) has two solutions of the form ξ1 (t) = eiµt ϕ(t). the driving can stabilize or destabilize the undriven oscillation.32 The parametrically driven harmonic oscillator where k(t) = mω 2 (t) is a periodic function with period T . the classical Floquet function ϕ(t) can be represented as a Fourier series. (5. n n=−∞ (5.3) By substituting x = ξ exp(−γt/2). on the level of the classical equations of motion.6) The Floquet index µ depends on the shape of the driving k(t) and is deﬁned only mod Ω. (5. ξi (t). where 2 ω 2 (t) = ω0 + ε cos Ωt. (5. µ0 . ξi0 (t).2) This is an experimentally important case in view of the fact that it describes the Paul trap [84]. respectively.7) c2 (µ + nΩ) = 1. ξi (t) and the Floquet index µ still depend on the damping γ.4) Already here.
5.3) with γ = 0 for the case of a Mathieu oscillator. In the shaded areas µ is complex and therefore one of the fundamental solutions (5. In terms of this function. which corresponds to stable solutions.9) 2 For constant frequency of the oscillator. µ becomes a multiple of Ω/2 and the motion is marginal stable. reads ∂G(t.5) is unstable. Returning to the original xcoordinate. t0 ).6) and (5. On the borderlines. respectively. the Floquet index 2 2 1/2 2 and the periodic function become µ = (ω0 − γ /4) and ϕ(t) = (ω0 − γ 2 /4)−1/2 . In the white areas the Floquet index µ is real.3) with initial conditions x(t0 ) = x0 and p(t0 ) = p0 .7).3) is constructed using Eqs. the solution of (5.1 The model and its classical dynamics 30 25 20 33 [Ω2 /4] 15 stable 2 ω0 10 5 0 5 0 5 10 15 unstable ε [Ω2 /2] Figure 5. this solution depends explicitly on the initial time t0 . (5. G(t.10) (5. which reproduces the results for a damped harmonic oscillator without driving. t0 ) = −x0 ∂t0 m Since the potential breaks continuous timetranslational invariance. k(t) = const = mω0 .3) read fi (t) = e−γt/2 ξi (t).1: Stability of equation (5. (5. 2. t0 ) p0 + G(t. we ﬁnd that the fundamental solutions of (5.n (5. t ) = e−γ(t−t )/2 ξ1 (t)ξ2 (t ) − ξ2 (t)ξ1 (t ) /2i = e−γ(t−t )/2 n. i = 1. .12) x(t. (5. (5.11) cn cn sin µ(t − t ) + Ω(nt − n t ) . The Green function for Eq.
In unstable zones or on the borderlines. 2 m A(t) = √ which satisfy the canonical commutation relation [A(t). t) of this invariant coincide—besides a timedependent phase factor—with the Floquet states of the system [86.1) possesses a T periodic Hermitian invariant operator. (5.2 The parametrically driven harmonic oscillator Floquet states in stable regimes It can be shown by group theoretical methods that the quantum mechanical quasienergy spectrum of a parametrically driven harmonic oscillator in a stable regime is equivalent to the energy spectrum of an undriven harmonic oscillator [83]. the quasienergy spectrum is equivalent to the energy spectrum of a parabolic barrier or of a free particle. 87. In a quantized version. They can be constructed in analogy to the energy eigenstates of the timeindependent harmonic oscillator: From the commutation relation (5. the socalled Lewis invariant [86] I(t) = A+ (t)A(t) = r(t) = 1 x r(t) − p r(t) ˙ 2 2 + x2 .15) x(t) = 0 0 A∗ ξ1 (t) + Aξ2 (t) .3) in the nondissipative case γ = 0 reads (5. the Floquet solutions for the Schr¨dinger equation are derived in the literature in various o ways [81.16) (5. 91].3).14) (5.16) one obtains √ A(t) ψα (x.34 5. The parametrically driven harmonic oscillator (5. t) = α + 1 ψα+1 (x. Here we sketch a derivation in the spirit of Ref.1). In the limit of zero driving amplitude they reduce to the familiar shift operators (A. (A. We restrict ourselves to the motion in stable regions.20) . r 2 (t) (5. A solution of the classical equation of motion (5.17) (5.13) 2m where A and A∗ are complex normal coordinates. A+ (t)] = 1. (5. t) = α ψα−1 (x. 86–90]. The latter cases result in a continuous quasienergy spectrum. [86]. The instantaneous eigenstates ψα (x. √ A+ (t) ψα (x. t).4) of the timeindependent harmonic oscillator. In these regions of the parametrically driven harmonic oscillator (5. 2 m i 0 ˙0 A+ (t) = − √ ξ2 (t)p − mξ2 (t)x . they are replaced by the conjugate pair of operators i 0 ˙0 ξ1 (t)p − mξ1 (t)x .19) (5. t). respectively.18) 0 0 ξ1 (t) ξ2 (t) = ϕ0 (t).
β+1 .26). .20). t) β ϕ0 (−t)δα. 2. . (5. Thus they do not lie within a single Brillouin zone. one ﬁnds the Floquet states φα (x. t) α! 0 ξ2 (t) 0 ξ1 (t) α/2 α (5. T Xαβ.6) for ϕ0 (t). . (5.β−1 + √ α ϕ0 (t)δα. t) into a 2π/Ωperiodic function and an exponential prefactor. α = 0.n are preferably evaluated in the spatial representation.26) einΩt Xαβ.10).β+1 .n = β c−n δα. to give Xαβ. Separating ψα (x.23) are chosen such that in the undriven limit they reduce to the eigenenergies of the harmonic oscillator. t) x φβ (x. The matrix elements of the position operator x with the Floquet states φα (t) .24) (5. (5.5.22) The corresponding quasienergies α = µ0 (α + 1/2) (5. t) = m/π Hα 2α α!ϕ0 (t) x 0 (t) ϕ m exp ˙0 im ξ1 (t) 2 x 0 2 ξ1 (t) . ∞ Xαβ (t) = −∞ dx φα (x. the periodicity of the Floquet states φα (t) has been used. 1. These states are solutions of the timedependent Schr¨dinger equation [86] and in the undriven limit reduce o to the position representation of the familiar eigenstates (A. t) = 0 and iterating according to (5. 0 To obtain Eqs.2 Floquet states in stable regimes 35 Solving A(t)ψ0 (x.29) 2m . read Xαβ (t) = φα (t)xφβ (t) = n (5.β−1 + √ α cn δα. we ﬁnd for I(t) the eigenfunctions ψα (x. which we will need to obtain the coeﬃcients of the master equation. .n 1 = T dt e−inΩt φα (t)xφβ (t) . (5.27) (5. t) = = (A+ (t)) √ ψ0 (x.n .25) and (5. The Fourier components Xαβ.25) (5. where Hα is the αth Hermite polynomial.21) 0 0 m/ ξ1 (t)ξ2 (t) exp m/π Hα x α α!ξ 0 (t) 2 1 ˙0 im ξ1 (t) 2 x 0 2 ξ1 (t) .28) = 2m by inserting the Fourier expansion (5.
the asymptotic state reads ∞ W∞ (x.3 The parametrically driven harmonic oscillator FloquetMarkov description in full RWA In the full rotatingwave approximation (RWA) introduced in Section 4.35) e Lα (2z 2 ). t) = is the Wigner function corresponding to φα (t) [89].29) in Eq.30) The eﬀective thermalbath occupation number N= n c0 n 2 (µ0 + nΩ) nth ( µ0 + n Ω) (5. p.36 5.3. we neglect all contributions with α − β = α − β in Eq. we switch to the Wigner representation. (5. (5. p. we obtain from (4. t) = α=0 ∞ αα Wα (x.2.25) the timeindependent master equation γ σαβ = ˙ (N + 1) 2 (α + 1)(β + 1)σα+1. we have to keep all terms with (α − β) = (α − β ). t).33) The basis {φα (t) } corresponds to the “generalized Floquet states” introduced in Ref.34) where (−1)α −z 2 (5.33). (5. Wα (x. Substituting Eq.32) σαβ = αβ = N +1 N +1 The density operator of the asymptotic solution is diagonal in this representation and reads ∞ ∞ (t) = α=0 ∞ αα φα (t) φα (t). (5. Thus in the present case of an equidistant quasienergy spectrum.23).β−1 − (α + β + 2)σαβ . i. Using the sum rule [92] ∞ α=0 κα Lα (x) = (1 − κ)−1 exp xκ κ−1 .β+1 − (α + β)σαβ 2 + N 2 αβσα−1. they are centered on the classical asymptotic solution and diagonalize the asymptotic density operator. To get the variances of (5.36) . this master equation coincides with the one for the undriven dissipative harmonic oscillator in rotatingwave approximation [50].e.25). (5. with the αth Laguerre polynomial Lα .31) reduces to N = nth ( ω0 ) in the undriven limit. (5. (4. Formally.. [26]. (4. There. p. π 1 0 0 0 0 ˙0 ˙0 ˙0 ˙0 z2 = mξ1 (t)ξ2 (t)x2 − ξ1 (t)ξ2 (t) + ξ1 (t)ξ2 (t) px + ξ1 (t)ξ2 (t)p2 /m . It has the stationary solution α 1 N ∞ ∞ δαβ .
(5. ˙ 0 ˙0 σpp (t) = m(N + 1/2)ξ1 (t)ξ2 (t). 0 0 (N + 1/2)ξ1 (t)ξ2 (t). 37 1 2 e−z /(2N +1) . p. we get an operator equation which only consists of position and momentum operators.44) (5. By substituting (5. . t) = It is a Gaussian with the variances σxx (t) = m 0 0 ˙0 ˙0 σxp (t) = (N + 1/2) ξ1 (t)ξ2 (t) + ξ1 (t)ξ2 (t) /2. this master equation is of Lindblad form [93] (see Appendix B. t).42) (5.30).40) To enable a comparison between the diﬀerent equations of motions for the dissipative quantum system. For a derivation. (5. π(2N + 1) (5.20) of the operators A and A+ . ] γ (N + 1) 2A A+ − A+ A − A+ A + 2 + N 2A+ A − AA+ − AA+ . but with the shift operators for Floquet states instead of the usual creation and annihilation operators.3 FloquetMarkov description in full RWA we obtain the asymptotic solution in Wigner representation as W∞ (x. p.14).38) (5. we also give for the master equation in RWA.19) and (5. p. we use the shift properties (5. (5. (5. 0 0 ˙0 ˙0 Dxp (t) = ξ1 (t)ξ2 (t) + ξ1 (t)ξ2 (t) (N + 1/2).30) for the density matrix elements σαβ i ˙ = − [HS (t).45) (5. Eq.1).28) yields for the Wigner function the diﬀerential equation ∂t W (x. with the diﬀerential operator LRWA (t) = − 1 γ p∂x + (∂x x + ∂p p) + k(t)x∂x m 2 γ 2 2 + Dxx (t)∂x + Dxp (t)∂x ∂p + Dpp (t)∂p 2 (5. to obtain the corresponding basisfree operator equation from the master equation (5. t) = LRWA (t) W (x.43) and the diﬀusion coeﬃcients 0 0 Dxx (t) = ξ1 (t)ξ2 (t)(N + 1/2)/m. Obviously.15).37) (5.39) (5. the corresponding partial diﬀerential equation in Wigner representation.25)–(A. Applying the transformations (A.5.46) ˙0 ˙0 Dpp (t) = m ξ1 (t)ξ2 (t)(N + 1/2).41) The dissipative part of this equation is the same as for the undriven dissipative harmonic oscillator [50].
for which the classical motion is integrable. This introduces an additional coupling term ∝ ppν .50) dτ K(τ ) G0 (t − τ. ]+ ] 2 γ γ − 2 Dpp [x. below. where a and bν ν are the usual annihilation operators of the system and the bath. Inserting this operator into Eq. In Wigner representation. (5. t ) + G0 (t.4 Basisindependent description beyond RWA In the present case of a bilinear system.51) . driven or not. indicated by the superscript 0 . (4. t ) for the corresponding undamped quantum system is given by the solution of the classical equation of motion in the limit γ → 0. the eﬀective master equation has the structure of (5. ] − γ [x. Note that within a Markov approximation.1). [p. t ) ∂t . [x.3). (5. This is in contrast to the nonMarkovian exact master equation [28]. In this latter case. t). with the periodically timedependent transport coeﬃcients ∞ (5. this corresponds to a timedependent diﬀusion coeﬃcient.42) containing derivatives with respect to x is a consequence of the RWA: Its eﬀect is equivalent to using the RWA coupling Hamiltonian HSB = ν gν (ab† + a† bν ) instead of (3. It consequently does not exhibit Lindblad form [93] (see Appendix B. (5. by returning to the original Markov approximation.12).48) Dpp (t) = − Dxp (t) = − γ mγ dτ K(τ ) 0 ∞ 0 ∂G0 (t − τ. ∞ G (t. the Heisenberg position operator xH (t. the master equation is periodic with the driving period T = 2π/Ω.6).49) (5. t ) = −x i i ˙ = − [HS (t). t ) = n. ]]. t ). (5. In our case the classical solution is given by (5. n n (5. This form of the master equation does not produce a positive semideﬁnite diﬀusion matrix. the knowledge of the classical dynamics opens a more direct access also to the quantal time evolution. Eq. The corresponding interactionpicture position operator reads p ∂G0 (t. leads to the master equation xH (t. In the next section we show how to avoid this RWA. see Eq. To evaluate these expressions. respectively. Speciﬁcally. (4.48) with coeﬃcients Dxp and Dpp which depend in a nonperiodic way on the time elapsed since the preparation at t0 . [p. we substitute the undamped limit of Eq.11).56).47) ∂t m where x and p now denote the position and the momentum operator.n =−∞ 0 c0 c0 sin µ0 (t − t ) + Ω(nt − n t ) . ]] + 2 Dxp [x. t =t (5. 5.1).38 The parametrically driven harmonic oscillator The fact that there are also dissipative terms in Eq.
55. Applying the transformations (A.53) is solved by contour integration in the upper half plane.5. Interestingly enough.4 Basisindependent description beyond RWA 39 The explicit time dependence in G(t.28) to the master equation (5.e.2. 5.54) π where C is the Euler constant. Its logarithmic divergence is regularized by a Drude cutoﬀ to obtain ∞ ∞ Dxp = − 2π P n=−∞ −∞ dω coth ω 2kB T 2 ω ωD . the 2 terms with ∂x x and ∂x are now absent.50) is completely diﬀerent. 2kB T (5. t ) results in a 2π/Ωperiodic time dependence of the coeﬃcients Dpp and Dxp .4. mγDxp coincides with the Drude regularized divergent part of the stationary momentum variance of a dissipative harmonic oscillator [54. In contrast to the FokkerPlanck equation with RWA in the last subsection. Averaging the transport coeﬃcients over one period of the driving is equivalent to the moderate rotatingwave approximation introduced in Section 4.45).17) and assuming an Ohmic bath. Dpp 1 = m 2 ∞ 0 cn (µ0 + nΩ) coth n=−∞ 2 (µ0 + nΩ) . we discuss the time evolution of the density operator in Wigner representation. (5. ∂t W (x. Expressing the resulting sums by the psi function ψ(x) = d ln Γ(x)/dx [92].1 Wigner representation and FokkerPlanck equation In order to achieve a description close to the classical phasespace dynamics. In addition. Thus the quasienergy spectrum approach is reﬂected solely by a drivinginduced modiﬁcation of the momentum diﬀusion Dpp . It originates from a principal part that has been neglected in the derivation of the FloquetMarkov equation in (4. we ﬁnd for Dpp in an average over one period of the driving. t) = L(t)W (x. p..3. we have to choose the cutoﬀ ωD much larger than the relevant frequencies µ0 + nΩ. The integral in Eq. 2 ω 2 − (µ0 + nΩ)2 ω 2 + ωD (5. After inserting the noise kernel (3. p.60].53) where P denotes Cauchy’s principal part. (5.55) . We have neglected terms of the order (µ0 + nΩ)/ωD . the cross diﬀusion Dxp in (5. i.25)–(A.18).52) This form makes explicit that the diﬀusion Dpp accounts for the quasienergies (µ0 + nΩ). The evaluation of the cross diﬀusion Dxp is more complex.48). I(ω) = mγω. and unrelated to the one in the RWA case (5. we obtain a cnumber equation of motion. we obtain Dxp = − ψ 1+ ωD 2πkB T +C . (5. t).
p. the diﬀusion matrix is not positive semideﬁnite. m Equation (5. so that (5. (5. or by using the formula for the conditional probability of a Gauss process [94].55) of this form with µ00 = 0 by the method of characteristics [95] in Appendix C. for any nonzero Dxp .58) dt [G(t.56) has the structure of an eﬀective FokkerPlanck operator. However.18). t )] . We construct a solution for (5. t ). 5. t) = e−µα t uα (x. uα (x.55) has solutions of Floquet form. it complies with the conditions of the Floquet theorem.56) L(t) = − p∂x + γ∂p p + k(t)x∂p + γDpp ∂p + γDxp ∂x ∂p . p. Since Eq. (5. p.56). correspondingly the FokkerPlancklike equation (5. This fact will be exploited in the following subsection to construct the solutions. p.55) with Eq. t ) ∂t . t + T ). t) = 1 2π σxx (t) σxp (t) σxp (t) σpp (t) x p t −∞ t −1/2 1 × exp − 2 with the variances σxx (t) = 2γDpp m2 2γDpp σxp (t) = m σxx (t) σxp (t) σxp (t) σpp (t) −1 x p (5. ∂t dt ∂ G(t. which contain the initial condition. In the present case. As is the case for the master equation from which it has been derived.4. t).55) with Eq.56) is not positive semideﬁnite requires to take a diﬀerent route. Eq. (5. the fact that the diﬀusion matrix of (5. vanish and we obtain the asymptotic solution W00 (x. 94]. In particular. In the limit t0 → −∞.59) (5. t ) −∞ t 2 (5.56) represents a diﬀerential equation with timeperiodic coeﬃcients. Consequently. a solution of the FokkerPlanck equation can be obtained directly by solving the equivalent Langevin equation [45.61) −∞ .55) with Eq. however. t) = uα (x. (5. dt G(t.60) 2 ∂ G(t. σpp (t) = −mγDxp + 2γDpp (5.56) has no equivalent Langevin representation. p.40 The parametrically driven harmonic oscillator with the diﬀerential operator 1 2 (5.2 WignerFloquet solutions The FokkerPlanck equation for the density operator in Wigner representation. (5. the coeﬃcients of the FokkerPlanck operator retain the periodicity of the driving. there exists a complete set of solutions of the form Wα (x. (5. oﬀers the opportunity to make full use of the wellknown and intuitive results for the corresponding classical stochastic system.57) henceforth referred to as WignerFloquet functions. the terms in the ﬁrst line of (C.
5. (5.11) for G(t. the functions n Wnn (x.3 Inﬂuence of the driving on the master equation The master equation in operator notation (5. . We can recover this solution by inserting the classical diﬀusion constant mkB T and the undriven limit ε → 0 for the classical solution.65).67). (5. t) = Q1+ (t) Qn (t) W00 (x. Nevertheless. .9). 1. (5.68) This spectrum is independent of the diﬀusion constants. (5. Due to Eqs. p.48) and the FokkerPlanck equation (5. .56) [96].64) (5. p. Taking the commutation relation (5. since it is a correction of order γ. The expression for the eigenfunctions in the hightemperature limit of the (undriven) classical Brownian harmonic oscillator in Refs. Q2+ (t + T ) = e(−γ/2−iµ)T Q2+ (t). they are formally independent of the Floquet basis. and therefore is the same as in the case of the classical parametrically driven Brownian oscillator [97].55). The operators Qi+ (t) have the properties [L(t) − ∂t . as expected for an operator of type (5. we construct further WignerFloquet functions: By solving the characteristic equations (see Appendix C). we ﬁnd the two timedependent differential operators Q1+ (t) = f1 (t)∂x + mf˙1 (t)∂p . (5.61) the diﬀerence in using Dpp and D = Dpp + γDxp [see Eq. By inserting the Fourier representation (5. 2+ (5.4. (5.14)] is meaningless. t).64) into account. t ). This allows for a detailed analysis of the diﬀerence between the Markovian approach with respect to the unperturbed spectrum and the quasienergy spectrum approach beyond mere diﬀerences in representation.66) they are of Floquet structure with the Floquet spectrum µnn = (n + n )γ/2 − i(n − n )µ. Q2+ (t)] = 0 and Q1+ (t + T ) = e(−γ/2+iµ)T Q1+ (t). Starting from W00 . (C.1.67) also solve Eq. n = 0.55) given in this section result from a Markov approximation with respect to the quasienergy spectrum. Q1+ (t)] = [L(t) − ∂t .4 Basisindependent description beyond RWA 41 Note that in (5. 98] is also of the structure (5. Q2+ (t) = f2 (t)∂x + mf˙2 (t)∂p .66) n.63) where the solutions fi (t) of the classical equation of motion are given by (5.65) (5. [96. 5. .59)–(5. given in Section 5. one ﬁnds that the variances are asymptotically time periodic. 2.62) (5.
t )F (t ).48). We obtain a master equation of the form (5. i. 1 Dpp = lim Dpp = m ω0 coth ε→0 2 ω0 2kB T . (5.70). t ) ∂G0 (t. the Markovian . Thus parametric driving of a dissipative harmonic oscillator modiﬁes the momentum diffusion in the master equation.42 The parametrically driven harmonic oscillator Parametrical driving The Markov approximation with respect to the unperturbed spectrum can be obtained from the (in general more complicated) quasienergy spectrum approach by replacing the coeﬃcients of friction and diﬀusion by their corresponding limits for zero driving amplitude ε. i.71) dt G0 (t. where the momentum diﬀusion coeﬃcient Dpp is replaced by its limit for ε → 0. (5. given by the third term.73) into (3. t t (5.69) In general Dpp = Dpp . k(t) = mω0 . HF (t) = HS (t) − xF (t).e. ∂t Thus we obtain a cnumber correction to the interactionpicture position operator (5. t ) + ∂t m m ∂ 2 G0 (t. x and can be integrated to yield the interactionpicture operators xH (t. which we verify by numerical studies in Section 5.48) and accordingly a FokkerPlanck equation of the form (5.47). After inserting (5. t ) pH (t.6.72) (5. Thus the level separations remain unaﬀected and we expect no change in the dissipative part of the master equation (5. t ) = −x 1 p ∂G0 (t.38). The classical equation of motion..70) 2 With HS (t) being a timeindependent harmonic oscillator.73) dt t ∂G0 (t.e.55). Additional additive driving The Markovian master equation within the quasienergy spectrum approach undergoes a further modiﬁcation when the parametric oscillator is subject to an additional additive driving −xF (t). t ) F (t ). It is known that the only eﬀect of the driving force F (t) on the (quasi) energy spectrum of a parametrically driven harmonic oscillator is an overall level shift [82].. now reads m¨ + k(t)x = F (t). the corresponding Markovian master equation in RWA for the dissipative system was already given in [26]. which is also obeyed by the interactionpicture position operator. Herein we generalize these results for the combined timedependent system Hamiltonian in (5. t ) + G0 (t. t ) = −x m +p + ∂t ∂t ∂t t (5.
] with an eﬀective total driving force ˜ F (t) = F (t) + 0 ∞ t−τ (5. γ(τ ) = 2γδ(τ ). Thus the equation of motion for the density operator has the structure i ˜ ˙ = . . With an Ohmic bath.3). t )F (t ).77) Note that the dissipative parts of (5. the inner integral in (5.3.48). the second one is a correction of the driving force due to interaction with the bath. The term in the ﬁrst line stems from the reversible part of the master equation (4. To obtain the conservative limit γ → 0 of these. ] + i 0 ∞ 43 (5.77) vanishes and ˜ we obtain F (t) = F (t). Thus in contrast to an explicit parametric time dependence k(t) in the quadratic part of the Hamiltonian. in this case. + F (t)[x.59) and get σxx (t) = − γDpp 2m2 cn cn n. (5. ∂t (5.1 Asymptotics The conservative limit In contrast to the Markov approximation with RWA in Section 5. .78) − 2f1 (t)f2 (t) eγt+i[2µ+(n+n )Ω]t eγt−i(n−n )Ωt 2 +f2 (t) γ − i(n − n )Ω γ + i[2µ + (n + n )Ω] In the limit of weak damping. that we must use a parametric timedependence to study diﬀerences in the dissipative parts resulting from the Markov approximation with respect to the energy spectrum versus the Markov approximation with respect to the quasienergy spectrum. we insert the Green function (5.11) into (5. given by the right hand side of Eq. the variances in both Markov approximations without RWA still depend on the friction γ. γ µ + nΩ for any integer n.76) are not aﬀected by the additive driving force F (t). the time dependence of an additive force. Note that this condition is violated in . (5.74) 2 m t−τ t dτ γ(τ ) [x. ] dt G0 (t − τ. (5.76) dτ γ(τ ) t dt ∂G0 (t − τ.n 2 f1 (t) eγt−i[2µ+(n+n )Ω]t γ − i[2µ + (n + n )Ω] . . does not change the Markovian master equation of the dissipative system. + F (t)[x. only the case n = n of the second term in the brackets remains. This makes explicit. 5.5 Asymptotics master equation emerges as i ˙ = . .5 5.75) The dots denote the old result for F (t) ≡ 0. t ) F (t ).5.5.
we get σxx (t) = B where B= n=−∞ Dpp 0 0 ξ (t)ξ2 (t). i.84) With the sum rule (5. . all diagonal elements Wnn (x. It is satisﬁed if the variances fulﬁll the inequality σxx (t) σxp (t) σxp (t) σpp (t) = Dpp B m 2 ≥ 2 /4. In an analogous way. (5.3..44 The parametrically driven harmonic oscillator parameter regions where the Floquet index becomes a multiple of Ω.e. For consistency. (5.55) with γ = 0.2 The hightemperature limit In the limit of high temperatures kB T ωD . In the reﬁned approach (Section 5. 5. the cross diﬀusion Dxp vanishes in the hightemperature limit. (5. t) are Floquet functions with the quasienergies µnn = 0. 5. Moreover. an equation of the form (5. t) can be viewed as dissipationadapted Floquet functions.83) which we have veriﬁed numerically for the case of the Mathieu oscillator. the FokkerPlanck equation is already of the required structure.4).79) c0 n 2 (5. With ψ(1) = −C [92]. p. we ﬁnd σxp (t) = B Dpp ˙0 0 0 ˙0 ξ (t)ξ2 (t) + ξ1 (t)ξ2 (t) . we expect the FokkerPlanck equation for the Wigner function to give the Kramers equation for the classical Brownian motion [97]. However. in this limit γ → 0.35) of the corresponding Schr¨dinger o equation. 2m 1 ˙0 ˙0 σpp (t) = BDpp ξ1 (t)ξ2 (t). Due to the degeneracy of the Floquet indices. which are also solutions of the coherent equation of motion. 2 1 m ∞ (5. For the position variance.82) Besides the prefactor.55) with diﬀusion constants Dxp = 0 and Dpp = mkB T .8).5. we check the positionmomentum uncertainty relation for the asymptotic solution. as is the case along the borderlines of the regions of stability in parameter space (cf.80) denotes a number of order unity.81) (5.1). they are diﬀerent from the Wigner representation of the stationary solutions (5. Eq. For Dpp . Fig. (5. we use coth x = 1/x + O(x) and get Dpp = mkB T n c0 n 2 (µ0 + nΩ). p. this reduces to Dpp = mkB T . The limγ→0 Wnn (x. this is no contradiction. these variances are the same as for the master equation with RWA in Section 5.
we choose 2 k(t) = m ω0 + ε cos Ωt .4). the classical Floquet index µ and the Fourier coeﬃcients cn are determined numerically by continued fractions [45].. 5.4 that the inﬂuence of the driving on the master equation results in a modiﬁcation of the momentum diﬀusion. obtained from a Markov approximation with respect to the unperturbed spectrum. reads kB T 1 2 kB T c0 =B . Note that within the unstable regimes. Appendix A). (5.3a and 5. Figure 5. The discrepan2 cies become most signiﬁcant for strong driving and large ω0 . Eq. Both for low driving 2 amplitude ε ω0 and high temperature T ω0 /kB . [28].86) and the ansatz (5. Speciﬁcally. the diﬀerence vanishes. We showed in Section 5.5 Ω is suﬃciently large. 5. we compare our approximate results to exact ones. i. in the hightemperature limit. we obtain the tridiagonal recurrence relation 2 εcn−1 + 2 ω0 − γ 2 /4 − (µ + nΩ)2 cn + εcn+1 = 0.87) From this equation. (5. we give the numerical results for the Mathieu oscillator. The chosen driving parameters ω 2 = 6.5. time and driving parameters are given in the units which are commonly used in mathematical literature [85] to obtain the scaled Mathieu equation ¯ x + (¯ 0 + 2¯ cos(2t))x = 0. (5. with the classical result in the limit γ → 0. which results from the quasienergy spectrum approach. This factor.53513 Ω/2).86) By inserting (5. obtained from the pathintegral solution in Ref. The temperature kB T = 0. The parameters ω0 and ε are varied along the full line in the inset.85) N+ = n 2 n Therefore the diﬀusion constants Dxx and Dxp remain ﬁnite and the FokkerPlanck operator (5.6) into (5. The variances σxx (t) and σpp (t) of the Markov approximations without RWA are compared to the exact results [28] in the Figs. The numerical values are given in units of the classical 2 momentum diﬀusion coeﬃcient mkB T .42) does not approach the Kramers limit for high temperatures.3. Variances are plotted in units of the corresponding ¨ ω2 ε groundstate variance for zero driving amplitude (cf. In the ﬁgures.2 compares the diﬀusion coeﬃcient Dpp . (5. Nevertheless the asymptotic variances in RWA coincide for high temperatures.3b. the variances and diﬀusion constants scale with N + 1/2. but with quantum eﬀects still . perturbation theory is not valid. Nevertheless.6 Numerical results In this section.6 Numerical results 45 In the quasienergy spectrum approach with RWA in Section 5.e. to the diﬀusion coeﬃcient Dpp .52) gives a smooth interpolation.5 Ω2 and ε = 7 Ω2 lie inside the ﬁfth stable zone (µ = 4.
but happen to occur in the regions with negative slope.3. As depicted in Fig.5.6. .4.2: The diﬀusion constants D pp for the simple (dotted) and Dpp for the improved (dashed) Markov approximation in units of the classical diﬀusion constant mk B T 2 for kB T = 0.1). The relative error ηxx (t) = Markov exact σxx (t) − σxx (t) exact σxx (t) (5. Results for the Markovian treatment within RWA. For the chosen parameters it is reduced by the use of the improved Markov scheme by approximately 30%. This reﬂects the breakdown of the weakcoupling approach. Fig. Note that the maximal deviations do not occur in the extrema.5 Ω. In the Figure we depict asymptotic times t > 100/Ω. given in Section 5. For this example. Nevertheless. The driving parameters are the same as in Fig. The asymptotic covariance elements retain the periodicity T = 2π/Ω of the external driving. the solution without RWA yields—up to a scale—a better overall agreement with the exact behavior over a full driving period T . where transient eﬀects have already decayed. 5. 5.3. are depicted for the position variance σxx (t) in Fig. appreciable. 5. We note that the improved Markovian treatment in Section 5.4.88) of the position variance for these two Markov approximations is depicted in Fig.5 h 2 ε = ω0 [mkB T ] 4 [Ω2 /4] 2 ω0 3 2 1 0 0 10 20 10 0 0 5 10 ε [Ω2 /2] 15 Dpp 20 30 2 ε = ω0 [Ω2 /2] Figure 5.46 The parametrically driven harmonic oscillator 6 5 kB T /¯ Ω = 0. The parameters ω0 and ε are indicated by the full line in the inset (cf. which accounts for the quasienergy diﬀerences. 5. agrees better with the exact prediction. the quality of agreement to the exact result is similar for both Markov approximations. the quality of both Markov approximations worsens with increasing dissipation strength γ. 5.
The scaled driving period T = 2π/Ω is indicated in panel (a).5 Ω2 .6 0.6 Numerical results 47 14 12 [¯ /2mω0 ] h (a) T 10 8 6 4 2 0 50 51 52 53 t [Ω/2] 54 55 σxx 12 (b) 10 [m¯ ω0 /2] h 8 6 4 2 0 50 51 52 53 t [Ω/2] 54 55 Figure 5.3 0.3: The asymptotic variances σxx (t) (a) and σpp (t) (b) with period T = 2π/Ω for the simple (dotted) and the improved (dashed) Markov approximation. 5.0 50 51 52 53 t [Ω/2] 54 55 Figure 5.1 0.4 0. σpp ηxx 0.5. compared to the exact result (full line) for the parameters ε = 7 Ω2 . kB T = 0.5 Ω and γ = Ω/20. 2 ω0 = 6.3a.2 0.4: Relative error ηxx (t) for the position variances of Fig.5 0. .
The driving parameters are ε = 7 Ω2 2 and ω0 = 6. In the crudest treatment introduced in Section 4.6: Position variances obtained with the Markov approximation with respect to the quasienergy spectrum with (dotted) and without (dashed) RWA. gain an explicit time dependence with the periodicity of the driving. refers to the degree to which changes in dynamical and spectral properties of the central system due to the driving are taken into account.5 Ω2 .5: The time averaged variance σxx t for the simple (dotted) and the improved (dashed) Markov approximation. the use of the improved FloquetMarkov . ω0 = 6.6 γ [Ω/2] 0.4 0. where the dissipative terms in the master equation are derived ignoring the explicit time dependence of the Hamiltonian.5 Ω. The energydomain quantity relevant for all subsequent developments is then the quasienergy spectrum. 0.8 1. instead of the unperturbed spectrum.48 The parametrically driven harmonic oscillator 8 [¯ /2mω0 ] h 6 4 2 0 Figure 5. and the driving only appears in the coherent term. compared to the exact result (full line) for γ = Ω/20 and kB T = 0.2 0. In the time domain.0 σxx t 12 10 [¯ /2mω0 ] h 8 6 4 2 0 50 51 52 53 t [Ω/2] 54 55 Figure 5. obtained within the Floquet formalism. An improved master equation results from the FloquetMarkov scheme which we obtained in Section 4. compared to the exact result (full line) for the 2 parameters ε = 7 Ω2 . Besides the diﬀerences in representation.5 Ω. the quantities entering the dissipative terms of the master equation.1.7 σxx Conclusion The principal distinction to be made among possible Markovian approaches to the driven dissipative dynamics. such as Heisenbergpicture operators of the central system.2 by coupling the central system and the driving as one whole to the heat bath. 5.5 Ω2 and kB T = 0.
by neglecting reservoirinduced virtual transitions between Floquet states of the central system that violate quasienergy conservation. these limit cycles are trivial and correspond to a ﬁxed point at the origin. the Floquet formalism is a useful device to construct and classify solutions. A time dependence arises only by the periodic variation of the shape of the asymptotic distributions. undergoes a renormalization which vanishes. the corresponding master equations have Floquet structure throughout.4 mainly results in a modiﬁed momentum diﬀusion that depends on the quasienergy spectrum instead of the unperturbed spectrum of the central system. the centers of gravity of the asymptotic quasiprobability distributions follow the corresponding classical limit cycles. speciﬁcally in terms of the Wigner representation of the density operator and its equation of motion. Even within the improved Markov approach.. then the equation of motion for the reduced density operator complies with the conditions for applicability of the Floquet theorem.e. and thus manifestly generates a dynamical semigroup. Apparently a drawback. They represent the quasiprobability distributions closest to the Floquet solutions of the corresponding classical FokkerPlanck equation. The diﬀerence becomes signiﬁcant in the limits of strong driving amplitude and low temperature. however. Since all variants of the Markov approximation discussed here truncate the memory of the central system on time scales shorter than the period of the driving. This is not the case if the RWA is avoided. allows for memory eﬀects of unlimited duration and thereby generally prevents the consistent deﬁnition of a propagator over a single period. the lack of a Lindblad structure in the master equation without RWA faithfully reﬂects the failure of the Markov approximation on short time scales. An additive timedependent external force. A signiﬁcant simpliﬁcation of the master equation is achieved by a rotatingwave approximation. They are not literally stationary but retain the periodic time dependence of the driving. The resulting master equation has Lindblad form. applied in addition to or instead of the parametric driving. As a consequence. Since all FokkerPlanck equations obtained are time periodic. The exact pathintegral solution. In the case of parametric driving. or WignerFloquet states in short. i. An analogous situation as with the Lindblad form of the master equation arises with its Floquet structure. here.e. as are the corresponding master equations. In this representation. i. Concluding from a numerical comparison of certain dynamical quantities. in contrast. for the . Additional insight is gained by discussing the dynamics in terms of phasespace distributions. can be written as eigenfunctions of a generalized nonunitary Floquet operator that generates the evolution of the density operator over a single period. Since we are here dealing with a linear system. with creation and annihilation operators acting on Floquet states. WignerFloquet states with Floquet index zero correspond to asymptotic solutions. integrated over a single period). their solutions may be written as eigenstates of a WignerFloquet operator (the FokkerPlanck operator evolving the Wigner function. ﬁner levels of approximation can be distinguished. If all coeﬃcients are at most periodically time dependent.5.7 Conclusion 49 approximation in Section 5. the solutions can be cast in Floquet form. in the case of an Ohmic bath.
Technical advantages of the Markov approximation in general and of its various ramiﬁcations—easy analytical and numerical tractability. even in parameter regimes where the respective approximations are expected to become problematic. However. the attributes “simple” and “improved” for the two basic Markovian approaches prove adequate. the diﬀerences in quality are not huge and the agreement with the exact solution is generally good. desirable formal properties such as Floquet or Lindblad form of the master equation—can justify to tolerate their quantitative inaccuracy. Results for the Markov approximation based on the quasienergy spectrum show consistently better agreement with the exact pathintegral solution than those for the Markov approximation with respect to the unperturbed spectrum.50 The parametrically driven harmonic oscillator speciﬁc case of the Mathieu oscillator. .
6 The harmonically driven doublewell potential In this chapter we use the FloquetMarkov scheme to investigate the interplay of chaos and dissipation in a bistable quantum system. we reveal the limitations of the threelevel approximation and identify additional features of the full dynamics not covered by it. 30–33. While in the coherent case the dynamics is well described in a threestate approximation. this attractor may be of quite diﬀerent nature. In particular. the coupling to the environment indirectly couples the three states to all other states. we have a mixed phase space. Tunneling is extremely sensitive to any disruption of coherence as it occurs due to the unavoidable coupling to the environment: In presence of dissipation.2.1 we introduce its Hamiltonian and the underlying symmetries. coherent tunneling becomes a transient that fades out on the way to an asymptotic state [11. On the basis of numerical results for the full driven double well with dissipation. Depending on friction strength and details of the system. . Chaotic tunneling comes about by an interplay of classical nonlinear. where the coexistence of regular and chaotic regions leads to a variety of uncommon coherence phenomena. In Section 6. we brieﬂy review coherent driven tunneling as well as its modiﬁcation caused by the inﬂuence of classical chaos. Thus. the socalled attractor [3]. the structures associated with classical attractors are smeared out on a scale but leave their trace in the asymptotic state of the corresponding dissipative quantum map [106]. 12]. the coherent exchange of probability between symmetryrelated regular islands that are separated by a chaotic layer. we consider the longtime asymptotics and the phasespace structure associated with it. Most prominent among them is chaotic tunneling [13–17. Switching on friction has a dramatic consequence for the classical phase space: A volume element contracts exponentially in time and therefore all trajectories converge towards a submanifold of phase space with zero volume. dynamics and quantum coherence. Moreover. We study coherent and dissipative chaotic tunneling in the vicinity of such singletdoublet crossings in Section 6. On a quantum level. If the dissipative dynamics is also chaotic.3. for suﬃciently strong friction. not by a static potential barrier. the attractor typically shrinks to a limit cycle or a set of isolated ﬁxed points. We study the classicalquantum correspondence of the asymptotic state in Section 6. The harmonically driven quartic double well will serve as our working model. The quasispectrum associated with chaotic tunneling exhibits a characteristic feature: Quasienergies of chaotic singlets intersect tunnel doublets which are supported by regular tori. For moderate driving near the classical resonances. typically bistable. 99–105]. the attractor has in general fractal geometry—it forms a socalled strange attractor. chaos already plays a significant role for the classical dynamics although the motion near the bottom of the wells is still regular.
It is deﬁned by the Hamiltonian H(t) = HDW + HF (t). ¯ 2 2 ¯ where the dimensionless quantities x. ˙ x = p.2) (6. of the x2 and the x4 term.52 The harmonically driven doublewell potential t = (n + 1/2)π/Ω t = 2πn/Ω V (x. we consider the quartic double well with a spatially homogeneous driving force.3) The potential term of the static bistable Hamiltonian HDW possesses two minima at 2 x = ±x0 .6) This implies that the classical dynamics is independent of the barrier height EB . x0 = (8EB /mω0 )1/2 . HDW = 2m 4 64EB HF (t) = Sx cos(Ωt). Apart from mere scaling.4) 1 3 1 ¯¯ ˙ ¯ ¯ (6. it has no free parameter. the classical phase space of HDW only depends on the presence or absence. 4 1 m2 ω 0 4 p2 2 − mω0 x2 + x . The inﬂuence of the driving on the classical phasespace structure is fully characterized by the rescaled amplitude and frequency of the driving. ω0 (6. −x0 6.1: Sketch of the driven double well potential described by the timedependent Hamiltonian (6.1). p/mω0 x0 and ω0 t. ¯ ¯ (6. Besides that.1) at various times.1) (6. p and t are given by x/x0 . ¯ ¯ respectively. harmonic in time. Fig. .1 The model As a prototypical working model. separated by a barrier of height EB (cf. The parameter ω0 denotes the (angular) frequency of small oscillations near the bottom of a well. This is obvious from the scaled form of the classical equations of motion. 6. Ω ¯ Ω= .5) p = x − x − F cos(Ωt). and the signs. F = S 2 8mω0 EB . (6. t) 0 EB −EB ω0 x0 x Figure 6.
79]. This enables a treatment within the FloquetMarkov scheme. can be chosen as real [8.7) and distinguishes the semiclassical from the deep quantum regime.1 Symmetries The model Hamiltonian (6. such that the variation of the potential at the bottom of the wells is much smaller than the barrier height.11)]. for the sinusoidal shape of the driving together with the initial phase chosen above. Timereversal symmetry It is well known that the energy eigenfunctions of an (undriven) Hamiltonian which obeys timereversal symmetry. p → −p. thus possesses discrete timetranslational invariance.6. Timereversal symmetry is typically broken by a magnetic ﬁeld (recall that a magnetic ﬁeld is described by an axial vector and changes sign under time reversion) or by an explicit timedependence of the Hamiltonian. x0 . The classical limits hence amounts to D → ∞. Eq. which allow for an improvement of numerical eﬃciency and also for a classiﬁcation of the Floquet states as even or odd. we ﬁnd two more discrete symmetries. apart from computational advantages. we restrict ourselves to moderate driving amplitudes. −t) . It is approximately given by D= EB . 6. ω0 (6.1 The model 53 In the quantummechanical case. This has. In addition. If now φ(x. t → −t (6. then φ∗ (x. This is evident from the classical scales for position.9) denotes the eﬀective quantum of action.10) is retained and the Floquet Hamiltonian obeys H(t) = H∗ (−t) [cf.8) = mω0 x2 0 (6. However. mω0 x0 . In the following. this holds no longer true: The ﬁnite size of Planck’s constant results in a ﬁnite number of doublets with energy below the barrier top.1. introduced in Chapter 4. This implies that the bistable character of the potential is retained at any time. 79]. t) is a Floquet state in position representation with quasienergy . the positionmomentum 0 uncertainty relation in the scaled phase space (¯. timereversal symmetry T: x → x.1) obviously is 2π/Ωperiodic in time. (2. and momentum. however. also direct physical consequences for the level statistics of quantum systems with chaotic classical counterpart [8. introduced above: The corresponding action scale is mω0 x2 and therefore. p) reads x ¯ ∆¯ ∆¯ ≥ x p where eﬀ eﬀ 2 = 1 8D (6.
. i. . which translates to φ(x. p. . Thus. . . the fact that the generalized parity acts on the composite Hilbert space results in a particularity: If φ(t) is e. . (6. two equivalent Floquet states from neighboring Brillouin zones possess diﬀerent generalized parity. i. . . ω) in the frequency regime. . ψ(t) = exp(−i t/ )φ(t) . 107.e. . Thus. . p → −p. t → t. 108].54 The harmonically driven doublewell potential also is a Floquet state with the same quasienergy. While such a discrete symmetry is of minor importance in classical physics. and phase. we have to diagonalize the even supermatrix . . time t mod(2π/Ω) or in the composite Hilbert space R ⊗ T . the Fourier coeﬃcients of the Floquet states can be chosen real. . Quasienergies from diﬀerent symmetry classes may intersect. it is suﬃcient to compute all eigenvectors of the Floquet Hamiltonian in the even subspace whose eigenvalues lie in the ﬁrst two Brillouin zones. thus allowing for a classiﬁcation of the Floquet states as even or odd. respectively. The invariance of the system under the generalized parity is also of considerable help in the numerical treatment of the Floquet matrix (2. · · · Ee + 2 Ω 0 Xeo 0 0 ··· ··· Eo + Ω 0 Xeo Xoe 0 ··· Ee Xeo 0 0 Xoe · · · . . This means that a classiﬁcation of the corresponding solutions of the Schr¨dinger equation. It is deﬁned by the operation PΩ : x → −x. ω) = φ∗ (x. However. To obtain a complete set of Floquet states. however. This means that we can always choose the Floquet states by linear combination such that φ(x.. . t) = φ∗ (x. −t). . whereas quasienergies with the same symmetry typically form avoided crossings [79]. Generalized parity The undriven Hamiltonian HDW is invariant under the parity P: x → −x. 100]. an even Floquet state.12) He = · · · ··· 0 0 Xeo Eo − Ω Xoe ··· ··· 0 0 0 Xoe Ee − 2 Ω · · · . . p → −p. then φ(1) (t) = exp(iΩt)φ(t) turns out to be odd. a more general. . dynamical symmetry remains [10.e.. . . . . the odd Floquet states are obtained by shifting the (even) ones from the second to the ﬁrst Brillouin zone. ..53) [16. as even or odd o requires a restriction to a single Brillouin zone. This symmetry is destroyed by a linerarly coupled driving ﬁeld. . .g. . .11) and represents a generalized parity acting in the extended phase space spanned by x. . . which changes their generalized parity. . t → t + π/Ω (6. its inﬂuence on the quantum mechanical quasispectrum { α (F )} is more distinct: It devides the composite Hilbert space in an even and an odd subspace. .. . The even Floquet states are given by the eigenvectors of He from the ﬁrst Brillouin zone. With the above choice of HF (t). .
Eq.. respectively. . Φ− acquire a relative phase exp(−i∆n t/ ) and ΦR .5 x2.1 x0. Eo = 0 0 E 4 5 . Therefore. . . .14) 6.6.. Thus. (6. .1. . we can gain a qualitative picture of its eigenstates from simple torus quantization: The unpaired tori correspond to singlets with positive energy. .13) x1.2 x5. The matrices E0 0 0 · · · E1 0 0 0 E2 0 · · · 0 E3 0 Ee = 0 0 E · · · .1 The model which for the same number of Floquet channels has only half original Floquet matrix (2.5 . .1 x2. each partner of which oscillates in either one of the two potential minima. There is a separatrix at E = 0. whereas the symmetryrelated pairs below the top of the barrier correspond to degenerate pairs of eigenstates.53).0 x5. .2). . ΦL are transformed n n n n into one another after a time π /∆n .. ··· ··· ··· . predicts that the partners of these pairs have small but ﬁnite overlap. purely quantum mechanical frequencyscale. . .0 x1. Neighboring pairs are separated in energy approximately by ω0 . The other set consists of unpaired trajectories. . . . . . . with E < 0. We can always choose the globals phases such that the superpositions 1 ΦR. . however. that encircle both wells in a spatially symmetric fashion.2 x1. Xeo = S 2 x0. driving. and dissipation With the driving HF (t) switched oﬀ. . . (6. Due to the integrability of the undriven double well. . . the states Φ+ . Xoe = S 2 55 the dimension of the ··· ··· ··· .3 x4. which reﬂects the almost harmonic potential shape near the bottom of the wells. As time evolves. . . . . .4 . . .0 x3.15) are localized in the right and the left well. denote the undriven Hamiltonian HDW and the coupling to the driving ﬁeld H1 = Sx/2.2 Tunneling. This introduces an additional. . . Typically..1 x4. .4 x5. each of which consists of an even and an odd state. Φ+ and Φ− .3 x2. comes in symmetryrelated pairs. . the tunnel rate ∆0 / of a particle which resides in the groundstate doublet. ··· ··· ··· .4 x3. .L = √ Φ+ ± Φ− n n n 2 (6. (6. The energies of the nth doublet are separated by a small tunnel n n splitting ∆n . the classical phase space generated by HDW exhibits the constituent features of a bistable Hamiltonian system. . . . It forms the border between two sets of trajectories: One set.3 x0. with E > 0.5 x4. . which are part of the supermatrix He . decomposed into the even and odd eigenstates of HDW . the particle tunnels forth and back between the wells with a frequency ∆n / . Exact quantization.2 x3. tunnel rates are extremely small compared . the true eigenstates come in doublets. . .
Increasing the amplitude of the driving from zero onwards has two principal consequences for the classical dynamics: The separatrix is destroyed as a closed . thus the splitting vanishes and tunneling is brought to a complete standstill by the purely coherent inﬂuence of the driving [10]. For adiabatically slow driving. and an additional energy scale is introduced. i. As an immediate consequence. In general. In the dissipative case. even if its inﬂuence on the classical phase space is minor. If the driving is faster. which is always larger than its unperturbed value ∆0 and results in an enhancement of the tunneling rate [107]. the coherent suppression of tunneling [10. A driving of the form (6. However. For example. tunneling is governed by the timeaverage of the instantaneous tunnel splitting. < ∆0 / ∼ Ω ω0 . there exist counterintuitive eﬀects. the opposite holds true: The relevant time scale is now given by the inverse of the quasienergy splitting of the groundstate doublet / 1 − 0 . 6. It has been found [107.3). It even happens that the quasienergies of the groundstate doublet (which are of diﬀerent generalized parity) intersect as a function of the driving amplitude F . reﬂecting the growth of the transition rates (4.23) [53].56 The harmonically driven doublewell potential to the frequencies of the classical dynamics. the eﬀective ﬁnite width attained by each discrete level. So far. Coherent tunneling is in this case well described within a twostate approximation [107. The small energy scales associated with make tunneling extremely sensitive to any disruption of coherence. It enters already on the level of classical mechanics since small oscillations near the bottom of the wells become resonant and classical chaos comes into play. as it occurs due to the unavoidable coupling to the environment. a parameter regime where classical motion is predominantly regular. all the more in the semiclassical regime we are interested in.109] that in this case for ﬁnite driving amplitude  1 − 0  < ∆0 . thus tunneling is always decelerated. Ω ∆0 / . This corresponds in a quantum description to resonant multiple excitation of interdoublet transitions until levels near the top of the barrier are signiﬁcantly populated. 109]. for driven tunneling in the vicinity of an exact crossing of the groundstate doublet. a twostate approximation of course fails for temperatures kB T ∼ ω0 . where thermal activation to higher doublets becomes relevant. how> ever. we have considered only driving frequencies much smaller than the frequency scale ω0 of the relevant classical resonances. can entail signiﬁcant consequences for the tunnel dynamics: It may enlarge the tunnel rate by orders of magnitude or even suppress tunneling at all. this time scale gets shorter for higher temperatures. Tunneling and related coherence phenomena are thus rendered transients that occur—if at all—on the way towards an asymptotic equilibrium state and fade out on a time scale tdecoh .3 The onset of chaos Driving with a frequency Ω ≈ ω0 has an even stronger inﬂuence on the dynamics of the bistable system.1. 107] can be stabilized with higher temperatures [76–78] until levels outside the doublet start to play a role. 12..e. the symmetry underlying the formation of tunnel doublets is generally broken.
As a whole. around the ﬁrst resonance introduces a second ladder of doublets into the spectrum. closed trajectories diverges for E → 0. By its sheer phasespace area.6. The appearance of a regular region. This opens the way for diﬀusive transport between the two potential wells. 100].112]. Fig. The tunnel doublets characterizing the unperturbed spectrum for E < 0 pertain to states located on pairs of symmetryrelated quantizing tori in the regular regions within the wells. They can then ﬂuctuate freely in the spectrum and thereby “collide” with other chaotic singlets or regular doublets. the corresponding eigenvalues detach themselves from the regular ladder to which they formerly belonged. there is an inﬁnite set of resonances of the driving with the unperturbed motion. the resonances accumulate towards the separatrix of the unperturbed system. (6.1). As soon as a pair of states is no longer supported by any toruslike manifold. Eq. the quantizing tori successively resolve in the chaotic sea. The gradual widening of the doublets proceeds as a smooth function of the driving amplitude [16. curve and replaced by a homoclinic tangle [110] of stable and unstable manifolds. Due to the nonlinearity of the potential. See Section 6. both inside and outside the wells [111. The corresponding doublets disappear as distinct structures in the spectrum as they attain a splitting of the same order as the mean level separation. it forms a chaotic layer in the vicinity and with the topology of the former separatrix (cf. With increasing size of the chaotic layer. Both major tendencies in the evolution of the classical phase space—extension of the chaotic layer and growth of the ﬁrst resonance—leave their speciﬁc traces in the quasienergy spectrum. 114]. including fractal [115] and vague tori [116].6). 6. 113.2: Tunneling phenomena and the according appropriate levels of description for the nondissipative driven doublewell potential. large enough to accommodate several eigenstates.1 The model 57 almost regular adiabatic energies quasienergies chaos twolevel description coherent destruction of tunneling ∆0 /¯ h multilevel description chaotic tunneling ω0 Ω Figure 6.1 for a detailed discussion. Size and shape of the ﬁrst resonance vary in a way diﬀerent from the . The bars depict the corresponding regimes of the driving frequency Ω. This function roughly obeys a power law [34. the ﬁrst resonance (the one for which the periods of the driving and of the unperturbed oscillation are in a ratio of 1:1) is prominent among the others and soon (in terms of increasing amplitude F ) exceeds the size of the “orderzero” regular areas near the bottom of each well [16]. Since the period of the unperturbed.
58 The harmonically driven doublewell potential − c − r + r (a) − r (b) − c (c) (d) Figure 6. For the parameters chosen in our numerical studies. in both directions. the familiar way tunneling fades out in the presence of dissipation is also signiﬁcantly altered.” Depending on temperature. See Section 6. the borderline between the chaotic layer along the former separatrix and the regular regions within and outside the wells is quite sharply deﬁned. it can also be destroyed on a much shorter time scale. For the tunneling dynamics. in turn.2. Singletdoublet crossings.8. below) on a scale of the chosen eﬀective quantum of action. Diﬀerent line types signify diﬀerent parity. As a consequence. the coherent dynamics can last much longer than for the unperturbed doublet. This gives rise to additional singletdoublet and even to doubletdoublet encounters. 6. 6. 6. the order of the regular doublet is restored in passing through the crossing. The corresponding doublet ladder therefore moves in the spectrum independently of the doublets that pertain to the main regular region. 6. The “coastal strip” formed by hierarchies of regular islands around higher resonances remains narrow (cf. Fig. higher resonances are negligible in size. despite the presence of the same dissipation as outside the crossing. Fig. below). Near a crossing.1 for the labeling of the levels. Note that only for conﬁgurations (a). from the typical tunnel splitting (cf. drastically change the nondissipative quasienergy scales and replace the twolevel by a threelevel structure. and of the chaotic singlets. This is reﬂected in timedomain phenomena ranging from the suppression of tunneling to a strong increase in its rate and to complicated quantum beats [31–33].3: Possible conﬁgurations of quasienergy crossings between a chaotic singlet and a regular doublet. In conﬁgurations (c).2. Therefore.6.2 Chaotic tunneling near singletdoublet crossings Near a crossing. 103] is therefore not signiﬁcant in our studies.1 Threelevel crossings Among the various types of quasienergy crossings that occur according to the above scenario. level separations deviate vastly. the role of states located in the border region [102.(b).(d). those involving a regular doublet and a chaotic singlet are the most com . it is reversed. main regular region. establishing “chaosinduced coherence.
As the third player. (6.18) φR. respectively. The global relative phases can be chosen such that the r r superpositions 1 − (6. a small splitting of the mean − + − ± energies pertaining to the regular doublet. and tunnel back and forth with a frequency ∆/ given by the tunnel splitting in the presence of the driving.16) . r − ψr (t) + (6. residing on a pair of quantizing tori in one of the regular (subscript r) regions. Without loss of generality. and for later reference in the context of the incoherent dynamics. so that its timeperiodic part φ− (t) c contains a large number of harmonics. We have assumed that the quasienergy splitting (as opposed to the unperturbed splitting) is − − + = ∆ > 0.21) . (6.17) =e −i( + r +∆)t/ φ− (t) r with even (superscript +) and odd (−) generalized parity. we have assumed that c = + +∆+∆c .20) For the singletdoublet crossings under study. In order to give a quantitative account of such crossings and the associated coherent dynamics. Neglecting the coupling with all other states. c + (6. Er − Er  Ec − Er . we suppose that r c there is a nonvanishing ﬁxed matrix element b≡ φ− HDW φ− r c > 0.6. [30]. In order to model an avoided crossing between φ− and φ− . Far to the left of the crossing. we introduce a Floquet state − ψc (t) = e−i( r +∆+∆c )t/ φ− (t) . r where ∆c  can be regarded as a measure of the distance from the crossing. we shall now discuss them in terms of a simple threestate model. The structure of the classical phase space then implies that the mean energy of the chaotic state should be close to the top of the barrier and far above that of the doublet.19) located mainly in the chaotic (subscript c) layer. we typically ﬁnd that ∆ b Ω.L (t) = √ φ+ (t) ± φr (t) r 2 are localized in the right and the left well. H3s = + + 0 ∆ b r 0 b ∆ + ∆c (6. we expect the following situation: There is a doublet of Floquet states + ψr (t) = e−i r t/ φ+ (t) . like for the quasienergies. We assume. devised much in the spirit of Ref. we model the system by the threestate (subscript 3s) Floquet Hamiltonian 0 0 0 . its generalized − parity is ﬁxed to be odd. respectively.2 Chaotic tunneling near singletdoublet crossings 59 mon. For the quasienergy.
Note that in the real crossing. Eq. This numerical example also shows that the idealized threestate model is not always strictly correct. while E0 remains unaﬀected (Fig. 6. Following the global tendency of widening of the splittings with increasing driving amplitude [16. 6. Eq. r c − − − φ2 (t) = φr (t) sin β + φc (t) cos β . The phasespace structure of the participating Floquet states (Figs. A comparison of the appropriately scaled threestate theory (Fig. we have numerically studied a singletdoublet crossing that occurs for the doublewell potential. as outlined above.1). (6. so that the exact crossing occurs to the left of the avoided one.60 The harmonically driven doublewell potential in the threedimensional Hilbert space spanned by {φ+ (t) . Therefore. 6. − 1.24). ∆c 2 For β → π/2. it may well happen that even far away from a crossing. 114]. The mean energy is essentially determined 2 1 c r − − by the phasespace structure. φ− (t) }. 0<β< . Its r r c Floquet states read φ+ (t) = φ+ (t) . To the far right of c 2 r the crossing. 6. since they pertain to states with opposite parity (cf. we obtain 2b π (6. c (6. 6. (6.3a. The angle β describes the mixing between the Floquet states φ− and φ− and is a measure of r c the distance to the avoided crossing.5) shows satisfactory agreement. The − quasienergies + and − must intersect close to the avoided crossing of − and 2 0 1 1 (Fig. Their quasienergies are + 0 (6.6.4a). 6. In order to illustrate the above threestate model and to demonstrate its adequacy. 6. with D = 4. i. Their crossing is exact. at a driving frequency Ω = 0.25) 2β = arctan . 0 r − φ1 (t) = φ− (t) cos β − φ− (t) sin β .e.5).23) and the mean energies are approximately given by + + E0 = E r .2 = + r 1 + ∆ + ∆c 2 1 2 ∆2 + 4b2 . − − − E1 = Er cos2 β + Ec sin2 β. cf.7) meets the assumptions of our threestate theory. there is also an exchange of E1 and E2 + in an exact crossing. 6. the exact eigenstates φ− and φ− have 1 2 interchanged their identity with respect to the phasespace structure [31–33].21). we retain the situation far left of the − crossing.015029 (Fig.24) where contributions of the matrix element b have been neglected. corresponding to −∆c b. the quasienergy of the chaotic singlet decreases as a function of F . with φ1 ≈ φ− . By diagonalizing the Hamiltonian (6..4b). we have φ− ≈ φ− and φ− ≈ φ− .22) = + r . φ− (t) .4) with this real singletdoublet crossing (Fig. − − − E2 = Er sin2 β + Ec cos2 β. . for β → 0 or ∆c b. φ− ≈ φ− . (6. Fig. 34.982 ω0 and amplitude F = 0.b). Here.
Eq.001 0.4: A singletdoublet crossing.0 − 1 − ( − + 0 0.9 are marked by dotted vertical lines.015 F 0. the energies for the case with coupling by full lines for even and dashed lines for odd states. respectively. Bold lines give the mean energies of the chaotic singlet and the groundstate doublet depicted in panel (a). Full and dashed lines indicate energies of even and odd states.014 α a b c 0. 6.016 0. Values of the driving amplitude used in Fig.015 F 0.6. Unperturbed energies are marked by dotted lines. according to a threestate model (6. (6. at D = 4 and Ω = 0. in terms of the dependence of the quasienergies (a) and the mean energies (b) on the driving amplitude F . 0. .1).982 ω0 .001 (a) + h 0 )/¯ ω0 (b) 2 1 − 2 − E2 − E1 Eα /¯ ω0 h − + 0 0 1 2 3 doublets + E0 0.2 Chaotic tunneling near singletdoublet crossings 61 4b quasienergy (a) 2b ∆ + 0 − 2 (b) mean energy − Ec − E2 2b 0 2b ∆c 0 − 1 − Er − E1 4b 10 5 0 ∆c /b 5 10 D 10 5 + E0 0 ∆c /b 5 10 Figure 6.014 0.5: Singletdoublet crossing found numerically for the driven double well.21) in terms of the dependence of the quasienergies (a) and the mean energies (b) on the coupling parameter ∆c /b.016 Figure 6.
.. .... ... .... .. The driving parameters F = 0.. ... .... ........ . .......5 (b) 0.. ..... . ... .. . ....... .015. ... ..... Ω = 0..... .. ........ ......... ... .. .............. ... Ω = 0. .. . ....... ...... .... ..... ..... .... .. .... .... .............................. ... . .... ... . .. .. .... ... .... . .. . ........ .... ... ... .... .. . ...... . .. .. .. ................... .....5 0. ...... ............ .. ... ... ..... ... ... ..... ... .. .. .. ............ ..... .. ..... ....... .... . . .. ..... . ..... .... . . ... .. . .. ... .... ....... ... . .. . .......... ... . . ..... . .. . ..0 0. ... . .. ....... . ....982 ω0 .. .......... . . .. ... ..... Eq.. ..... ... . .. .... ... ... ......... ...... .. .... .. . ..... .. ....... .. . . ..5 1.. ..... . .... .. .5 0.. ...... . ........ . ... ......... .. ........ at t = 2πn/Ω. .... ....... .......1). .... .. .. .... ...... .. . .... .. ....... .............. .. ............ .. .... .... ...... ... .. . .... ... . ... .. .. .. . .. ...... . .. ......5 0. ........... . ...5 x/x0 1. ...... ... ....... .. ... .......... ..5 p/mω0 x0 0. . .. .... . . . ......... ... . .. .. .... .. ............. . .... . ......... ............ .. . .. ..... ..... .. ... .... .... ............... . . ... . ....... .....6: Stroboscopic classical phasespace portraits. .. .... ........... ........... .0 0..... .. ......... ...... . .... . . ..... ..... ........... .... ..... ..... ...... ........ ...... ... .. The driving parameters F = 0.... . .. ... ... .. .... ... . ......... .. .. ... ........ .... .. . ....... . ...... . .............. . . ..... .. 1... ..... . ... .. . ... ... . ...... ... ... . .......... .... ....... ........ . Eq. ............. .... .... . .... . .. ..... .. . . ..... . ........ . ............ . ... ... . .... .. .. . .... ...... . ... .. .. .... ...... .. ... ... .... . ... . . .. . of the harmonically driven quartic double well. ... ..... . ........... .. .. . ..... ....... ... . . ... .................. . .. .... .. . .0 0. ......... ........ .. ...... ... ..... ... ..... ... . ..... .. ..... .. . ... .... .. .. . are chosen at the the center of the singletdoublet crossing under study... . . ... ..... .. . ...... ...... ....... .. . . .. . ... ....... . .......... .... . . . .. . . . .. . . .... ..... ... .. .. .... ... . . ... ... ...... ... .. .... . ........... .. .. .. . ....... ...... .. . ........5 x/x0 1. ..... . ...... ... ... ....... .. . .. ........ . ......... ....0 1.. .. ..... ........ .. .. ... ........... ........ ......... ....0 0. ... .. . ... .. .. . . ... ....... . .... .. ... . .. . 1....... .... ........ ... .. . ........ .. . .... ..... . . . ...... ... . ...... . ........... .. .. ..... ..... .. ...... .....5 1.. . ..... . ..... .. . ... ........ .. .. ..... ... ........ .... . ..... ...... .... . ... .... . .... .. ... .. ... ... .. ........ ..............5 .... .. ..... .. ..... .. . ...... . .. ...... . ...... ... ..0 1..... . . . ....... ..... .... .... ....... .. .....0 0. ...... .. . . ... ... . ....... .. . ... ........ ....... . .. ..... . ....... .. .. .. ...... .. .......... .... ... ...... ...... .... . ..... .. ... .. ..... . . ...... . . .... ........... .. .......... .. ........... ... . . ........ (6... .. .. .. ... .... ........ .. ...... .5 .. ... .. ... .......... . ........... .. . . ... .. . .... .. ... .. ...... ....... ........... . ...... ....... . .. .. .. ......... .... ... ....... ... ... ... . .. .. ... .. . ..... .. .0 0..... ... are in suﬃcient distance to the singletdoublet crossing such that the mixing between the regular and the chaotic state is negligible......... ..... ... ..... . ....0 0. . . ..982 ω0 . .... ... . . ... .... .. ..... .... ... .... .. ...... ..................... . ... . ... .. .. ........ ... . .014. .. .......... ..... .... .. .. .... . .... .. .. . . ...7: Contour plots of the Husimi functions for the Floquet states φ− ≈ φ− (a) r 1 and φ− ≈ φ− (b) of the harc 2 monically driven quartic double well.. ......... . Figure 6. .. . .... .. ... ..... . ............. . . .. .... .. ..... ....5 (a) 0. ... ..... .... . . . ....... .. .... . . . .5 1.. ............ .. .. ........ .. ..... ... .. ... .. . ..... . ... . .....0 0... ........... . ... . ... . . .... . . .. ............. ... ..... ..... ...... ...... . ...... ...... ... ... ..... . .... ... . ..62 The harmonically driven doublewell potential 0.... . .. .. . .. ..0 0.... at stroboscopic times t = 2πn/Ω... ..5 p/mω0 x0 0....... ....... .... ... . . ..... ...5 1..... ... . . .. . .. ...... .0 1. .... .1)... .......... ..... . (6.... . .......... . .... . ..... . .. ...... .. . . .... . The rectangle in the lower left corner depicts the size of the eﬀective quantum of action eﬀ . . ..... ... . . . .. . ....... . ....... .. .. ............ . . .. .. . ....... ....... ...... .. .... . ... ..... .... .. . . .. ..... ......... ...... .. .5 Figure 6..... ..... .... ... .. . .. . . .... .... ........... ..... .. ...5 p/mω0 x0 0..... ....... ..... ............... ...... .... ....... ......5 x/x0 1.. .. ... .... ... . . . . ...... . ... . .. .. ..... . . ..... . .... ...... ........... .... .... . .. . ......... ...... ....... .... ..... ...... .
3a. In that case. or φc . the doublet splitting does not exactly return to its value on the opposite side (see Fig.01 0.0 0.3c. It is even possible that an exact crossing of + and − does not take place 0 1 at all in the vicinity of the crossing. The arrows indicate the locations of the exact and the avoided crossing within a threelevel crossing of the type sketched in Fig. 6. it corresponds to the decomposition of φR in the basis (6. To study the dynamics of the tunneling process. Therefore.025 F Figure 6.26) 0 1 2 2 It is constructed such that at t = 0. 6. Nevertheless. PL (t) =  φL (t)ψ(t) 2 ( −− 1 1 − cos 1 = 2 + 0 )t + 0 )t sin2 β (6. (6. the above scenario captures the essential features. respectively.18).22) at ﬁnite distance from the crossing.982 ω 0 .27) cos2 β − cos ( − 2 − + 0 )t sin2 β .8).d). we focus on the state + − − 1 ψ(t) = √ e−i 0 t/ φ+ (t) + e−i 1 t/ φ− (t) cos β + e−i 2 t/ φ− (t) sin β . to be PR (t) =  φR (t)ψ(t) 2 1 ( − − + )t ( −− 0 = 1 + cos 1 cos2 β + cos 2 2 ( − − − )t 2 + cos 1 − 1 cos2 β sin2 β .2 Chaotic tunneling near singletdoublet crossings 63 10 2 n=2 3 10 10 avoided n=1 n=0 4 5 ∆n /¯ ω0 h 10 10 6 7 8 10 exact 10 10 9 0. we ﬁnd the probabilities for its evolving into φR .015 0.8: Splitting of the lowest doublets for D = 4 and Ω = 0. the relation of the quasienergies in the doublet gets reversed via the crossing (Fig.02 0. (6. it is initially localized in the regular region in the right well and follows the time evolution under the Hamiltonian (6.6. 6. From Eqs.21).005 0. φL .22). (6.
64
The harmonically driven doublewell potential
1.0 0.8
P (tn ) (a) PR PL
0.6 0.4 0.2 0.0
5
Pc
0
10 2 10 tn = 2πn/Ω [1/ω0 ]
5
3 10
5
1.0 0.8
P (tn )
(b)
0.6 0.4 0.2 0.0
5 5 5
0
10 2 10 tn = 2πn/Ω [1/ω0 ]
3 10
1.0 0.8
P (tn )
(c)
0.6 0.4 0.2 0.0
5 5 5
Figure 6.9: Stroboscopic time evolution of a state initially localized in the right well, in the vicinity of the singletdoublet crossing shown in Fig. 6.5, in terms of the probabilities to be in the right well (which here is identical to the return probability, marked by full lines), in the reﬂected state in the left well (dashed), or in the chaotic state ψc (dotted). Parameter values are as in Fig. 6.5, and F = 0.0145 (a), 0.0149 (b), 0.015029 (c).
0
10 2 10 tn = 2πn/Ω [1/ω0 ]
3 10
6.2 Chaotic tunneling near singletdoublet crossings + cos (
− 1
65
−
− 2 )t
− 1 cos2 β sin2 β , cos2 β sin2 β.
Pc (t) =  φc (t)ψ(t) 2 ( −− = 1 − cos 1
− 2 )t
We discuss the coherent dynamics of the threestate model for diﬀerent distances to the crossing and illustrate it by numerical results for the real crossing introduced above. In suﬃcient distance from the crossing, there is only little mixing between the regular and the chaotic states, i.e., sin β 1 or cos β 1. The tunneling process + then follows the familiar twostate dynamics involving only φr and φ− , with r tunnel frequency ∆/ (Fig. 6.9a). Close to the avoided crossing, cos β and sin β are of the same order of magnitude, and φ− , φ− become very similar to one another. Both now have support in the 1 2 chaotic layer as well as in the symmetryrelated regular regions and thus are of a hybrid nature. Here, the tunneling involves all the three states and must at least be described by a threelevel system. The exchange of probability between the two regular regions proceeds via a “stopover” in the chaotic region [15, 30–33]. The three quasienergy diﬀerences that determine the time scales of this process are in general all diﬀerent, leading to complicated beats (Fig. 6.9b). However, for ∆c = −2∆, the two quasienergies − − + and + − − are de1 0 0 2 generate. At this point, which marks the center of the crossing, the number of diﬀerent frequencies in the threelevel dynamics reduces to two again. This restores the familiar coherent tunneling in the sense that there is again a simple periodic exchange of probability between the regular regions [31–33]. However, the rate is much larger if compared to the situation far oﬀ the crossing, and the chaotic region is now temporarily populated during each probability transfer, twice per tunneling cycle (Fig. 6.9c).
6.2.2
Dissipative chaosassisted tunneling
The crucial eﬀect of dissipation on a quantum system is the disruption of coherence: a coherent superposition evolves into an incoherent mixture. Thus, phenomena based on coherence, such as tunneling, are rendered transients that fade out on a ﬁnite time scale tdecoh . In general, for driven tunneling in the weakly damped regime, this time scale gets shorter for higher temperatures, reﬂecting the growth of transition rates [53]. However, in the vicinity of an exact crossing of the groundstate quasienergies, the coherent suppression of tunneling [10, 12, 107] can be stabilized with higher temperatures [76–78] and increasing friction [57, 58] until levels outside the doublet start to play a role. We have studied dissipative chaosassisted tunneling, using again the real singletdoublet crossing introduced in Sec. 6.2.1 (see Fig. 6.5) as our working example. The time evolution has been computed numerically by iterating the dissipative quantum map (4.31) for the improved master equation in
66
The harmonically driven doublewell potential
1.0
tr ρ2 (tn )
0.8 0.6 0.4 0.2 0.0
(a)
1.0
P (tn )
0.98 0 0 10
5 5 5
2 10
4
2 10 3 10 4 10 tn = 2πn/Ω [1/ω0 ]
5
5 10
5
1.0
tr ρ2 (tn )
(b)
0.8 0.6 0.4 0.2 0.0
5 5 5 5 5
0
10
2 10 3 10 4 10 tn = 2πn/Ω [1/ω0 ]
5 10
Figure 6.10: Occupation probabilities as in Fig. 6.9a,c, but in the presence of dissipation. The dashdotted line shows the time evolution of tr 2 . The parameter values are D = 4, Ω = 0.982 ω0 , γ = 10−6 ω0 , kB T = 10−4 ω0 , and F = 0.0145 (a), 0.015029 (b). The inset in (a) is a blow up of the rectangle in the upper left corner of that panel.
P (tn ) tr ρ2 (tn )
1.0 0.8 0.6 0.4 0.2 0.0
6 7
Figure 6.11: Time evolution of the return probability PR (full line) and the coherence function tr 2 (dashdotted) during loss and regain of coherence. The parameter values are as in Fig. 6.10b.
PR (tn )
0
5 10 tn = 2πn/Ω [1/ω0 ]
10
For the relaxation towards the asymptotic state.10). we approximate tdecoh by the decay rate of tr 2 . The time scale trelax of the approach to the asymptotic state is given by the reciprocal of the smallest real part of the eigenvalues of the dissipative kernel. a measure of coherence (see Appendix B. also the slower transitions within doublets are relevant. i. this attractor remains time dependent but shares all the symmetries of the central system.27) 2 1 and Fig. The high mean energy of this singlet results in an enhanced decay of coherence at times when φc is populated (Fig. periodicity and generalized parity. 6. In general.3) is dissipative. This means that we have eliminated the explicit time dependence of the attractor by representing it in the Floquet basis and introducing a mild rotatingwave approximation. Eq. averaged over a time tp . we focus on the . n should be so large that the coherence decays substantially during the time tp (in our numerical studies to a value of approximately 0.11). are time independent and so the asymptotic solution also is. 6. the coeﬃcients of the master equation (4. the chaotic singlet induces an exact crossing of the groundstate quasienergies (see Fig. i. At the center of the avoided crossing. valid within the moderate rotatingwave approximation. here. the tunnel splitting increases signiﬁcantly—the essence of chaosassisted tunneling. At F ≈ 0.2.12). we have chosen the density operator (0) = φR φR . 6.3 Asymptotic state As the dynamics described by the master equation (4. (4.9. For this procedure to be meaningful. 6. 1 tdecoh d 1 tp dt =− tr 2 (t ) tp 0 dt 1 tr 2 (0) − tr 2 (tp ) . the chaotic singlet becomes populated periodically with frequency  − − − / . we have chosen the propagation time tp as an nfold multiple of the duration 2π / − − −  of the chaotic 2 1 beats. Therefore. This indicates that transitions from states with mean energy far above the ground state play a crucial role. Eq.e. a state localized in the right well.8). the corresponding time scale trelax can be much larger than tdecoh (Fig. = tp (6.28) (6.22) for the matrix elements αβ . Outside the singletdoublet crossing we ﬁnd that the decay of coherence and the relaxation take place on roughly the same time scale (Fig. the decay of coherence becomes much faster and is essentially independent of temperature.6. it converges in the longtime limit to an asymptotic state ∞ (t). 6.22).29) Because of the stepwise decay of the coherence (Fig.2). resulting in a stabilization of coherence with increasing temperature. However.013.e.10).2 Chaotic tunneling near singletdoublet crossings 67 moderate rotatingwave approximation.9). cf. 6. (6. 6. To obtain quantitative estimates for the dissipative time scales. To gain some qualitative insight into the asymptotic solution. During the tunneling. In the vicinity of a singletdoublet crossing. As an initial condition.
It describes two kinds of thermal transitions: decay to states with lower energy and. whereas at the center of the avoided crossing (F ≈ 0. They give the rates of the direct transitions from φα to φα .013. Thus Lαα.982 ω0 . In the case of zero driving amplitude. The only nonvanishing Fourier component is then cα. The ratio of the direct transitions forth and back then reads ( α− α) Lαα.αα kB T . dashed vertical line) the decay of coherence is accelerated.02 10 [1/ω0 ] 8 (b) kB T = 10−4 hω0 ¯ kB T = 10−3 hω0 ¯ kB T = 10−2 hω0 ¯ 10 trelax 7 10 6 Figure 6.012 0.α α only consists of a single term proportional to N ( α − α ).28 and 4.014 F 0.01 0. α=α. Ω = 0.30) of the dissipative kernel. (6. Within the full rotatingwave approximation. Near the exact crossing (F ≈ 0.02 diagonal elements Lαα.012 0.018 0. these are the only nonvanishing contributions to the master equation which aﬀect the diagonal elements αα of the density matrix.α α = 2 n Nαα .29. thermal activation to states with higher energy.0 .n Xαα . temperature as given in the legend.α α = exp − Lα α . the Floquet states φα reduce to the eigenstates of the undriven Hamiltonian HDW .01 0.68 The harmonically driven doublewell potential 10 [1/ω0 ] 8 (a) kB T = 10−4 hω0 ¯ kB T = 10−3 hω0 ¯ kB T = 10−2 hω0 ¯ tdecoh 10 7 10 6 0.016 0. given in Eqs. and the quasienergies α reduce to the corresponding eigenenergies Eα . The parameter values are D = 4. 0.31) . (6.n 2 . 4.014 F 0.016 0. if the energy diﬀerence is less than kB T .12: Time scales of the decay of the coherence measure tr 2 (a) and of the relaxation towards the asymptotic solution (b) near the singletdoublet crossing.015. full vertical line) coherence is stabilized.018 0. γ = 10−6 ω0 .
its coherence tr 2 (see Appendix B. both should behave similarly.22). It equals unity only if the attractor is a pure state. The longtime limit of the corresponding classical dynamics converges to one of two limit cycles. the system tends to occupy the ground state only.13a). 6. Thus a decay to states with “higher” quasienergy (recall that quasienergies do not allow for a global ordering) becomes possible due to terms with n < 0. In the vicinity of a singletdoublet crossing the situation is more subtle. 1 2 0 if the temperature is signiﬁcantly above the splitting 2b of the avoided crossing. thermal activation from φ+ to φ− . becomes possible. we have  α− α Ω. as a consequence of symmetry. alternatively.α α is given by a sum over contributions with quasienergies α − α + n Ω. 6. In a stroboscopic map they correspond to two isolated ﬁxed points. the splitting is the main diﬀerence between the two partners of the quasienergy doublet. An important global characteristic of the asymptotic state is its Shannon entropy S = − tr( ∞ ln ∞ ) or.32) In particular. one expects an equal population of the doublets even in the limit of zero temperature (Fig. In particular. According to Eq. Due to the high mean energy − Ec of the chaotic singlet.6. where the splitting has no signiﬁcant inﬂuence. Here.. accompanied by depletion via the states below 0 1. However. According to the above scenario. The eﬀects under study are found for a driving with a frequency of the order of unity. the odd partner. Eq. in a regime with ∞ strong driving but preserved doublet structure. This behavior is qualitatively diﬀerent from the asymptotic limit of the dissipative quantum dynamics near the center of the crossing and shows that the occupation of the levels outside the singlet and the doublet at asymptotic times is a pure quantum eﬀect. Thus asymptotically.13b.2). reﬂecting the incoherent population .αα is dominated by contributions with n < 0.e. The ∞ value of the latter gives approximately the reciprocal of the number of incoherently occupied states. In the limit kB T → 0. This is in contrast to the undriven case. close to the top of the barrier.c). (6. Thus φ− and φ− are depleted and mainly φ+ will be populated. with respect to dissipation. (6. of the doublet mixes with a chaotic singlet. i. the system tends to occupy Floquet states comprising many Fourier components with low index n. Physically. However. Therefore. For a strong driving. these states have low mean energy. the occupation probability decays monotonically with the energy of the eigenstates. and thus acquires components with higher energy. say. they describe dissipative transitions under absorption of drivingﬁeld quanta. each of which is located close to one of the potential minima. and Lα α .2 Chaotic tunneling near singletdoublet crossings We have detailed balance and therefore the steadystate solution αα 69 (∞) ∼ e− α /kB T δαα . far oﬀ the threelevel crossing. cf.42).2 − Ec . (2. all these states become populated in a steady ﬂow (Fig. each Floquet state φα contains a large number of Fourier components and Lαα. we expect tr 2 to assume the value 1/2. the decay back to the ground state can also proceed indirectly via other states with mean energy below − Ec . Thus for a quasienergy doublet. Correspondingly.
0. γ = 10−6 ω0 .015029 (c). 0 1 2 3 4 5 6 7 8 9 10 1112 1314 15 Floquetstate index α .0145 (b). temperature as given in the legend. and F = 0.70 The harmonically driven doublewell potential 10 0 φ+ 0 φ− 1 (a) kB T = 10−4 hω0 ¯ kB T = 10−3 hω0 ¯ kB T = 10−2 hω0 ¯ 10 αα 1 φ− 2 10 2 10 3 0 1 2 3 4 5 6 7 8 9 10 1112 1314 15 Floquetstate index α 0 10 φ+ 0 φ− 1 (b) kB T = 10−4 hω0 ¯ kB T = 10−3 hω0 ¯ kB T = 10−2 hω0 ¯ 10 αα 1 φ− 2 10 2 10 3 0 1 2 3 4 5 6 7 8 9 10 1112 1314 15 Floquetstate index α 0 10 φ+ 0 φ− 2 (c) kB T = 10−4 hω0 ¯ kB T = 10−3 hω0 ¯ kB T = 10−2 hω0 ¯ 10 αα 1 φ− 1 10 2 10 3 Figure 6. The parameter values are D = 4.13: Occupation probability αα of the Floquet states φα in the longtime limit. Ω = 0.982 ω0 .013 (a). 0.
0.018 0.15: Coherence of the asymptotic state in the vicinity of a singletdoublet crossing for F = 0.018 0.6.01 0.5 0.14: Coherence (a) and Shannon entropy (b) of the asymptotic state in the vicinity of a singletdoublet crossing for diﬀerent temperatures as given in the legend.0 kB T = 0 kB T = 10−4 hω0 ¯ kB T = 10−3 hω0 ¯ kB T = 10−2 hω0 ¯ 0.6 0. The other parameter values are D = 4.982 ω0 .02 3 kB T = 0 kB T = 10 −4 (b) hω0 ¯ kB T = 10−3 hω0 ¯ 2 S kB T = 10−2 hω0 ¯ 1 0 Figure 6.0 1.8 tr ρ2 ∞ 0.0 (a) 0.0 10 6 10 5 10 10 T [¯ ω0 /kB ] h 4 3 10 2 10 1 Figure 6. Ω = 0.982 ω0 .014 F 0.0 (b) 0.012 0.016 F 0. Ω = 0.014 0.02 1.012 0.01 0.013 (a) and F = 0.2 0. and γ = 10−6 ω0 . tr ρ2 ∞ .0 (a) tr ρ2 ∞ 0. and γ = 10−6 ω0 .015029 (b): exact calculation (full line) compared to the values resulting from a threelevel description (dashed) of the dissipative dynamics.016 0.5 0.4 0.2 Chaotic tunneling near singletdoublet crossings 71 1. The other parameter values are D = 4.
for the case of regular classical dynamics.3.3 Signatures of chaos in the asymptotic state In recent work it has been demonstrated that a phasespace representation of quantum mechanics. the threestate model results in a completely diﬀerent type of asymptotic state (Fig. 6. 6. its value should be close to unity for temperatures kB T 2b and much less than unity for kB T 2b (Figs. it is important to take a large set of levels into account. This phenomenon amounts to a chaosinduced coherence or incoherence.15). The failure of the threestate model in the presence of dissipation clearly indicates that in the vicinity of the singletdoublet crossing. The crucial role of the decay via states not involved in the threelevel crossing can be demonstrated by comparing it with the dissipative dynamics including only these three levels (plus the bath). the phasespace representation of the asymptotic state of a dissipative quantum map exhibits the structures of the corresponding classical attractor [106]. like the Husimi or Wigner distribution. m (6. Moreover. and a loss of coherence for temperatures above the splitting.30. reveals the structures of the corresponding classical phase space [5. it has support in the whole chaotic layer. we have S ≈ ln 2 and at the center of the crossing the entropy exhibits a signiﬁcant temperature dependence. Thus outside the crossing. the analogies have their limitations due to the Heisenberg uncertainty principle which does not allow for arbitraryly ﬁne phasespace structures for a quantum system and results in coarsegraining over a “phasespace unit” 2π .5) and obtain x= ˙ 1 p. the Husimi function of an eigenstate (or of a Floquet state if the system is driven) is localized in phase space along the corresponding quantizing torus. In particular. assumes approximately the value ln n for n incoherently populated states. 6. In the vicinity of the singletdoublet crossing where the doublet structure is dissolved. the motion changes from regular to chaotic. we add an Ohmic friction force Fγ = −γp to the conservative equations (6.14b). The asymptotic classical dynamics of the driven dissipative doublewell potential is for suﬃciently strong driving particularly sensitive to the friction strength: With decreasing friction.14a. At the crossing. The corresponding Shannon entropy (Fig. This means that the crossing of the chaotic singlet with the regular doublet leads to an improvement of coherence if the temperature is below the splitting of the avoided crossing. one is even able to classify quantummechanical states as regular or chaotic according to their localization in phase space [120].4).117–120].72 The harmonically driven doublewell potential of the groundstate doublet. However. 6.1 Classical attractor To describe the classical dissipative dynamics of the driven double well. for chaotic motion. 6. 6. respectively. If the classical dynamics is mixed.15). (6.33) .
it may even happen that the dissipative dynamics is chaotic and the attractor possess fractal geometry. however. the dynamics is asymptotically conﬁned to an attractor. For periodically driven dissipative systems.5 0. the attractor is in general also timedependent with the period of the driving and is properly rendered by its stroboscopic map [121–123]. (6. (6.9 ω0 .34) As friction always decelerates a particle. we obtain an exponential contraction of a phasespace volume V —a constituent feature of dissipative ﬂows. The type of geometry can be characterized as fractal or regular according to its Hausdorﬀ dimension dH which is deﬁned by the scaling assumption N ∼ l−dH . γ/ω0 p = −γp − ˙ ∂V (x.3 Signatures of chaos in the asymptotic state 2. Therefore. forming a socalled strange attractor. thus destroys the timereversal symmetry (6.33).0 1. l → 0. . an attractor consists of limit cycles or isolated ﬁxed points. a formation in phase space with zero volume to which all suﬃciently close trajectories from the socalled basin of attraction converge for long times.0 0.36) Here. It evolves by having each point on its surface ∂V follow an orbit generated by (6.09. Accordingly. N is the number of squares with width l needed to cover the whole attractor. 6.6. For suﬃciently weak dissipation. Fig.0 5 10 3 2 5 10 2 2 5 10 1 2 5 10 0 Figure 6. which yields by the divergence theorem [3] dV = dt dx dp V ∂x ∂p ˙ ˙ + ∂x ∂p = −γV.6). ∂x (6. dissipation breaks the reﬂection symmetry at the xaxis of phasespace portraits which we found for the chosen initial phase of the driving (cf. it distinguishes between future and past. The lack of timereversal symmetry in presence of friction is even more evident from the time evolution of a volume element V of phase space.34).10) of the conservative system.35) Thus. Ω = 0.5 dH 73 1. Depending on the values of the driving parameters and the friction strength. t) . (6.16: Hausdorﬀ dimension of the classical attractor for F = 0.
. ... ..... . . . . . .. .... ... . .. ... .. .. .. . .. ...... .. . 0.... .... . . .. ..... . . . .. .... . . . .. ... .. ... . .. .. . .. .. .... . . ... . ... ... ... .. . . . .... .. ... ............. ... .. ......... .. . . . . .... .. ... .. .... . . .... . .. ..5 0... .. ..... .... ...... .. .. .. .. ... . .. . ..... . .. . ..... .. .. ....... . ....... . ... .... .... . ...... .. . ...... . ..... ... .. . . .5 1... . .. . ... ... .. . ...5 . .. . .... ... ..... .. ... . .... . . ...5 . . .. .. .... .. .... .. ..34)... ... .. ..... . . .. .. ...... . .... . ..... . .. .. ...... ......5 0. .. .. ....... ....... ... . . .. ... ... .. .. . .. .... .... .... .... . ... . .... ..... . ...... . .. .... ..... . ... .. ..5 0.... .. . . ..... .. ..... .. . . .... .... . . . ... . .. . ..... . ...... ... ... . .. ... ............. . .. .... ..5 1. ...... ... .. . . .... . .. (6... .. ....... .. ... . .. . ... ...... . ..........33).. . . .. ..... ... .... . ...... .. .. .. ... .... ... ........ . .. . ... . . ... . ... .. .. ..... . . .... ........... ...... . .. .... ... . . . . ... ... . .... . . . .. . ...... . ...... .. ..... ....... . .. ... Figure 6.. . .... .5 0. . .. . ... ... . ... . .. . .. . . .. ... .... . ....... . ... .5 p/mω0 x0 0. ....... . ... . . ... . .... . ..... .. .. . ... .... . . .. . ..... ..... . ...0 0..... .. . . ... ...... .. .... . . .. . .. .... .. .0 x/x0 0. .. ..... ... .. .. .... . .. ... .. . ... .... . . .. ... ........... .. . .... . .. ... ...... ..... .... ... .. ..... ...... .5 0. .3 ω0 (a)...... .. . .... ... . ....... ... . . .. .. ... ...... .. .. ...74 The harmonically driven doublewell potential (a) 0. .. . ... ..... .. .. . Eqs........0 0. .. .. . . . ..... . ..... .. .. .17: Stroboscopic classical phasespace portrait at t = 2πn/Ω...... ..... ... . ....... ... 0..... ... . . .. .... ... . . .. ..... .. ..... ....... ..... . .... . . . .. . .....2 ω0 (b). . . .. . .... .. .. ... . ... . .. .. .. . .. .. . . . .... ..... . ....... . .. ...... ... for the driving amplitude F = 0.. .. . ... ... ... . ... . . ....... . ..5 p/mω0 x0 0.. . .... ..03 ω0 (c). of the dissipative harmonically driven quartic double well... . . . . . . .. ..... In panels (a) and (b) the stroboscopic portrait is marked by a full dot and the broken lines show the corresponding limit cycles.. .....0 1. .. . . .. . . ...... . 1........ ...... .. ..... . . . ...0 0. .... . . ......... . ... . .. . . .. .. . . . ... .. ............ .. ....... ... . . .5 . .. ... .. .. . .. . ...... . .. .. ... . .... . ...... .. ... . .. .... . . .. ... ...5 1. ... ... . . . . . .. . .. ...... .... .... . .... .. ...... .. ... .. .... . . . ........ .... .. ... . .. ..... . . .. . .... .. . ....... ..... ..... . . . . .... 1. . ... ........... . .. . . . .. ..... . ... . ...... . . ..... .. . .. ... .. ... . .... ..... ..... . .9 ω0 .0 1.. ..... .. . ....... .. . ..... .... .. .... . . . . . ..... .. ....... . 1. . ..0 1..... .... .. ....0 x/x0 0.. ..... .. .. .. ... .. .. ... ... ... .... . . . . . .. . ... .. .... .. ... .. . .. 0. (6... . . .. . ... . . . .... . ... ........ ... . ...5 (c) . .. .. . .. .. . . . .... ... .... ... .. ... .. . . . .... .... ... ......... ...5 1. .. ...... .. . . . . .. . . ..... . ... .. . ... ..5 1. . . . . . .. ... .... .. . .... . . .. ....... .. ... .... . .. ... ... . . .. .. ..5 p/mω0 x0 0.... . ...... ...... . . .... . .. ..0 ....0 x/x0 0........ ...... . . . .. . .. .. .. . ..... . ............ .. . .. . .... ........ . .... . .. . .5 (b) 0. .. ... . .09 and frequency Ω = 0. . .... The friction strength is γ = 0. . ........ . . . ... ... . . .. . ...... . .. ........0 .. . ........ 1. ..... . . .. . ... ... ... .... ....... .... ...0 0..... .. . . ... . ..
6. 6.17c).3. 6. 6.17). It becomes larger with decreasing friction γ. This feature is in contrast to the Husimi representation of the Floquet states in absence of dissipation (cf.. since it would result in a diagonal asymptotic state (see Section 4. we cannot characterize the quantum attractor by a Hausdorﬀ dimension.09 and Ω = 0.6) are in absence of dissipation already completely resolved in the chaotic sea.6. This demonstrates that a description within a full rotatingwave approximation is insuﬃcient.17a. This transition is.3. continuous formations correspond to integer values of dH .2 ω0 (Figs.06 ω0 .19). 6. Thus. exp(S Q ). Thus. its quantummechanical counterparts obey no reﬂection symmetry at the xaxis. the occupied phasespace area is 2π exp(S Q ).19b) is still mainly located near the ﬁxed points of the classical stroboscopic map. 6.3 Signatures of chaos in the asymptotic state 75 It is computed numerically by box counting. manifest by a strange attractor (Fig. 6. Like the phasespace portrait of the dissipative classical dynamics (Fig. 124] (see Appendix A. 6. These “quantum attractors” clearly reﬂect the structures of the corresponding classical asymptotic state as well as their qualitative change from isolated ﬁxed points to a strange attractor. gives approximately the number of minimum uncertainty states covered by the Husimi function. its Hausdorﬀ dimension dH has no signiﬁcant timedependence [121]. The Wehrl entropy of the asymptotic state for our numerical example for diﬀerent values of the eﬀective quantum of action is depicted in Fig. In the semiclassical regime. A more suitable measure for the qualitative shape of the quantum attractor is the Wehrl entropy S Q of its Husimi representation [120.2).2 Quantum attractor In the quantum case. 6. thus they are smeared out in the Husimi representation of the asymptotic state (Figs.18. in the quantum case not as sharp as in the classical case: Although the asymptotic state for γ = 0.9 ω0 for diﬀerent friction strength γ is depicted in Fig.3. it covers a broader phasespace area that already indicates the shape of the strange attractor. Consequently. since diagonal representations share the symmetries of the basis. as expected. Although the attractor of the driven dissipative double well is periodically timedependent with the period of the driving. 6. Near γ ≈ 0. Fig.18b.16.7) and is caused by ﬁnite oﬀdiagonal elements of the asymptotic density matrix in Floquet representation. the classical dynamics undergoes with decreasing γ a transition from regular motion (Fig.2). however. for . Its exponential. For this driving amplitude and frequency. Because the selfsimilar structures at an arbitrary small length scale of the classical attractor are washed out in the quantum case. The underlying classical structures in the Husimi functions become more distinct for smaller values of the eﬀective quantum of action eﬀ = 1/8D. the selfsimilar ﬁne structures of a strange attractor are in contradiction to the positionmomentum uncertainty relation. 6. reﬂecting the increasing dispersion of the Husimi functions.e. 6. Fig.17b) to chaos. 6.20. For dH < 2 the attractor has zero volume. oﬀdiagonal matrix elements play a signiﬁcant role for the asymptotic state. The Hausdorﬀ dimension of the classical attractor for the parameter values F = 0. i. the regular islands near the bottom of the wells (cf.
. . ... . ... . ....... .. . ... . . .... ... . ... .. . . .... .... . . .. ......... .. ... . ... .. .... ... . .. .. .. .............. .. .. . ..... . . ... . . .. ... . . .. . . ... ... .. . .. . The rectangle in the lower left corner depicts the size of the eﬀective quantum of action eﬀ = 1/8D.. . . .. ..... . ... ....... . ..... .. .. ..5 1......... .. . .. . ....... .. .. ... ... ....... . .. .. .... .... . . . .. . . ..... .... . ...... .. .. .. 0.. .... . .... ... . ....0 0. ........ Ω = 0... .. .. ..... . ... . ... . .0 x/x0 0. . ... .... . ..17. . ......... .... .... ..... .. .. ... .9 ω0 ... . .... .... .. . .... . ... .. ... . .. .. ..5 0. .... . .. ...... . .... .. . . . ... . . .. .. . . ... ....... ...... .. . .. ... . ... . . . . ..... 6... ... ... . . ...0 x/x0 0.... .. . . ... ..... . .. 0.. .. . .. ........ ..... . .... . .... . ... .. . ... .. . . ... . γ = 0. . .. .. . . .. .. .. . .. .... . . .. ... ... ........ . . ... . . .5 (b) 0. .. .. . . ..0 0.. ... .. . .. .. .... .. . .. .... ... .... ... ... . . .. ... . . . . .. ...... .. ... ..... ..... ... . .. .. . .. . ....... .... ..... ... . .... ... .. ......... .... .. ...... . .... . .. . .... . . .. . .... . . .. .. .. . . .. . . ... ..... . . . .. ... . . . ...0 1.. ... .... ... . . . . ...... .5 1. . . . .... .. ... ...... .... . .. ...0 x/x0 0.. .... ... . . . .. ....... ......... The eﬀective action is D = 6.... .... .... ... .... ... .. .. . . ... . .. .... .. superposed on the corresponding classical phasespace portrait...... .... .. 1... . . ... . .. .. . .. . .. . .. .. .. .. ... . . ......... . .. ....... .. ... .. 6. . . .. . .. ... .. .. ... . . . ... . ..... .5 . . . .... . .. . ....... .. . ..... .. . ... ..... The parameter values F = 0... .. .. ... .. ....... . .. . ...76 The harmonically driven doublewell potential (a) 0. .. ... . . . .. ... . ... ........ . .. ........ . ... .. .... .... .. ... . ..... .. . ...... .. ...3 ω0 (a)...... . .... . . .... .. . ... .... .. .... ..... . .5 p/mω0 x0 0.... . . . ........ ... . .. .. . ....... .. . .... ..... .... .. . . . ..5 0....... ... . . ..... .... ...... . . ... . ... . . ... . ........ .. .... .5 1... ... . Fig.. . ..... ..... . ... .. . ... . .. ...... . .... . ... 0.. .... . .. . ....... . .. . .... . .. . . ..... .. ..... ... . .... .... .... . . . . .. . . . .. n → ∞. ..... . ....5 p/mω0 x0 0..... . ........ ... ........ .... ... . ... .. .. ........ . ...... ...... .... ....5 0.. .. .. .. ... . .. .. .03 ω0 (c) are as in Fig. .. ... .. ... .. . .. .... .. ...... . .... .. .... ..... ... ... .. ... . . .. ...... . .. . ... . ..0 .. ..... .. . . ... . . 1..... .. . .... .. . . ... . . . . . 1.0 0.. . . ... .. . .... .. . .. ........ .. .. .... ..... .17... . . ... ....... . .. ..09..... . ..... . ... . ...... ..... .. . ......... ...... .. ... ....... .. ... .... ... ... ... . .. .. . . .. ...... . ... ..... . .. .. . ...0 0.... . . ...... . ..... .... .. . ..... ...... .... .... ......5 . . .. ..5 (c) . . ........ . ........ . ..... . ...... . ... ..... .. .. .. . ......... . . . ... . ... .. ..... .. .. ... .. ... .0 1. . .. ............ ... . . . .5 1.5 ... .. .. ... ... . .. .. . .. ... . . ... . . .. ...... .... . . 1...... .......... .. . ..... . ... . .. .... .. ... . . .. .... .. .. . .. .. ..... . ... . .. . ..... ... ... .........5 0. . .. . .... ... ... . ..5 p/mω0 x0 0.. ..5 1..... .. .. . . . .. ... . .5 0. . ... ....18: Contour plot of the Husimi function of the quantum attractor (full lines) at t = 2πn/Ω... ... ..... . ... ..... .0 .... .. .. .. .. . . Figure 6.. . . . . .. ... .. . . ....... ... .... .. ..... ... .. . ..0 1. . . ..... .. ... . .. . . ............. . . ... . .... . ... .... ..... .....2 ω0 (b). ... .. .... ...............
. . ... . ..0 1... ... ... . . ...... .. ... ....... . ... . ..... . . ........ . .. . .. ..... .. . .. .. ......... ... .. . . .. ...... . ... .. . ........... .. .. . . .......... .. ... . . ... .... . . . . .... .. .....5 0.... ...... . . ....5 .3 Signatures of chaos in the asymptotic state 77 (a) 0... ... ... .... .. .. .. . . . . .. .... . . . ... .... .. . .....0 0..5 1.... .. ..0 x/x0 0. ... ..... .. . . ... ... ...... ....... ...5 p/mω0 x0 0. 1. . .... ... . .. . .... .. . . ... . . ...18 for the eﬀective action D = 12.... . . . . . .... .... .. .. .. .. . . .. . ... . ...... ..... .. . ... . .. .... .. .... .. . . Figure 6... . ... .. .... ... .......... ... .. ... ........ ... . ..... . ... .... .. ...... .. .......... ..... . . .... . . ... . .. .. . . . ...... . .... . ..0 0..... .. ....... . . .... .... 1.... .. . . ... .. ..... .. .. ..... . .... ..... .. .. . . . .... . . . ......... .. ... . ........ .... .. .. ... . . . ... ... ... . . .... .. . ....... . .. .. ... .... .. ....... . ..... . .. . . .... . ........... .5 1. . .. ... . . .. . .. . ........ ...5 (b) 0..... .. . . ..6. ..... .. . ... ..... . .. .. . . .. .... .... . . .... ... . . .... .. .. . ............... . . ... .. . . . . . ..... . . ..... . .... .5 0. . ... .. . ... ... ... . . ........ ... ... ... ... .. ........ .......... . . ... 0...... . . .. . .. .... . . . ... . .. . ..... . . . . . . ..... . .. ... ... . .5 1. ... .. . ....... . ... . .... . .. . ........ . ... . .. .. .... . .... . . ... ...... . .. ... . . ......... ..... . ........... . .. .. ... .. . ... . ... . ... . .. .. .. . . . ... . ... ... .. 1. .. .. .. . .. .... . .... . . ..... .. . . .. . .... . ..... . ....... . ... .5 p/mω0 x0 0.. ................. ..0 x/x0 0. . ...... . .. . . .. .... .. ..... .. ...... . ......5 (c) . .. .. . ... . . . ... . . .... .... . ... ... . . . .. ... ... ... ... . . . ... .... . .... . ... . ... .....5 0.. . .. . .. .. . ... . . . ......... ....... .... .... .. .. .. .... .. .. . . ... . ..... .. . .... .. .. ....... .. . . .... . . .......... ..... ............ . ... ... .. .. . . .5 .... ..... .... . . .. . . . ..... ... . ... . . ... .... .. .. . . . .... ... .. . ... .. .0 .. .. ... . ..... . ... .. . . ... . ... 1.. ....... ............. .... ... ...... . . ..... . .. . .. .. . . . .... ... . .. .... . . ..... .. . ....... .. . ... .... .. . .. .... . . . ... . . . .. .... ......... .. ..... . . .. ... . . ... .. ...5 0. .. .. .. . . . . . ..... .... . . ... .. ... . . . . . ..... . . ..... ........... .... ....... . .... . . .. .. . .. .. .. .. .. ..... . .. .... .. . . .. ..19: Same as Fig.... ..... ...0 1. .. ..0 .... .. .. . . ..5 0. . . ........ . . . ... . .. .. .. .... . .. . . ... ... .. .5 ... . .. .... ... .. ... .. . . ..0 0.. .. .... . . .. .... .... . . . ... . .. .. . .. .. .... . .. ...... .... .. ... ...5 p/mω0 x0 0..0 0...... .... . .... . . ... . .. .. . ... .... ... .. .. ..... . . . ...... ... . . .. .... .......... ...... ...... ....... ... ... . .. . . ... ... .. ..... .... . ... ...... ....... . ... .... .0 x/x0 0. .. . ... ... ... . ... ........ ... .. . .. .... .. .. ...... .... ................ . ........0 1. .5 1.. . .... .. ....... .. .. .. . . . 6....... . . .... . . . ... . . .. ...... . . .. . ... . ......... .. . . . . .... ... .. . . .. ....... ... ..... . ... ... . .. ........ . ..... .... .. .. ... .. ....... ...... ..5 1. .
Other parameters like in Fig. > Note that for γ ∼ 0. we observe a kink of the entropy near γ ≈ 0. 2 2 5 SQ 10 3 2 5 10 10 1 2 5 10 0 γ/ω0 a suﬃciently large value of the eﬀective action D. 6.78 The harmonically driven doublewell potential 4 3 D = 12 D=6 D=3 2 Figure 6.1 ω0 . the Markov approximation becomes inaccurate. where the classical attractor undergoes a transition from a set of isolated ﬁxed points to a strange attractor. Nevertheless.06 ω0 .20: Wehrl entropy of the asymptotic state of the dissipative quantum map for diﬀerent values of the eﬀective quantum of action eﬀ = 1/8D.16. . we obtain the qualitative behavior which we expected from classical considerations. since γ is of the order of the mean level spacing and the condition (3.33) is violated for at least some of the transitions between Floquet states.
we have incidentally obtained the Floquet solutions of the FokkerPlanck equation for the corresponding classical Brownian motion. A Markovian approach to quantum dissipation. thus obtained a partial diﬀerential equation for the Wigner function that corresponds to the density operator. vanishes for strictly Ohmic damping. Concluding from numerical results for the case of a Mathieu oscillator. The diﬀerence becomes signiﬁcant in the limits of strong driving and low temperature. provides a faithful description of an isolated singletdoublet crossing. the separatrix is replaced by a chaotic layer. reﬂecting an eﬀective twostate behavior. based on the Floquet solutions of the coherent dynamics. An additional additive timedependent force undergoes a renormalization which. which is the essence of chaotic tunneling. Even for arbitrarily small driving amplitude. We have derived this FloquetMarkov approach from an exact pathintegral expression and have applied it to the parametrically driven harmonic oscillator and the driven doublewell potential. Here. Well outside the crossing. both timescales are of the same order. To solve the master equation. Nevertheless. has proven welladapted to the description of such systems. The center of the . It turned out that the dissipative part of the Markovian master equation depends quantitatively on whether the driving is included in its derivation or not: Considering the driving mainly results in a modiﬁed momentum diﬀusion that depends on the quasienergy spectrum instead of the unperturbed spectrum of the central system without the driving. As a simple intuitive model to compare against. we have transformed it to Wigner representation. and derived an analytical expression for the Floquet solutions of the resulting FokkerPlancklike equation. the inﬂuence of states located in the chaotic region alters the splittings of the regular doublets and thus the tunnel rates. the attributes “simple” and “improved” for the two basic Markovian approaches prove adequate.7 Summary and outlook In this thesis. but the motion near the bottom of the wells remains regular. classical chaos plays a signiﬁcant role for the coherent dynamics. We have studied chaotic tunneling in the vicinity of crossings of chaotic singlets with tunnel doublets under the inﬂuence of an environment. Dissipation introduces new time scales to the system: one for the loss of coherence and a second one for the relaxation to an asymptotic state. we have constructed a threestate system which in the case of vanishing dissipation. In doing so. The study of the parametrically driven harmonic oscillator has been devoted mainly to a thorough understanding of the diﬀerent approximation schemes. we put focus on a special class of system: a particle which moves in a onedimensional potential under the inﬂuence of a heat bath and of an external ﬁeld which is periodic in time. however. A quantum system with more complex dynamics is the quartic doublewell potential under the inﬂuence of a driving with frequency near resonance.
With increasing driving amplitude. it remains chaotic. “external” states of the central system. the quantum attractor undergoes a smooth transition: The structure of the strange attractor is already felt by the Husimi function for parameter values where the classical attractor consists only of two isolated ﬁxed points. the geometry of the classical attractor is fractal or regular. the quantum attractor. accordingly. but for strong friction it becomes regular. A semiclassical description of the dissipative quantum system may circumvent this problem. As a result. chaotic Bloch tunneling along extended potentials with a large number of unit cells instead of just two. In the presence of driving. Many more phenomena at the overlap of chaos. even if that is barely visible in the corresponding classical phasespace structure. However. which requires to take very many levels into account. These phenomena are typically observed in the far semiclassical regime. the asymptotic state is no longer literally a state of equilibrium. incoherent processes create a steady ﬂow of probability involving states within as well as outside the crossing. This is so because the coupling to the heat bath enables processes of decay and thermal activation that connect the states in the crossing with other. expressed for example by its coherence tr 2 . Thus. tunneling fades out much faster. They include fourstate crossings formed when two doublets intersect. are markedly diﬀerent at the center of the crossing as compared to ∞ the asymptotic state far away from the crossing. the dynamics near the bottom of the wells. tunneling. This has striking consequences for the dissipative classical dynamics: For suﬃciently small dissipation. and the inﬂuence of decoherence on a multistep mechanism of chaotic tunneling. This clearly reﬂects the failure a full rotatingwave approximation.80 Summary and outlook crossing is characterized by a strong mixing of the chaotic state with one state of the tunnel doublet. respectively. in absence of dissipation. We have observed the signatures of this qualitative diﬀerence in the asymptotic state of the corresponding quantum dynamics. . decoherence becomes far more eﬀective and. For the observation of these semiclassical structures. in contrast to the sudden change of the classical behavior. Accordingly. becomes fully chaotic. the composition of the asymptotic state. The high mean energy of the chaotic state introduces additional decay channels to states outside the threestate system. oﬀdiagonal matrix elements of the asymptotic state in Floquet basis proved crucial. Rather. demonstrates clearly that a threestate model of the singletdoublet crossing is insuﬃcient once dissipation is eﬀective. and dissipation await being unraveled. The study of the asymptotic state.
7) (A.8) (A.9) These justify the denotation creation and destruction operator (of a quantum) or shift operators (between eigenstates) for a+ and a. we use its eigenfunctions as a basis set for numerical computations. the harmonic oscillator plays an important role as an exactly solvable model as well as an approximation to a smooth potential minimum. By recursion of (A. = ωHO a+ a + 2 (A. 2m ωHO 1 p. the ground state of a harmonic oscillator.2) and of the closely related coherent states and quasiprobabilities. a+ ] = 1. displaced in phase space (coherent state).2) of the Hamiltonian is achieved by the transformation a= a+ = x= mωHO x+i 2 mωHO x−i 2 1 p.9). p] = i results the bosonic commutation relation [a. p=i 2 (a+ + a). The form (A. the socalled number states (a+ )n (A.A The harmonic oscillator In many ﬁelds of physics. 2m ωHO (A. we give a synopsis of basic properties of the harmonic oscillator.6) 2mωHO m ωHO + (a − a). forms the initial state for the propagation of the density matrix in Chapter 6.4) (A.5) (A. In this appendix. which yields for the energy eigenstates n the relations [125] √ an = n n − 1 . described by the Hamiltonian HHO = 2 1 2 mωHO 2 p + x 2m 2 1 . (A. √ a+ n = n + 1 n + 1 . From [x. In this work. Moreover.10) n = √ 0 n! .3) (A.1) (A.
2m 2 thus < p ∼ pn = < x ∼ xn = 2n ωHO m .15) To visualize the inﬂuence of a ﬁnite basis set. thus their value is a numerical artefact caused by using a ﬁnite basis set. thus formally approximates inﬁnite matrices by ﬁnite ones. as matrix elements of powers of the position operator for these states obey a simple analytical expression resulting from (A. which is deﬁned by a0 = 0. we have depicted some eigenvalues of the truncated Hamiltonian PN HDW PN for N = 100 over the scaling parameter ωHO /ω0 of the basis functions in Fig.13). the energies depend on the scaling parameter. where PN projects on the subspace spanned by the ﬁrst N basis functions {n }n=0. Thus.14) (A. we eﬀectively diagonalize—instead of the Hamiltonian H—the truncated Hamiltonian PN HPN . the eigenfunctions of the harmonic oscillator form a wellsuited basis set.13) (A. This subspace. The state n in a semiclassical interpretation [126. Therefore.9).12) (A. it is restricted to phasespace areas which obey 1 p2 2 < + mωHO x2 ∼ n ωHO .12)..127] is a quantized torus with action (n + 1/2). one uses N number states (A. .11) A.14) and (A.82 The harmonic oscillator are constructed from the ground state 0 .N ..5)–(A. The numerical computations in Chapter 6 were performed using number states with an oscillator frequency ωHO = ω0 N ω0 16EB 1/3 . (A. 2n . A. In numerical calculations.1.10) as a (incomplete) basis set. mωHO (A. (A. This results in the conditions p2 N = N ωHO . a state with energy E can be approximated reasonably by a linear combination of the ﬁrst N number states only if its corresponding classical torus is contained in this region of phase space. corresponds to a ﬁnite region of phase space. according to (A.16) and N was chosen according to the required numerical precision. 2 2ωHO 16EB ωHO E< (A.15). wave functions and operators are decomposed into a complete set of basis functions. Dealing with polynomial potentials. Consequently. 2m 2 4 N ω0 N 2 2 ω0 E < V (xN ) = − + .1 Number states as a basis set For numerical computations. Outside the limits (A.
p) = 1 2π 2 dξ dξ ezξ ∗ −z ∗ ξ χs (ξ).15).17) a quantummechanical state cannot be localized in phase space with arbitrary precision.1: Some eigenvalues of the truncated Hamiltonian PN HDW PN for N = 100 and D = 4 (full lines). z ∈ C (A.19) (A. they have minimal uncertainty and approximate a point in phase space at best.2 Coherent states 83 100 80 60 40 20 0 1 10 Figure A. 2 (A.18) obey zxz = zpz = 2 Re z. mωHO 2m ωHO Im z. 2 5 E/¯ ω0 h 10 ωHO /ω0 0 2 5 A.A. z∆x2 z = .3 Quasiprobabilities The unique representation of a density operator as a phasespace function is closely related to the question on quasiclassical states. 129] + ∗ z = eza −z a 0 .20) Thus according to (A. A. The most prominent example from a variety of possibilities [130–134] is the sparameterized quasiprobability or CahillGlauber distribution [135] Ws (x.14) and (A. (A. The coherent states (or Glauber states) [128. The broken lines give the limits of convergence according to (A.21) . as would be possible in classical mechanics.2 Coherent states Due to the Heisenberg uncertainty principle ∆x ∆p ≥ 2 (A.17). 2mωHO m ωHO z∆p2 z = .
a negative s with small absolute value is often used to ensure numerical convergence.28) For powers of x and p they hold iteratively. there exists a corresponding diﬀerential operator acting on Ws (x. It is obvious from these operator correspondences that.84 The harmonic oscillator χs (ξ) = tr eξa z=x + −ξ ∗ a+sξ ∗ ξ/2 . p) [137]. (A. the sparameterized quasiprobability depends on the choice of the oscillator frequency ωHO . z ∗ =z=0 (A. Thereby the operator ordering is ﬁxed by the parameter s as the sordered product (a ) a + n m s = ∂ ∂z n ∂ − ∗ ∂z m s exp za+ − z ∗ a + z ∗ z 2 . In numerical computations of Wigner functions or their reconstruction from experimental data. except for the case s = 0. p).29) It is independent of the oscillator frequency ωHO . From Eq.1 Wigner function For s = 0. Ws results in the Wigner function [130.23) mωHO + ip 2 1 . In general.27) (A. p).3. thus basis independent. (A. For each operator acting on the density matrix . Ws (x.25) (A.24) which gives an interpolation between normal ordering (a+ )n am = {(a+ )n am }1 and antinormal ordering am (a+ )n = {(a+ )n am }−1 of creation and annihilation operators [136]. Quasiprobabilities are used for the calculation of expectation values alike classical phasespace distributions. .22) we obtain the relations x ←→ p ←→ x ←→ p ←→ s i ∂p − ∂x 2 2mωHO i sm ωHO p − ∂x − ∂p 2 2 s i ∂x x − ∂p − 2 2mωHO i sm ωHO p + ∂x − ∂p 2 2 x+ Ws (x. p). Ws may also assume negative values and for positive s may even be singular—thus a strict probabilistic interpretation is not possible. p). (A. (A.21) runs over real and imaginary part of ξ. 2m ωHO It includes the Wigner and the Husimi function as limiting cases. (A. s ∈ [−1. p) = 1 2π dx eipx / x + x /2 x − x /2 = W0 (x. A. 1].26) (A. Ws (x. The integration in (A. p).22) (A. Ws (x. 133] W (x.21) with Eq.
the Husimi function of an eigenstate is located on the corresponding quantizing torus. For driven systems. The fact that already the diagonal matrix elements hold the full information on the quantum state reﬂects the overcompleteness of the coherent states [128].23). due to the positivity of the density operator [133]. it is desirable to have a direct measure for localization properties. This allows for a classiﬁcation of single eigenstates as regular or chaotic if the classical dynamics is mixed. . (A.A. Q(x. 124]. the respective assignment of Floquet states to regions in classical phase space holds true [120].31) The number of minimum uncertainty states occupied by the Husimi function is approximately given by exp(S Q ). thus the occupied phasespace area is 2π exp(S Q ).2 Husimi function and Wehrl entropy The Husimi function is deﬁned as the expectation value of the density operator with coherent states [131] and coincides with the quasiprobability W−1 .3. Thus. p) = 1 2π z z = W−1 (x. p)]. p) is given by (A.3 Quasiprobabilities 85 A. In a semiclassical limit. p) ln[2π Q(x. For a classiﬁcation of quantum mechanical states according to their phasespace structure. for a coherent state the Wehrl entropy assumes its minimum value Q Smin = 1. for the case of irregular classical dynamics. (A. the Husimi function of a state is localized in phase space along the corresponding Lagrangian manifolds. One possibility is provided by the Wehrl entropy S Q of the state which is deﬁned as the entropy of the corresponding Husimi function [120. SQ = − dx dp Q(x. in case of regular classical dynamics. it is smeared out over the whole chaotic layer [5]. It is nonnegative. Consequently. p).30) where z(x.
86 .
139. are not of this socalled Lindblad form. i (B. Therefore. Namely. 140]. . of course. have to be conserved during time evolution. 0 ≤ pi ≤ 1.4) positivity will be violated until ∆x becomes larger than λdB [68–71. as well as its Hermitecity. can at best describe it by a density operator [138]. (B. 141]. i. Lindblad proved [93] that a Markovian master equation with constant coeﬃcients meets this requirement. positivity and a total probability which equals unity. ] + † γi 2Qi Qi − Q† Qi − Q† Qi . B. if and only if it is of the form i ˙ = − [H. which are introduced phenomenologically. This apparent contradiction was resolved only recently: A master equation of the form (4.3) violates positivity only for initial conditions that do not meet the requirements under which it has been derived.1 Lindblad form The conditions on a physically meaningful density operator. ∆x < λdB = / 4mkB T . that many Markovian master equations occurring in the literature [11. all the others vanish. induce dissipative transitions of the system.B The density operator An observer. (B. Eqs. 71.2). dissipative eﬀects on a length scale l < λdB cannot be described selfconsistently within a Markov approximation.e. It turned out.2) pi = tr = 1. thus generates a socalled completely positive dynamical semigroup. i i (B. who is not fully aware of the state of a system. 70. however. thus they do not ensure positivity of an arbitrary density operator at any future time.1) and (B. Its eigenvalues pi give the probability for the system to reside in the corresponding eigenstate. including our master equation (4.3) i The operators Qi .e. Thus. the wavefunction) is known.3). where the full quantummechanical information (i. one of the probabilities pi equals unity.1) (B. In the limit of a pure state. the eigenvalues of a proper density operator have to suﬃce the intrinsic restrictions of probabilities. if the system is prepared with a position variance ∆x smaller than the thermal de Broglie wavelength.
Its value approximately gives the reciprocal of the number of incoherently populated states and equals unity if the system resides in a pure state. [142].7) is a proper measure for the coherence of a density operator. A numerically less expensive. (B. (B. the related quantity C = tr 2 = 1 − Slin (B. Nevertheless. thus for the ability to observe interference eﬀects.6) introduced by Zurek et al. The entropy also gives a proper measure for the coherence of a system. It arises formally by Taylor expansion of (B. However.5) i Consequently. . This deﬁnition agrees. with the entropy known from statistical thermodynamics. In the case of many incoherently populated states. besides a factor kB . for a pure state S = 0.2 The density operator Coherence and entropy The lack of information inherent in a density operator can be measured by the Shannon entropy S=− pi ln pi = − tr( ln ). related quantity is the “linearized entropy” Slin = tr (1 − ) = 1 − tr 2 .5) if describes an almost pure state. all pi 1 and both entropies diﬀer drastically.88 B. it has the disadvantage that its direct numerical computation requires diagonalization of the density operator.
(C. St ) = 0 (C.3) We denote the partial derivatives of S(X. we solve the equation of motion (5. t) = dXdP eixX+ipP eS(X. respectively. t) as W (x.3).t) .2) to get an expression for St . Equation (C. P. Instead of equation (C.8). where F is given by F = St − XSP + γP SP + ω 2 (t)P SX + γDpp P 2 + γDxp XP.2) are given by ˙ ∂F = 1. ∂P dω 2 (t) ∂F ˙ =− P SX . (C. The characteristic equations [95] of (C. ∂SP ∂F ˙ = SP − γDxp P.55) for the Wigner function by the method of characteristics. SP . So we only have to solve (C.5) and (C.2).P. we ﬁnd ¨ ˙ P − γ P + ω 2 (t)P = 0. The solutions of these equations can be traced back to the fundamental solutions fi (t) of the classical equation of motion (5. we will use (C.8) (C. t) with respect to X.1) By this ansatz.7) (C. P . SP .5) (C. P.C Solution of the FokkerPlanck equation In this appendix.6) (C.4) (C. St = − ∂t dt (C. equation (5. P. X= ∂SX ∂F ˙ P = = γP − X.4) signiﬁes that the characteristics can be parameterized by the time t. SX . t= ∂St ∂F ˙ = ω 2 (t)P.9). p.9) whose solutions give the characteristics of the partial diﬀerential equation (C.6).10) .55) is transformed to the quasilinear partial diﬀerential equation F (X. and St .5)–(C. and t by SX . From (C.2) for S(X. We write W (x. t. t). p. SX = − ∂X ∂F ˙ SP = − = −γSP − ω 2 (t)SX − 2γDpp P − γDxp X. (C.
Now together with Eq.2). t )P (t ). σpp (t. which can be integrated to P (t ) = G(t.20) 2 σxp (t. X(t) = c1+ eγt f˙2 (t) − c2+ eγt f˙1 (t). t0 )P 2 . where ci+ denote integration constants. (C.13) (C. t0 ) = 2γD t0 t dt [G(t.7) SP (t) = c1− f˙1 (t) + c2− f˙2 (t) − 2γD By inserting t dt G(t. t) = c1− f1 (t) + c2− f2 (t) X + c1− f˙1 (t) + c2− f˙2 (t) P 1 1 − σxx (t. we get a result for SX and SP that only depends on the endpoints of the characteristics.15) dt t0 ∂G(t.12) which is the classical equation of motion with an inhomogeneity. t )X(t) + S(X.13) with the Green function (5. t ) t0 t 2 (C.19) (C. . t ). St ). From (C. (C. we integrate (C. dt G(t. t0 (C. t).11) and (C. t ) P (t). (C. P. t0 )XP − σpp (t. ∂t (C.18) σxx (t. we ﬁnd a timedependent solution for the Wigner function W (x.11) (C.21) t0 By inserting S(X. t ) P (t ) + γDxp P (t). t0 )X 2 − σxp (t. 2 2 with t (C.1). t ) ∂t .14) With the integration constants ci− . t0 ) = −γDxp + 2γD (C.16) ∂G(t. t) into (C. t) = (SX . (C. p. Therefore the solutions for X and P read P (t) = −c1+ eγt f2 (t) + c2+ eγt f1 (t). The eﬀective diﬀusion constant D is given by D = Dpp + γDxp .90 Solution of the FokkerPlanck equation This is simply the classical equation of motion with a negative damping constant. P.8) we ﬁnd for SX ¨ ˙ SX + γ SX + ω 2 (t)SX = −2γDP. we have an expression for grad S(X.12). ∂t dt ∂ G(t. P. (C.7) and (C. t )] . t0 ) = 2γD ∂ G(t. SP .10) to t SX (t) = c1− f1 (t) + c2− f2 (t) − 2γD and get by use of (C.17) ∂t obtained from Eqs.
−ix. i. (C. t0 )∂p . p.. P. Therefore. −ip. are ˆ proportional to the ci+ . there is no ambiguity concerning the ordering of operators. The operators Qi+ (t). St ) vanish [95]. t0 )∂p − if2 (t) p + σxp (t. SX . used in Section 5.Solution of the FokkerPlanck equation 91 The integration constants ci± are constant along the characteristics by construction.e. SP . 2 1 = f2 (t)∂x + f˙2 (t)∂p . P. SX . whose nullspace is the solution of the equation of motion. −i∂p . p. t0 )∂x + σpp (t. t) commute with ˆ the operator ∂t − L(t). t. c2− = −if˙1 (t) x + σxx (t. t) ˆ is a solution of (5.55). the Poisson brackets between the expressions ci± (X. then ci± W (x. t0 )∂x + σpp (t. . By transforming back from Fourier space to real space.23) (C.4. For the ci± we ﬁnd ˆ ˆ c1+ = ˆ c2+ ˆ c1− ˆ 1 f1 (t)∂x + f˙1 (t)∂p .24) Note that because of the linear structure of the characteristic equations. SP . 2 = if˙2 (t) x + σxx (t.22) (C. t) is also a solution. t0 )∂p . t0 )∂x + σxp (t.2 to construct the Floquet solutions of the FokkerPlanck equation. t) and F (X. the ci± are shift operators in the subspace of solutions. t0 )∂x + σxp (t. Thus. t0 )∂p ˆ + if1 (t) p + σxp (t. if W (x.25) (C. one ﬁnds that the operators ci± ≡ ci± (−i∂x .
92 .
Rep. 1199 (1962). Rev. Utermann. (N. Rev. Rev. H¨nggi. M. [12] M. (N. 273 (1994). 2413 (1984). Phys. Schuster. P. 1979). 363 (1990). E 50. Casati and J. Lett. Jung. 2nd ed. and P. Semiclassical level spacings when regular and chaotic orbits coexist. Zur Deutung der Molekelspektren III: Bemerkungen uber das ¨ Schwingungs. Long Time Behavior in the Quantized Standard Map with Dissipation. 43. Chirikov. Casati. L. Dynamical quasidegeneracies and separation of regular and irregular quantum levels. Dittrich. 65. Phys. Weinheim. Bohigas. Rev. [4] E.Y. Graham. A 17. 1515 (1984). Berlin. J. 516 (1991). [13] O. Robnik. 145 (1994). H¨nggi. Phys. 1479 (1990). 67. Rev. Dittrich and R. Phys. in Stochastic Behavior in Classical and Quantum Hamiltonian Systems. Rev. 53. Dyson. A. Grossmann. Berry and M. Lett. V. (VCH. and P. Vol. S. 3. Ballentine.und Rotationsspektrum bei Molekeln mit mehr als zwei Kernen. Z. Phys. 219 a (1998). Phys. Leggett. 1967). B. Phys.References [1] G. Phys. [16] R. S. edited by G. and D. Phys. V. Tomsovic. Rev. T. Ann. and D. Lett. Ullmo. Quantum Tunnelling in a Dissipative System. L. Rev. Random matrices and the statistical theory of energy levels (Academic Press. Phys. Chaosassisted tunneling. Lett. [3] H. Takahashi and N.Y. Ford (Springer. 93 of Lecture Notes in Physics. Caldeira and A. 55. Dittrich. Izrailev.) 149. Quantum tunneling and chaos in a driven anharmonic oscillator. p. 5 (1990). Hund. [14] O. [8] M. E. 2927 (1990). Ullmo. Tomsovic. Deterministic chaos: an introduction. Driven Quantum Tunneling. Lett. Coherent Destruction of a Tunneling. Saitˆ. 805 (1927). [15] S.) 200. Lett. Mehta. Phys. Bohigas. Heller. [2] T. [10] F. T. [9] F. Phys. [7] M. Classical transport eﬀects on chaotic levels. [6] F. Ann. O.G. [5] K. Ullmo. F. Tomsovic and D. New York. J. Math. Lin and L. 645 (1985). Ford. BoundState Eigenfunctions of Classically Chaotic Hamiltonian Systems: Scars of Periodic Orbits. and J. 374 (1983). [17] W. 65. Chaos and Husimi Distribution Function in Quano tum Mechanics. J. Tunneling and the Onset of Chaos a in a Driven Bistable System. . 304. 334. E 49. The Threefold Way: Algebraic Structure of Symmetry Groups and Ensembles in Quantum Mechanics. [11] A. 1989). H¨nggi. Phys. Phys. 64. Grifoni and P.
J. [21] R. (N. P. and D. Ann. Eksp. [22] R. Kohler. E 50. Manifestations of classical phase space structures in quantum mechanics. [31] M. H¨ bner. Kohler. Latka. [33] M. Sch¨n. Chaosinduced avoided level crossing and tunneling. Dittrich. T. H¨nggi. [Sov. [28] C. V. and L. Walther. G. Phys. Control of dynamical tunneling in a bichromatically driven pendulum. D. Rev. The theory of a general quantum system interacting with a linear dissipative system. Brownian parametric quantum oscillators with dissia pation. [29] J. P. 319 (1986). Phys. Rep.94 References [18] J. 223. I. 33. G. [35] G. and W. Teor. and P. G. E 50. Latka. Vernon. [30] O. R. Feynman and F. J. and T. Rev. [27] S. E 55. U. [32] M. Phys. Rev. L. 74. and L. L. Latka. and B. P. E 52. Rapoport. Zoller. B. P. H¨nggi. 141. Phys. Zerbe and P. Ann. Graham and R. Utermann.) 234. West. P. Sirko. Some general properties of quasienergetic spectra of quantum systems in classical monochromatic ﬁelds. N. S. (N. 4091 (1995). 47 (1883). 1942 (1959). B 11. 43 (1993). Phys. Weinheim. H¨nggi. Dittrich. Rev. Lett. Phys. J. ı. Rev. J. R3299 (1994). Bohigas. 12. Rev. A 25. Rev. West. Phys. Chem. J. Cirac and P.L. A 50.) 24. Fainshtein. Grigolini. Phys. 596 (1994). L. 1533 (1995). Phys. 300 (1994). Atoms in a laser ﬁeld. Smilansky. 1338 (1960). Rapoport. L303 (1992). and B. [25] T. P. Coherent and incoherent a chaotic tunneling near singletdoublet crossings. Generalized QuasiEnergies and Floquet States for u a Dissipative System. a o Quantum Transport and Dissipation (WileyVCH. Ovsiannikov. P. Phys. Grigolini. Magalinski˘ Dynamical model in the theory of the Brownian motion. B. [19] V. Sup. Tomsovic. Phys. P. 341 (1989). Ingold. Quantum Computations with Cold Trapped Ions. L. H¨nggi. Phys. Grigolini. Dittrich. [20] R. Zwerger. Ensemble method in the theory of irreversibility. Bl¨ mel. [26] R. H.Y. Rev. Ann. Rep. 36. Classicalquantum correspondence for barrier crossing in a driven bistable potential. de l’Ecole Norm. Lett. Phys. and K. 62. JETP 9. 2561 (1978). Phys. 118 (1963). and B. Chaos and avoided level crossing. u Microwave excitation of Rydberg atoms in presence of noise. Manakov. [24] N. Manakov. [34] S. Zh. Ullmo. Plata and J. FloquetMarkovian description of the a parametrically driven. 7219 (1998). Gomez Llorente. R. M. . Floquet. Phys. Graham. Yamada.Y. Phys. Zwanzig. West. 1381 (1959)]. E 58. Fiz. 1998). Kramer. 1071 (1994). dissipative harmonic quantum oscillator. 300 (1997). [23] A. J.
739 (1989). Howland. Quantum Dissipative Systems. [42] W. The FokkerPlanck Equation. Chem. Math. Molinari. and G. Comments At. Classical Mechanics. DupontRoc. [37] S. Steady States and Quasienergies of a QuantumMechanical System in an Oscillating Field. Magnus and S. H. Helv.I. Berlin. Mod. Louisell. Timeindependent scattering theory for general timedependent Hamiltonians. edited by G. H¨nggi. New York. Time dependence in quantum mechanics—Floquet theory and the Berry phase. [50] W. Vol. [47] U. (AddisonWesley. Grynberg. Phys. CohenTannoudji. Weiss. Phys. 1992). S. Ann. 99. Acta 66. Reading. 2 of Series in Modern Condensed Matter Physics (World Scientiﬁc. [45] H. Prog. Risken. 207. 1955). t ) method: Theory. [53] P. H. 62. 1980). [51] F. Reactionrate theory: ﬁfty years after a Kramers. 95 of Springer Tracts in Modern Physics. A 7. [41] D. in Quantum Statistics in Optics and SolidState Physics. Chem. 66 of Springer Tracts in Modern Physics. Moiseyev. Berlin. H¨hler (Springer. Vol. [43] J. Press. Adv. B979 (1965). 2nd ed. Shirley. Berlin. The solution of the timedependent Schr¨dinger o equation by the (t. Phys. Sambe. Casati and L. Phys. [38] J. Princeton. Winkler. Phys. 1982). Vol. H¨hler o (Springer. 87 (1995). 73. 4590 (1993). [40] H. [52] H. Vol. [39] J. 138. Singapore. 1979). 18 of Springer Series in Synergetics (Springer. Atom photon interaction: basic processes and applications (Wiley. computational algorithm and applications. edited by G. [44] C. Solution of the Schr¨dinger Equation with a Hamiltonian Perio odic in Time. Mathematical foundations of quantum mechanics (Princeton Univ. Grabert. 1984). J. Quantum Statistical Properties of Radiation (Wiley & Sons. Theor.References 95 [36] G. Peskin and N. 251 (1990). Moiseyev. Phys. in Projection Operator Techniques in Nonequilibrium Statistical Mechanics. 1993). Phys. 1973). 315 (1974). J. Rev. Talkner. Borkovec. Stationary Scattering Theory for Timedependent Hamiltonians. [48] H. o 1973). 31. Moore. 98. Mol. Goldstein. P. Rev. Suppl. . New York. 2203 (1973). Generalized Floquet theoretical approach to intenseﬁeld multiphoton and nonlinear optical processes. and M. von Neumann. Haake. “Quantum Chaos” with TimePeriodic Hamiltonians. Chu. Rev. [46] N. J. Hill’s Equation (Dover. Phys. New York. 3 (1993). 287 (1989). [49] U.
SchimanskyGeier and T. p. 1060 (1994). Pechukas. Talkner. Riseborough. Phys. [62] A. Chem. 517 (1993). [69] P. and G. Chem. edited by W. 97. [59] R. Grabert. Phys. S. 123. W. 8365 (1993). 1997). Exact Results for a Damped Quana tum Mechanical Harmonic Oscillator. J. 1 (1981). 106. Makri. P. Feynman and A. Quantum Brownian Motion: The Funtional Integral Approach. Theory. B258. I. New York. R. [57] D. [60] H. Gans (Plenum Press. Phys. [64] P. 471 (1985). Phys. 1965). H¨nggi. [68] P. J. Physica o A 199. Schulman. Chem. Voth. Schramm. edited by L. [71] L. Lett. Zwanzig.L. On a Quasiclassical Langevin Equation. Stat. Nonlinear Generalized Langevin Equations. Rep. Berlin. Benguria and M. On HighTemperature Markovian Equation for Quantum Brownian o Motion. Kac. Schmid. Control of dissipative tunnelling dynamics by continuous wave electromagnetic ﬁelds: Localization and largeamplitude coherent motion. [70] L. 2286 (1997). Vol. U. [63] G. Phys. NATO ASI “Largescale molecular systems”. 168. in Stochastic Dynamics. Lett. A 31. 87 (1984). Phys. 46. J. A. 1 (1993). Vol. Lett. Phys. P¨schel o (Springer. Tensor propagator for iterative quantum evolution of reduced density matrices. 15. 4600 (1995). . CaldeiraLeggett master equation and medium temperatures. Europhys. Feynman Path Integral of Quantum Mechanical TransitionState Theory. Low Temp. Quantum Theory of the Damped Harmonic Oscillator. [61] R. Phys. 22. [56] G. in Proc. Rev. Rev. Phys. Grabert. Makri and D. Z. 1991). 609 (1982). 49. Kac. Stabilization of localized states in dissipative tunneling systems interacting with monochromatic ﬁelds. [66] L. and P. [55] P. New York. Quantum Langevin Equation. P. Generalized Langevin Equations: A Useful Tool for the Perplexed a Modeller of Nonequilibrium Fluctuations?. On the Quantum Langevin Equation. H¨nggi. Makarov and N. B 55. Pechukas. [65] R. 73. Rev. E. Weiss. J. Techniques and Applications of Path Integrals (Wiley & Sons. Hibbs. Di´si. Stat. J. Ingold. Phys. Weiss. Reduced Dynamics Need Not Be Completely Positive. Phys. p. 803 (1987). Phys. and U. 484 of Lecture Notes in Physics. 9. 102. 46. [58] N. Ford and M. Makarov. E. Rev. 5863 (1995). New York. Di´si. 1981).96 References [54] H. 115 (1988). Quantum Mechanics and Path Integrals (McGrawHill. P. Makri. E 52. J. [67] N. 215 (1973).
3596 (1995). S. [Sov. S. Phys. P. 1995). L.I. 3497 (1993). Popov. and W. 1991). Phys. B. Bl¨ mel. Theory and Applications of Mathieu Functions (Dover Publications Inc. Dittrich. Lewis. Driven Tunneling: a a New Possibilities for Coherent and Incoherent Quantum Transport. in 25 Years of NonEquilibrium Statistical Mechanics. 531 (1990). 10. Riesenfeld. A 44. Zh. Graham. Classical and Quantum Systems with TimeDependent HarmonicOscillatorType Hamiltonians. Phys. W. 1798 (1986). Dynamical localization in the microwave interaction of Rydberg atoms: The inﬂuence of noise. 445 of Lecture Notes in Physics. [73] M. H¨nggi. Quantum Signatures of Chaos. Teor. 510. Rev. J. Grouptheoretical aspects of the variable frequency oscillator problem. H¨nggi. Fiz. [85] N. B. Phys. Smilansky. Phys. Teor. 589 (1968). Lett. 55. [78] T.. H¨nggi. 66. Weiss. A. M. R. JETP 29. and H. Oelschl¨gel. 62. 1964). . [79] F. Math. Electromagnetic traps for charged and neutral particles. Chu. 636 (1967). and R. [75] R. Rev. J. Utermann. Phys. 527 (1991). Rev. Berlin. [74] T. 1684 (1969). Popov and A. McLachlan. A 33. Rev. Fiz. 269. 57. R. Lett. [84] W. Perelomov. [86] H. Phys. M. 22. Haake. Rev. Jr. Sassetti.References 97 [72] M. [76] T. Teor. Dittrich. and U. [77] B. 1. Oelschl¨gel. FloquetLiouville supermatrix approach: Time development of densitymatrix operator and multiphoton resonance spectra in intense laser ﬁleds. 54 of Springer Series in Synergetics (Springer. [Sov.. Weiss. New York. JETP 30. Phys. Wang. and U. Driven Dissipative Tunneling. B. Oelschl¨gel. P. Phys. Dittrich. Vol. R. Brey (Springer. Rev. Mod. E 48. Eksp. Grifoni. 18. Stockburger. Phys. Phys. Jr. 360 (1969). M. [88] L. 910 (1970)]. Nonlinear response of a periodically driven damped twostate system. and P. H¨nggi. Buchleitner. Z. Berlin. [82] V. Quantum Motion in a Paul Trap. Acta Physica Polonica B 24. Zh. 4521 (1991). [81] V. Global and Local Dissipation in a Quantum Map. An Exact Quantum Theory of the TimeDependent Harmonic Oscillator and of a Charged Particle in a TimeDependent Electromagnetic Field. E 52.S. S. Eua a rophys. Eksp. M. Perelomov. Lewis. Sassetti. 845 (1993). Rev. Parametric exitation of a quantum oscillator. Lett. [87] H. A. 75 (1985). Grifoni. u Walther. [83] V. Paul. 5 (1993). p. 1458 (1969). K. Brown. [80] R. T. Fiz. and P. Graham. Mat. Popov and A. Damped periodically driven quana a tum transport in bistable systems. and S. M. Ho. S. Cooperative eﬀects in the a nonlinearly driven spinboson system. Vol. Sirko. B 59. edited by J. Perelomov. J. Parametric exitation of a quantum oscillator II. 719 (1969)]. U.
Kamke. 88. Z. 48. [101] E. Utermann. 206 (1982). Lett. [95] E. a E 49. Roncaglia. 1994). [106] T. M. D. A. M. Ryzhik. [99] W. Phys. Phys. San Diego. Rev. D. Korsch.98 References [89] G. Semiclassical description of tunneling in mixed systems: Case of the annular billiard. Dittrich and R. W. 3661 (1995). [90] J. Frischat and E. Schleich. 1979). [104] F. Hartley and J. F. Dittrich. B 77. and Products. SchimanskyGeier. Wigner Functions in the Paul Trap. Man’ko. and T. Quantum Semiclass. Gomez Llorente. B. Doron and S. Rev. B.J. Coherent States for the timedependent harmonic oscillator. Series. Ullmo. 307 (1995). G. Opt. 1013 (1994). 802 (1994). Bonci. Mirbach. and L. E 52. I. Monteoliva. Doron. Commun. Tunneling versus chaos in the kicked Harper model. Ballentine. (Academic Press. Zerbe. H’walisz. 6th ed. Rep. Diﬀerentialgleichungen. [94] P. Colored a noise driven systems with inertia. Grigolini. Europhys. (Teubner. 2529 (1996). and H. Stuttgart. Phys. and P. Physica A 91. J. Leyvraz and D. [103] S. H¨nggi. P. A 45. 4. Quantum Eﬀects in the Steady State of the Dissipative Standard Map. Phys. [92] I. Titulaer. Glauber. H¨nggi and H. 1421 (1998). Lett. Math. Mixed dynamics and e tunneling. J. M. 471 (1989). Phys. Jung. P. [93] G. Rev. 321 (1978). Guti´rrez. Symmetries a and Linear Response. Lindblad. [97] C. P. Lin and L. 73. Ray. The level splitting distribution in chaosassisted tunneling. R. . [102] E. E 57. 263 (1987). Vol. West. L. V. D 25. Tunnel Splittings and Chaotic Transa port in Periodically Driven Bistable Systems. 5th ed. Dynamical tunneling in mixed systems. Schrade. Phys. [96] L. Frischat. Brownian parametric oscillators. Rev. Phys. M. [100] P. H¨nggi. Rev. Talkner. Rev. A 57. I. J. Phys. [98] U. Lett 75. Quantum tunneling and regular and irregular quantum dynamics of a driven doublewell oscillator. Stochastic Processes: Time Evolution. and J. Physica B 194196. R. J. Global and local dynamical invariants and quasienergy states of timeperiodic Hamiltonians. Graham. H¨nggi. Rev. [105] R. 3626 (1994). B. M. S. Zanardi. II: Partielle Diﬀerentialgleichungen. Jung. E. P. 746 (1998). Thomas. Phys. Rev. and P. Phys. 7. Table of Integrals. and R. A 29. 119 (1976). 382 (1982). Phys. Gradshteyn. 3637 (1992). On the Generators of Quantum Dynamical Semigroups. Izrailev. P. A systematic solution procedure for the FokkerPlanck equation of a Brownian particle in the high friction case. [91] D. Phys. 4736 (1995).
H¨nggi. 1987). Grossmann. [117] S.J. 77. B. Rep. Rev. Phys. P. Evolution and exact eigenstates of a resonant quantum system. Approximate constants of motion for classically chaotic vibrational dynamics: Vague tori. A 20. 571 (1992). Phys.J. [115] L.X. Phys. Vol. 1439 (1985). Physica D 21. Großmann and P. New York. Singapore. 217. Phys. A. A 34. H¨nggi. Zheng. Chem. Vol. C. Phys. [118] S. 5204 (1982). Moon and G. Math. On the relation between classical and quantummechanical entropy. 38 of Applied Mathical Sciences (Springer. [119] B. [111] D. J. semiclassical quantization. 1992). a Europhys. Chem. [113] M. 7 (1986). 55. F. Time Evolution and Eigenstates of a Quantum Iterative System.J. 145 (1997). [109] F. Wilkinson. CrossWell Chaos and Escape Phenomena in Driven Oscillators. Phys. 1. . Rev. Gorin. Lett. Tunneling in a Periodically a Driven Bistable System. Wehrl. 1983). Rev. Z. Korsch. Peres. p. Phys. edited by H. 315 (1991). 55. The fractal dimension of the twowell potential strange attractor. The Transition to Chaos: In Conservative and Classical Systems: Quantum Manifestations (Springer. Phasespace localization and level spacing distributions for a driven rotor with mixed regular/chaotic dynamics. Regular and Stochastic Motion. P. B. Mirbach. Wilkinson. Escande. 341 (1986). and classical intramolecular energy ﬂow. Reichl and W. Shi. Lin (World Scientiﬁc. SzemplinskaStupnicka. Lett. 635 (1987). New York. [110] A. Li. 269 (1985). Reinhardt. 165 (1985). Tunnelling between tori in phase space. J. H. [124] A. Shirts and W. 158 (1991). 67. [123] W. and B. [108] A.References 99 [107] F. Semiclassical quantization of KAM resonances in timeperiodic systems. Reps. E. Chang and K. Phys. 6579 (1994). 18. Lett. Mirbach and H. Physica D 17. Reichl. in Directions in Chaos. Lett. J. T. Jung.X. Stochasticity in classical Hamiltonian systems: universal aspects. Liebermann. J. 99 (1985). Lichtenberg and M. 17. [120] T. [112] L. [121] F. Phys. Dynamical quasidegeneracies and quantum tunneling. Phys. J. 353 (1979). C. and P. Li. Phys. [116] R. Localization in a Driven TwoLevel Dynamics. Nonlinear Dynamics 3. A 27. E. Shi. 16. Narrowly avoided crossings. [114] M. M. B 84. [122] F. 121. Rev. Fractal Basin Boundaries and Homoclinic Orbits for Periodic Motion in a TwoWell Potential. Korsch. 225 (1992).J. J. Dittrich. Moon and G. Chang and K.
Rev. Husimi. Gardiner. 177. Beiglb¨ck (Springer. P. [139] R. I. Habib. Bhaduri. Vol. Phys. 237 (1977). . o 1987). R. Berlin.) 167. Semiclassical mechanics in phase space: a study of Wigner’s function. edited by W. u Wiesbaden. Phys. On the Quantum Correction for Thermodynamic Equilibrium. Lett. [127] M. Quantum Brownian Motion and its Classical Limit. 40. Soc. (N. Rev. J. see Appendix C therein. 264 (1940). Coherent and Incoherent States of a Radiation Field. R. Messiah. Soc. Alicki and K. 286 of Lecture Notes in Physics. 106. (Springer. [134] C. Hillery. Ann. and J. 749 (1932). P. Einf¨hrung in die Grundlagen der Quantentheorie. Rev. Quantum Mechanics.100 References [125] A. New York. Phys. Vol. Proc. Vol. Paz. 400 (1991). [131] K. Phys. The Failure of the Quantum Regression Hypothesis. K. Rep. [138] E. [136] K. P. Density Operators and Quasiprobability Distributions. 1883 (1969). 1857 (1969). Rev. 2766 (1963). 131. [135] K. Ambegaokar. Rev. Phys. 13 of Springer Series in Synergetics. Brack and R. 121 (1984). 390 (1986). [140] P. Drummond and C.Y. Cahill and R. Gardiner. Phys. Ordered Expansions in Boson Amplitude Operators. Vol. [141] V. 1985). 96 of Frontiers in Physics (AddisonWesley. Japan 22. J. 2nd ed. Handbook of Stochastic Methods. S. F. Coherent states via decoherence. New York. 6th ed. Phys. Wigner. Berlin. [142] W. 3rd ed. Lendi. Phys. 1187 (1993). in Quantum Dynamical Semigroups and Applications. Lett. 70. Glauber. Wigner. H. W. Rev. O’Connell. 1965). 2353 (1980). M. Scully. Distribution Functions in Physics: Fundamentals. (Aula. 1 (1927). V. Sudarshan. 10. Z. G. Semiclassical Physics. Equivalence of semiclassical and quantum mechanical description of statistical light beams. Math. J. [130] E. Weyl. 277 (1963). Proc. Talkner. [133] M. Phys. E. Zurek. A 13. (Wiley & Sons. [128] R. Glauber. [137] H. [126] M. Cahill and R. Phys. E. 1988). Generalised P representations in quantum optics. [129] E. W. [132] P. Glauber. Fick. Berichte der Bunsengesellschaft 95. Berry. and E. 1997). 177. J. A 287. Phys. C. Quantenmechanik und Gruppentheorie. 46. J.
GertLudwig Ingold. I enjoyed many discussions on dissipative quantum mechanics and driven quantum systems with Milena Grifoni. and Andr´ Wobst. but not least. I gained a lot from their experience. Christine Zerbe provided the numerical code for the exact solution of the dissipative. and Haifa. W¨ rzburg. During the time I spent in Augsburg. Peter Schmitteckert. Di 511/1 and Di 511/2 as well as for the possibility to participate in conferences in Freiburg. Michael Thorwart. but also kept it (mostly :) well tuned. I had e lots of fruitful discussions about eﬃcient computing and objectoriented programming. Thomas Dittrich. Ludwig Hartmann. I would like to thank Prof. and Dietmar Weinmann. Berlin. Especially Gert has always been a competent and interested partner for discussions and questions during his Teerunde. u . even while staying at several remote places all over the world. Dr.Acknowledgment First. Dresden. Dr. and Sonja Thunnessen were of indispensible help in proofreading and improving the English of this thesis. I’m grateful to Thomas also for collaborating with me. With him. Peter H¨nggi and Prof. The members of the groups Theoretische Physik I and Theoretische Physik II— present and former ones—provided a stimulating and pleasant working atmosphere. Ralf Utermann not only built up a great computer environment. Ralf Utermann. Thomas Dittrich a for accepting me as a Doktorand and for giving me the opportunity to work on an intriguing project. parametrically driven harmonic oscillator. GertLudwig Ingold. I’m grateful to the DFGSchwerpunkt “Zeitabh¨ngige Ph¨noa a mene und Methoden in Quantensystemen der Physik und Chemie” for founding my position at the Universit¨t Augsburg from September ’95 to February ’99 under a grant no. Last.