Introduction to decoherence theory

Klaus Hornberger
arXiv:quant-ph/0612118v1 14 Dec 2006

Arnold Sommerfeld Center for Theoretical Physics

Ludwig-Maximilians-Universität München,
Theresienstraße 37, 80333 Munich, Germany

Dec 14, 2006

1 The concept of decoherence

This introduction to the theory of decoherence is aimed at readers with an

interest in the science of quantum information. There, one is usually content
with simple, abstract descriptions of non-unitary “quantum channels” to ac-
count for imperfections in quantum processing tasks. Nonetheless, in order to
justify such models of non-unitary evolution and to understand their limits
of applicability it is important to know their physical basis. Accordingly, I
will emphasize the dynamic and microscopic origins of the phenomenon of
decoherence, and will relate it to concepts from quantum information where
applicable, in particular to the theory of quantum measurement.
The study of decoherence, though based at the heart of quantum theory,
is a relatively young subject. It was initiated in the 1970’s and 1980’s with
the work of H. D. Zeh and W. Zurek on the emergence of classicality in the
quantum framework. Until that time the orthodox interpretation of quantum
mechanics dominated, with its strict distinction between the classical macro-
scopic world and the microscopic quantum realm. The mainstream attitude
concerning the boundary between the quantum and the classical was that this
was a purely philosophical problem, intangible by any physical analysis. This
changed with the understanding that there is no need for denying quantum
mechanics to hold even macroscopically, if one is only able to understand
within the framework of quantum mechanics why the macro-world appears
to be classical. For instance, macroscopic objects are found in approximate
position eigenstates of their center-of-mass, but never in superpositions of
macroscopically distinct positions. The original motivation for the study of
decoherence was to explain these effective super-selection rules and the ap-
parent emergence of classicality within quantum theory by appreciating the
crucial role played by the environment of a quantum system.
Hence, the relevant theoretical framework for the study of decoherence
is the theory of open quantum systems, which treats the effects of an un-
2 Klaus Hornberger

controllable environment on the quantum evolution. Originally developed to

incorporate the phenomena of friction and thermalization in the quantum
formulation, it has of course a much longer history than decoherence theory.
However, we will see that the intuition and approximations developed in the
traditional treatments of open quantum systems are not necessarily appropri-
ate to yield a correct description of decoherence effects, which may take place
on a time scale much shorter than typical relaxation phenomena. In a sense,
one may say that while the traditional treatments of open quantum systems
focus on how an environmental “bath” affects the system, the emphasis in
decoherence is more on the contrary question, namely how the system affects
and disturbs environmental degrees of freedom, thereby revealing information
about its state.
The physics of decoherence became very popular in the last decade, mainly
due to advances in experimental technology. In a number of experiments the
gradual emergence of classical properties in a quantum system could be ob-
served, in agreement with the predictions of decoherence theory. Needles to
say, a second important reason for the popularity of decoherence is its rel-
evance for quantum information processing tasks, where the coherence of a
relatively large quantum system has to be maintained over a long time.
Parts of these lecture notes are based on the books on decoherence by
E. Joos et al. [1] and on open quantum systems by H.-P. Breuer & F. Petruc-
cione [2], and on the lecture notes of W. Strunz [3]. Interpretational aspects,
which are not covered here, are discussed in [4, 5], and useful reviews by
W. H. Zurek and J. P. Paz can be found in [6, 7]. This chapter deals exclusively
with conventional, i.e., environmental decoherence, as opposed to spontaneous
reduction theories [8], which aim at “solving the measurement problem of
quantum mechanics” by modifying the Schrödinger equation. These models
are conceptually very different from environmental decoherence, though their
predictions of super-selection rules are often qualitatively similar.

1.1 Decoherence in a nutshell

Let us start by discussing the basic decoherence effect in a rather general

framework. As just mentioned, we need to account for the unavoidable cou-
pling of the quantum system to its environment. Although these environmen-
tal degrees of freedom are to be treated quantum mechanically, their state
must be taken unobservable for all practical purposes, be it due to their large
number or uncontrollable nature. In general, the detailed temporal dynamics
induced by the environmental interaction will be very complicated, but one
can get an idea of the basic effects by assuming that the interaction is suffi-
ciently short-ranged to admit a description in terms of scattering theory. In
this case, only the map between the asymptotically free states before and after
the interaction needs to be discussed, thus avoiding a temporal description of
the collision dynamics.
 Introduction to decoherence theory 3

Let the quantum state of a the system be described by the density oper-
ator ρ on the Hilbert space H. We take the system to interact with a single
environmental degree of freedom at a time–think of a phonon, a polaron, or a
gas particle scattering off your favorite implementation of a quantum register.
Moreover, let us assume, for the time being, that this environmental “particle”
is in a pure state ρE = |ψin ihψin |E , with |ψin iE ∈ HE . The scattering opera-
tor Stot maps between the in- and out-asymptotes in the total Hilbert space
Htot = H ⊗ HE , and for sufficiently short-ranged interaction potentials we
may identify those with the states before and after the collision. The initially
uncorrelated system and environment turn into a joint state,

[before collision] ρtot = ρ ⊗ |ψin ihψin |E (1)

[after collision] ρ′tot = Stot [ρ ⊗ |ψin ihψin |E ]S†tot . (2)

Now let us assume, in addition, that the interaction is non-invasive with

respect to a certain system property. This means that there is a number
of distinct system states, such that the environmental scattering off these
states causes no transitions in the system. For instance, if these distinguished
states correspond to the system being localized at well-defined sites then the
environmental particle should induce no hopping between the locations. In the
case of elastic scattering, on the other hand, it will be the energy eigenstates.
Denoting the set of these mutually orthogonal system states by {|ni} ∈ H, the
requirement of non-invasiveness means that Stot commutes with those states,
that is, it has the form
X
Stot = |nihn| ⊗ Sn , (3)
n

where the Sn are scattering operators acting in the environmental Hilbert

space. The insertion into (2) yields
X
ρ′tot = hm|ρ|ni|mihn| ⊗ Sm |ψin ihψin |E S†n (4)
m,n
X (m) (n)
≡ ρmn |mihn| ⊗ |ψout ihψout |E
m,n

and disregarding the environmental state by performing a partial trace we get

the system state after the interaction.
X
ρ′ = trE (ρ′tot ) = |mihn|ρmn hψin | S†n Sm |ψin iE (5)
m,n
| {z }
(n) (m)
hψout |ψout iE

Since the Sn are unitary the diagonal elements, or populations, are indeed
unaffected,

ρ′mm = ρmm , (6)

4 Klaus Hornberger

while the off-diagonal elements, or coherences, get multiplied by the overlap

of the environmental states scattered off the system states m and n,
(n) (m)
ρ′mn = ρmn hψout |ψout i. (7)
This factor has a modulus of less than one so that the coherences, which
characterize the ability of the system state to display a superposition between
|mi and |ni, get suppressed1 . It is important to note that this loss of coherence
occurs in a special basis, which is determined only by the scattering operator,
i.e., by the type of environmental interaction, and that it occurs to a degree
that is determined by both the the environmental state and the interaction.
This loss of the ability to show quantum behavior due to the interaction
with an environmental quantum degree of freedom is the basic effect of deco-
herence. One may view it as due to the arising correlation between the system
with the environment. After the interaction, the joint quantum state of system
and environment is no longer separable, and part of the coherence initially lo-
cated in the system now resides in the non-local correlation between system
and the environmental particle; it is lost once the environment is disregarded.
A complementary point of view argues that the interaction constitutes an in-
formation transfer from the system to the environment. The more the overlap
in (7) differs in magnitude from unity, the more an observer could in prin-
ciple learn about the system state by measuring the environmental particle.
Even though this measurement is never made, the complementarity princi-
ple then explains that the wave-like interference phenomenon characterized
by the coherences vanishes as more information discriminating the distinct,
“particle-like” system states is revealed.
To finish the introduction, here is a collection of other characteristics and
popular statements about the decoherence phenomenon. One often hears that
decoherence (i) can be extremely fast as compared to all other relevant time
scales, (ii) that it can be interpreted as an indirect measurement process, a
monitoring of the system by the environment, (iii) that it creates dynamically
a set of preferred states (“robust states” or “pointer states”) which seemingly
do not obey the superposition principle, thus providing the microscopic mech-
anism for the emergence of effective super-selection rules, and (iv ) that it helps
to understand the emergence of classicality in a quantum framework. These
points will be illustrated in the following, though (iii) has been demonstrated
only for very simple model systems and (iv ) depends to a fair extent on your
favored interpretation of quantum mechanics.

1.2 General scattering interaction

In the above demonstration of the decoherence effect the choice of the interac-
tion and the environmental state was rather special. Let us therefore now take
1
The value |ρmn | determines the maximal fringe visibility in a general interference
experiment involving the states |mi and |ni, as described by the projection on a
general superposition |ψθ,ϕ i = cos (θ) |mi+eiϕ sin(θ) |ni.
 Introduction to decoherence theory 5

Stot and ρE to be arbitrary and carry out the same analysis. P Performing the
trace (5) in the eigenbasis of the environmental state, ρE = ℓ pℓ |ψℓ ihψℓ | E ,
we have
 †
 X
ρ′ = trE Stot [ρ ⊗ ρE ]Stot = pℓ hψj |Stot |ψℓ iE ρhψℓ |S†tot |ψj iE
j,ℓ
X
= Wk ρWk† , (8)
k

where the hψj |Stot |ψℓ i are operators in H. After subsuming the two indices
j, ℓ into a single one, we get the second line with the Kraus operators Wk
given by
√
Wk = pℓk hψjk |Stot |ψℓk i (9)

It follows from the unitarity of Stot that they satisfy

X †
Wk Wk = I. (10)
k

This implies that Eq. (8) is the operator-sum representation of a completely

positive map Φ : ρ 7→ ρ′ (see Sect. 3.1). In other words, the scattering trans-
formation has the form of the most general evolution of a quantum state that
is compatible with the rules of quantum theory. In the operational formula-
tion of quantum mechanics this transformation is usually called a quantum
operation [9], the quantum information community likes to call it a quantum
channel . Conversely, given an arbitrary quantum channel, one can also con-
struct a scattering operator Stot and an environmental state ρE giving rise to
the transformation, though it is usually not very helpful from a physical point
of view to picture the action of a general, dissipative quantum channel as due
to a single scattering event.

1.3 Decoherence as an environmental monitoring process

We are now in a position to relate the decoherence of a quantum system to

the information it reveals to the environment. Since the formulation is based
on the notion of an indirect measurement it is necessary to first collect some
aspects of measurement theory [10, 11].

(a) Elements of general measurement theory

Projective measurements

This is the type of measurement discussed in standard textbooks of quantum

mechanics. A projective operator |αihα| ≡ Pα = P2α = P†α is attributed to each
6 Klaus Hornberger

possible outcome α of an idealized measurement apparatus. The probability

of the outcome α is obtained by the Born rule

Prob(α|ρ) = tr (Pα ρ) = hα|ρ|αi, (11)

and after the measurement of α the state of the quantum system is given by
the normalized projection

α Pα ρPα
ρ → M (ρ|α) = . (12)
tr (Pα ρ)

The basic requirement that the projectors form a resolution of the identity
operator,
X
Pα = I, (13)
α

ensures the normalization of the corresponding probability distribution Prob (α|ρ).

If the measured system property corresponds to a self-adjoint operator A
the Pα are the projectors into its eigenspaces, so that the expectation value is

hAi = tr (Aρ) .

If A has a continuous spectrum the outcomes are characterized by intervals

of a real parameter, and Rthe sum in (13) should be replaced by a projector-
valued Stieltjes integral dP(α) =I, or equivalently by a Lebesgue integral
over a projector-valued measure (PVM) [10, 11].
It is important to note that projective measurements are not the most
general type of measurement compatible with the rules of quantum mechanics.
In fact, non-destructive measurements of a quantum system are usually not
of the projective kind.

Generalized measurements

In the most general measurement situation, a positive (and therefore her-

mitian) operator Fα > 0 is attributed to each outcome α. Again, the collection
of operators corresponding to all possible outcomes must form a resolution of
the identity operator,
X
Fα = I. (14)
α

One speaks of a positive operator-valued measure

R (POVM), in particular in
the case of a continuous outcome parameter, dF(α) = I , and the probability
(or probability density in the continuous case) of outcome α is given by

Prob(α|ρ) = tr (Fα ρ) . (15)

 Introduction to decoherence theory 7

The effect on the system state of a generalized measurement is described by

a nonlinear transformation
P †
α k Mα,k ρ Mα,k
ρ → M (ρ|α) = (16)
tr (Fα ρ)
involving a norm-decreasing completely positive map in the numerator (see
Sect. 3.1) and a normalization which is subject to the consistency requirement
X †
Mα,k Mα,k = Fα . (17)
k

The operators Mα,k appearing in Eq. (16) are called measurement operators,
and they serve to characterize the measurement process completely. The Fα
are sometimes called “effects” or “measurement elements”. Note that different
measurement operators Mα,k can lead to the same measurement element Fα .
A simple class of generalized measurements are unsharp measurements,
where a number of projective operators are lumped together with probabilis-
tic weights in order to account for the finite resolution of a measurement device
or for classical noise in its signal processing. However, generalized measure-
ments schemes may also perform tasks which are seemingly impossible with
a projective measurement, such as the error-free discrimination of two non-
orthogonal states [12, 13].

Efficient measurements

A generalized measurement is called efficient if there is only a single summand

in (16) for each outcome α,

M ρ M†α
M (ρ|α) = α , (18)
tr M†α Mα ρ

implying that pure states are mapped to pure states. In a sense, these are
measurements where no unnecessary, that is no classical uncertainty is in-
troduced during the measurement process, see the following Sect. 1.3(b). By
means of a (left) polar decomposition and the consistency requirement (17)
efficient measurement operators have the form
p
Mα = Uα Fα , (19)

with an unitary operator Uα . This way the state after efficient measurement
can be expressed in a form which decomposes the transformation into a “raw
measurement” described by the Fα and a “measurement back-action” given
by the Uα . √ √
Fα ρ Fα
M (ρ|α) = Uα U†α (20)
|{z} tr (Fα ρ) |{z}
back-action | {z } back-action
“raw measurement”
8 Klaus Hornberger
√
In this transformation the positive operators Fα “squeeze” the state
along the measured property and expand it along the other, complementary
ones, similar to what a projector would do, while the back-action operators
Uα “kick” the state by transforming it in a way that is reversible, in principle,
provided the outcome α is known. Note that the projective measurements (12)
are a sub-class in the set of back-action-free efficient measurements.

(b) Indirect measurements

In an indirect measurement one tries to obtain information about the sys-

tem in a way that disturbs it as little as possible. This is done by letting
a well-prepared microscopic quantum probe interact with the system. This
probe is then measured by projection, i.e., destructively, so that one can infer
properties of the system without having it brought into contact with a macro-
scopic measurement device. Let ρprobe be the prepared state of the probe, Stot
describe the interaction between system and probe, and Pα be the projectors
corresponding to the various outcomes of the probe measurement. The prob-
ability of measuring α is determined by the reduced state of the probe after
interaction, i.e.,

 
Prob(α|ρ) = trp Pα ρ′probe = trp robe Pα trsys (Stot [ρ ⊗ ρprobe ]S†tot ) . (21)

By pulling out the system trace (extending the projectors to Htot = H ⊗ Hp )

and using the cyclic permutability of operators under the trace we have
 
Prob(α|ρ) = tr S†tot [I ⊗ Pα ]Stot [ρ ⊗ ρprobe ] = tr (Fα ρ) (22)

with microscopically defined measurement elements

 
Fα = trprobe S†tot [I ⊗ Pα ]Stot [I ⊗ ρprobe ] > 0 (23)
P
satisfying α Fα = I . Since the probe measurement is projective, we can
also specify the new system state conditioned on the click at α of the probe
detector,

M (ρ|α) = trprobe [Mtot (ρtot |α)]

!
[I ⊗ Pα ]Stot [ρ ⊗ ρprobe ] S†tot [I ⊗ Pα ]
= trprobe
tr(Pα [ρ ⊗ ρprobe ])
X Mα,k ρM†α,k
= . (24)
tr (Fα ρ)
k

In the last step a convex decomposition

P of the initial probe state into pure
states was inserted, ρprobe = k wk |ψk ihψk | . Taking Pα = |αihα| we thus get
a microscopic description also of the measurement operators,
 Introduction to decoherence theory 9
√
Mα,k = pk hα|Stot |ψk i. (25)

This shows that an indirect measurement is efficient (as defined above) if the
probe is initially in a pure state, i.e., if there is no uncertainty introduced
in the measurement process, apart form the one imposed by the uncertainty
relations on ρprobe .
If we know that an indirect measurement has taken place, but do not know
its outcome α we have to resort to a probabilistic (Bayesian) description of the
new system state. It is given by the sum over all possible outcomes weighted
by their respective probabilities,
X X
ρ′ = Prob(α|ρ)M (ρ|α) = Mα,k ρM†α,k . (26)
α α,k

This form is the same as above in Eqs. (8) and (9), where the basic effect of
decoherence has been described. This indicates that the decoherence process
can be legitimately viewed as a consequence of the information transfer from
the system to the environment. The complementarity principle can then be
invoked to understand which particular system properties lose their quantum
behavior, namely those complementary to the ones revealed to the environ-
ment. This “monitoring interpretation” of the decoherence process will help
us below to derive microscopic master equations.
1.4 A few words on nomenclature

Since decoherence phenomena show up in quite different sub-communities of

physics, a certain confusion and lack of uniformity developed in the terminol-
ogy. This is exacerbated by the fact that decoherence often reveals itself as
a loss of fringe visibility in interference experiments—a phenomenon, though,
which may have other causes than decoherence proper. Here is an attempt of
clarification:
• decoherence: In the original sense, an environmental quantum effect af-
fecting macroscopically distinct states. The term is nowadays applied to
mesoscopically different states, as well, and even for microscopic states, as
long as it refers to the quantum effect of environmental, i.e. in practice
unobservable, degrees of freedom.
However, the term is often (ab-)used for any process other reducing the
purity of a micro-state.
• dephasing: In a narrow sense, this describes the phenomenon that coher-
ences, i.e., the off-diagonal elements of the density matrix, get reduced in a
particular basis, namely the energy eigenbasis of the system. It is a state-
ment about the effect, and not the cause. In particular, dephasing may
be reversible if it is not due to decoherence, as revealed e.g. in spin-echo
experiments.
This phrase should be treated with great care since it is used differently
in various sub-communities. It is taken as a synonym to “dispersion” in
10 Klaus Hornberger

molecular physics and in nonlinear optics, as a synonym to “decoherence”

in condensed matter, and often as a synonym to “phase averaging” in
matter wave optics. It is also called a T2 -process in NMR and in condensed
matter physics (see below).
• phase averaging: A classical noise phenomenon entering through the de-
pendence of the unitary system evolution on external control parameters
which fluctuate (parametric average over unitary evolutions). A typical
example are the vibrations of an interferometer grating or the fluctuations
of the classical magnetic field in an electron interferometer due to techni-
cal noise. Empirically, phase averaging is often hard to distinguish from
decoherence proper.
• dispersion: Coherent broadening of wave packets during the unitary evo-
lution, e.g. due to a velocity dependent group velocity or non-harmonic
energy spacings. This unitary effect may lead to a reduction of signal os-
cillations, for instance, in molecular pump-probe experiments.
• dissipation: Energy exchange with the environment leading to thermal-
ization. Usually accompanied by decoherence, but see Sect. 3.4(b) for a
counter-example.

2 Case study: Dephasing of qubits

So far, the discussion of the temporal dynamics of the decoherence process was
circumvented by using a scattering description. Before going to the general
treatment of open quantum systems in Sect. 3 it is helpful to take a closer
look on the time evolution of a special system, where the interaction with a
model environment can be treated exactly [14, 2].

2.1 An exactly solvable model

Let us take a two-level system, or qubit, described by the Pauli spin operator
σz , and model the environment as a collection of bosonic field modes. In
practice, such fields can yield an appropriate effective description even if the
actual environment looks quite differently, in particular if the environmental
coupling is a sum of many small contributions2 . What is fairly non-generic in
the present model is the type of coupling between system and environment,
which is taken to commute with the system Hamiltonian.
The total Hamiltonian thus reads
~ω X X † 
Htot = σz + ~ωk b†k bk +σz gk bk + gk∗ bk (27)
2
k k
| {z } | {z }
H0 Hint

2
A counter-example would be the presence of a degenerate environmental degree
of freedom, such as a bistable fluctuator.
 Introduction to decoherence theory 11

with the usual commutation relation for the mode operators of the bosonic
field modes, [bi , b†k ] = δik , and coupling constants gk . The fact that the sys-
tem Hamiltonian commutes with the interaction, guarantees that there is no
energy exchange between system and environment so that we expect pure
dephasing.
By going into the interaction picture one transfers the trivial time evo-
lution generated by H0 to the operators (and indicates this with a tilde). In
particular,
X
e int (t) = eiH0 t/~ Hint e−iH0 t/~ = σz
H (gk eiωk t b†k + gk∗ e−iωk t bk ) , (28)
k

where the second equality is granted by the commutation [σz , Hint ] = 0. The
time evolution due to this Hamiltonian can be formally expressed as a Dyson
series,
 Z 
e i t ′e ′
U(t) = T← exp − dt Hint (t )
~ 0
X∞  n Z t h i
1 1 e int (t1 ) · · · H
e int (tn ) ,
= dt1 · · · dtn T← H (29)
n=0
n! i~ 0

where T← is the time ordering operator (putting the operators with larger
time arguments to the left). Due to this time ordering requirement the series
usually cannot be evaluated exactly (if it converges at all). However, in the
present case the commutator of H e int at different times is not an operator, but
just a c-number,
X 2
e int (t) , H
[H e int (t′ )] = 2i |gk | sin ( ωk (t′ − t)) . (30)
k

As a consequence, the time evolution differs only by a time-dependent phase

from the one obtained by casting the operators in their natural order,
 Z 
e iϕ(t) i t ′e ′
U(t) = e exp − dt Hint (t ) . (31)
~ 0
This is discussed in the note3 , which also shows that the phase is given by
Z t Z t h i
i e int (t1 ) , H
e int (t2 ) .
ϕ (t) = 2 dt1 dt2 Θ (t1 − t2 ) H (32)
2~ 0 0
3 e (t) for the case [H
To obtain the time evolution U e (t) , H I
e (t′ )] = c define the
operators
Z
1 t ′ e ` ′´
Φ (t) = dt H t
~ 0
e (t). This way U (t) describes the “additional” motion due
and U (t) = exp[iΦ (t)]U
to the time ordering requirement. It satisfies
12 Klaus Hornberger

One can now perform the integral over the interaction Hamiltonian to get
!
1 X †

e =e
U(t) iϕ(t)
exp σz ∗
αk (t)bk − αk (t)bk (33)
2
k

with complex, time-dependent functions

1 − eiωk t
αk (t) := 2gk . (34)
~ωk
e
The operator U(t) is diagonal in the eigenbasis of the system, and it describes
how the environmental dynamics depends on the state of the system. In par-
ticular, if the system is initially in the upper level, |ψi = | ↑i, one has
Y  
e αk (t)
U(t)| ↑i|ξ0 iE = eiϕ(t) | ↑i Dk |ξ0 i =: eiϕ(t) | ↑i|ξ↑ (t)iE , (35)
2
k

and for the lower state

Y  
e αk (t)
U(t)| ↓i|ξ0 iE = eiϕ(t) | ↓i Dk − |ξ0 i =: eiϕ(t) | ↓i|ξ↓ (t)iE . (36)
2
k

Here we introduced the unitary displacement operator s for the k-th field mode,

Dk (α) = exp(αb†k − α∗ bk ), (37)

which effect a translation of the field state in its attributed phase space. In
particular, the coherent state |αik of the field mode k is obtained from its
ground state |0ik by |αik := Dk (α)|0ik [15]
The equations (35) and (36) show that the collective state of the field
modes gets displaced by the interaction with the system and that the sense
of the displacement is determined by the system state.
„» – «
d iΦ(t) −iΦ(t) 1 e (t) e−iΦ(t) U (t) .
∂t U (t) = e e + eiΦ(t) H
dt i~

The derivative in square brackets has to be evaluated with care since the H e (t) do
not commute at different times. By first showing that [A, ∂t A] = c ∈ C
implies
∂t An = nAn−1 ∂t A − 12 n (n − 1) cAn−2 one finds
»Z t –
d iΦ(t) −1 iΦ(t) e 1 ` ´
e = e H (t) + 2 eiΦ(t) e t′ , H
dt′ H e (t) .
dt i~ 2~ 0
` 2 ´−1 R t ′
Therefore, we have ∂t U (t) = 2~ dt [He (t′ ) , H
e (t)] U (t), which can be inte-
0
grated to yield finally
„ Z t Z t1 h i«
e (t) = exp − 1
U dt 1 dt 2 He (t 1 ) , e
H (t 2 ) e−iΦ(t) .
2~2 0 0
 Introduction to decoherence theory 13

Assuming that the states of system and environment are initially uncor-
related, ρtot (0) = ρ ⊗ ρE , the time-evolved system state reads4
 
e
ρ̃(t) = trE U(t)[ρ e † (t) .
⊗ ρE ]U (38)

It follows from (35) and (36) that the populations are unaffected,

h↑ |ρ̃(t)| ↑i = h↑ |ρ̃(0)| ↑i,

h↓ |ρ̃(t)| ↓i = h↓ |ρ̃(0)| ↓i,

while the coherences are suppressed by a factor which given by the trace over
the displaced initial field state,
!
Y
h↑ |ρ̃(t)| ↓i = h↑ |ρ̃(0)| ↓i trE Dk (αk (t))ρE . (39)
k
| {z }
χ(t)

Incidentally, the complex suppression factor χ (t) is equal to the Wigner char-
acteristic function of the original environmental state at the points αk (t), i.e.,
it is given the Fourier transform of its Wigner function [16].

(a) Initial vacuum state

Q
If the environment is initially in its vacuum state, ρE = k |0ih0|k , the
|ξ↑ (t)iE and |ξ↓ iE defined in (35), (36) turn into multi-mode coherent states,
and the suppression factor can be calculated immediately to yield:
!
Y Y 2
|αk (t)|
χvac (t) = h0|Dk (αk (t))|0ik = exp −
2
k k
!
X 2 1 − cos (ωk t)
= exp − 4 |gk | (40)
~2 ωk2
k

For times that are short compared to the field dynamics, t ≪ ωk−1 , one ob-
serves a Gaussian decay of the coherences. Modifications to this become rele-
vant at ωk t ∼
2
= 1, provided χ (t) is then still appreciable, i.e., for 4 |gk | /~2 ωk2 ≪
1. Being a sum over periodic functions, χvac (t) is quasi-periodic, that is, it will
come back arbitrarily close to unity after a large period (which increases ex-
ponentially with the number of modes). These somewhat artificial Poincaré
recurrences vanish if we replace the sum over the discrete modes by an integral
over a continuum with mode density µ,
4
In fact, the assumption ρtot (0) = ρ ⊗ ρE is quite unrealistic if the coupling is
strong, as discussed below. Nonetheless, and it certainly represents a valid initial
state.
14 Klaus Hornberger
X Z ∞
f (ωk ) −→ dωµ(ω)f (ω), (41)
k 0

for any function f . This way the coupling constants gk get replaced by the
spectral density,
2
J(ω) = 4µ(ω) |g(ω)| . (42)
This function characterizes the environment by telling how effective the cou-
pling is at a certain frequency.
(b) Thermal state

If the environment is in a thermal state with temperature T ,

e−HE /kB T Y X∞

−~ωk /kT
ρE = ρth =
= 1 − e e−~ωk n/kB T |nihn|k,
tr e−HE /kB T k n=0
| {z }
(k)
=ρth
(43)
the suppression factor reads5
Y   Y 2  !
(k) |αk (t)| ~ωk
χ(t) = tr Dk (αk (t))ρth = exp − coth . (44)
2 2kB T
k k

This factor can be separated into its vacuum component (40) and a ther-
mal component, χ(t) = e−Fvac (t) e−Fth (t) , with the following definitions of the
vacuum and the thermal decay functions:
X 1 − cos (ωk t)
Fvac (t) := 4 |gk |2 (45)
~2 ωk2
k
X    
2 1 − cos (ωk t) ~ωk
Fth (t) := 4 |gk | coth −1 (46)
~2 ωk2 2kB T
k

2.2 The continuum limit

Assuming that the field modes are sufficiently dense we replace their sum by
an integration. Noting (41), (42) we have
Z ∞
1 − cos (ωt)
Fvac (t) −→ dωJ(ω) (47)
~2 ω 2
Z0 ∞    
1 − cos(ωt) ~ω
Fth (t) −→ dωJ(ω) 2ω2
coth −1 . (48)
0 ~ 2kBT
5
This` can be found
´ in a small exercise `by using
´ the Baker-Hausdorff relation with
exp αb† − α∗ b = exp(−|α|2 /2) exp αb† exp (−α∗ b), and the fact that coher-
ent states satisfy the eigenvalue equation
√ b|βi = β|βi, have the number rep-
2 n
resentation hn|βi = exp(−|β| /2)β / n!, and form an over-complete set with
I R
= π −1 d2 β|β ihβ|.
 Introduction to decoherence theory 15

So far, the treatment was exact. To continue we have to specify

√ the spectral
density in the continuum limit. A typical model takes g ∝ ω, so that the
spectral density of a d-dimensional field can be written as [17]
 d−1
ω
J(ω) = aω e−ω/ωc (49)
ωc
with “damping strength” a > 0. Here, ωc is a characteristic frequency “cutoff”
where the coupling decreases rapidly, such as the Debye frequency in the case
of phonons.

(a) Ohmic coupling

For d = 1 the spectral density (49) increases linearly at small ω (“Ohmic

coupling”). One finds
a
Fvac (t) = log(1 + ωc2 t2 ), (50)
2~2
which bears a strong ωc dependence. Evaluating the second integral requires to
assume that the cutoff ωc is large compared to the thermal energy, kT ≪ ~ωc :
 
a sinh (t/tT )
Fth (t) ≃ 2 log (51)
~ t/tT

Here tT = ~/ (πkB T ) is a thermal quantum time scale. The corresponding

frequency ω1 = 2/tT is called the (first) Matsubara frequency, which also
shows up if imaginary time path integral techniques are used to treat the
influence of bosonic field couplings [17].
For large times the decay function Fth (t) shows the asymptotic behavior
a t
Fth (t) ∼ [as t → ∞]. (52)
~ 2 tT
It follows that the decay of coherence is characterized by rather different
regimes. In the short time regime (t < ωc−1 ) we have the perturbative behavior
a 2 2
F (t) ≃ ω t [for t ≪ ωc−1 ], (53)
2~2 c
which can also be obtained from the short-time expansion of the time evolution
operator. Note that the decay is here determined by the overall width ωc of
the spectral density. The intermediate region, ωc−1 < t < ω1−1 , is dominated
by Fvac (t) and called the vacuum regime,
a
F (t) ≃ log (ωc t) [for ωc−1 ≪ t ≪ ω1−1 ].
~2
Beyond that, for large times the decay is dominated by the thermal suppres-
sion factor,
16 Klaus Hornberger

a t
F (t) ≃ [for ω1−1 ≪ t]. (54)
~ 2 tT
In this thermal regime the decay shows the exponential behavior typical for
the Markovian master equations discussed below. Note that the decay rate
for this long time behavior is determined by the low frequency behavior of
the spectral density, characterized by the damping strength a in (49), and is
proportional to the temperature T .

(b) Super-Ohmic coupling

For d = 3, the case of a “super-Ohmic” bath, the integrals (47), (48) can
be calculated without approximation. We note only the long-time behavior of
the decay.
 2  
kB T ′ kB T
lim F (t) = 2a ψ 1+ <∞ (55)
t→∞ ~ωc ~ωc
Here ψ (z) stands for the Digamma function, the logarithmic derivative of the
gamma function. Somewhat surprisingly, the coherences do not get completely
reduced as t → ∞, even at a finite temperature. This is due to the suppressed
influence of the important low-frequency contributions to the spectral density
in three dimensions (as compared to lower dimensions). While such a suppres-
sion of decoherence is plausible for intermediate times, the limiting behavior
(55) is clearly a result of our simplified model assumptions. It will be absent if
there is a small anharmonic coupling between the bath modes [18] or if there
is a small admixture of different couplings to Hint .

2.3 Dephasing of N qubits

Let us now discuss the generalization to the case of N qubits, which do not
interact directly among each other. Each qubit may have a different coupling
to the bath modes.
N
X −1
~ωj (j) X
N
X −1 X  (j) † (j)

Htot = σz + ~ωk b†k bk + σz(j) gk bk + [gk ]∗ bk (56)
j=0
2 j=0
k k
| {z }
H0

Similar to above the time evolution in the interaction picture reads

 
1
N
X −1 X  (j) † (j)

e = eiϕ(t) exp 
U(t) σ (j) αk bk − [αk ]∗ bk  ,
2 j=0 z
k

where the displacement of the field modes now depends on the N -qubit state.
As an example, take N = 2 qubits and only a single vacuum mode. For
the states
 Introduction to decoherence theory 17

|φi = c11 | ↑↑i + c00 | ↓↓i (57)

and

|ψi = c10 | ↑↓i + c01 | ↓↑i (58)

we obtain
(1)
e α (t) + α(2) (t) −α(1) (t) − α(2) (t)
U|φi|0iE = c11 | ↑↑i| iE + c00 | ↓↓i| iE
2 2
and similarly
(1)
e α (t) − α(2) (t) −α(1) (t) + α(2) (t)
U|ψi|0iE = c10 | ↑↓i| iE + c01 | ↓↑i| iE ,
2 2
where the α(1) (t) and α(2) (t) are the field displacements (34) due to the first
and the second qubit.
If the couplings to the environment are equal for both qubits, say, because
they are all sitting in the same place and seeing the same field, we have
α(1) (t) = α(2) (t) ≡ α (t). In this case states of the form |φi are decohered
2
once the factor hα (t) | − α (t)iE = exp(−2 |α (t)| ) ≃ 0. States of the form
|ψi, on the other hand, are not affected at all, and one says that the {|ψi}
span a (two-dimensional) decoherence-free subspace. It shows up because the
environment cannot tell the difference between the states | ↑↓i and | ↓↑i if it
couples only to the sum of the excitations.
In general, using the binary notation, e.g., | ↑↓↑i ≡ |1012 i = |5i, one has

hm|ρ̃(t)|ni = hm|ρ̃(0)|ni (59)

   
N
X −1 X  (j) (j)

× tr exp  (mj − nj ) αk (t)b†k − [αk (t)]∗ bk  ρth 
j=0 k

where mj ∈ {0, 1} indicates the j-th digit in the binary representation of the
number m.
We can distinguish different limiting cases:

(a) Qubits feel the same reservoir

If the separation of the qubits is small compared to the wave lengths of the
(j)
field modes they are effectively interacting with the same reservoir, αk = αk .
One can push the j-summation to the α’s in thisPcase, so that, compared to
the single qubit, one merely has to replace αk by (mj − nj )αk . We find
  2 
PN −1 
χmn (t) = exp −  j=0 (mj − nj )  (Fvac (t) + Fth (t)) (60)

with Fvac (t) and Fth (t) given by Eqs. (47) and (48).
18 Klaus Hornberger

Hence, in the worst case, one observes an increase of the decay rate by
N 2 compared to the single qubit rate. This is the case for the coherence
between the states |0i and |2N − 1i, which have the maximum difference in
the number of excitations. On the other hand, the states with an equal number
of excitations form a decoherence-free

subspace in the present model, with a
N
maximal dimension of N/2 .

(b) Qubits see different reservoirs

In the other extreme, the qubits are so far apart from each other that each
field mode couples only to a single qubit. This suggests a re-numbering of the
field modes,
(j)
αk −→ αkj ,
and leads, after transforming the j-summation into a tensor-product, to
 
NY−1 Y
(kj )
χmn (t) = tr  Dkj (mj − nj ) αkj (t) ρth 
j=0 kj
 
N
Y −1
 2 
= exp −|mj − nj | (Fvac (t) + Fth (t))
j=0
| {z }
=|mj −nj |
 
 
 NX −1 
 
= exp − |mj − nj | (Fvac (t) + Fth (t)) . (61)
 
 j=0 
| {z }
Hamming distance

Hence, the decay of coherence is the same for all pairs of states with the
same Hamming distance. In the worst case, we have an increase by a factor
of N compared to the single qubit case, and there are no decoherence-free
subspaces.
An intermediate case is obtained if the coupling depends on the position r j
of the qubits. A reasonable model, corresponding to point scatterings of fields
(j)
with wave vector k, is given by gk = gk exp (ik · rj ), and its implications are
studied in [19].
The model for decoherence discussed in this section is rather exceptional
in that the dynamics of the system can be calculated exactly for some choices
of the environmental spectral density. In general, one has to resort to approx-
imate descriptions for the dynamical effect of the environment, and we turn
to this problem in the following section.
 Introduction to decoherence theory 19

3 Markovian dynamics of open quantum systems

Isolated systems evolve, in the Schrödinger picture and for the general case of
mixed states, according to the von Neumann equation,
1
∂t ρ = [H, ρ]. (62)
i~
One would like to have a similar differential equation for the reduced dynamics
of an “open” quantum system, which is in contact with its environment. If we
extend the description to include the entire environment HE and its coupling
to the system, then the total state in Htot = H ⊗ HE evolves unitarily. The
partial trace over HE gives the evolved system state, and its time derivative
reads
d   1
∂t ρ = trE Utot (t)ρtot (0)U†tot (t) = trE ([Htot , ρtot ]) . (63)
dt i~
This exact equation is not closed and therefore not particularly helpful as it
stands. However, it can be used as the starting point to derive approximate
time evolution equations for ρ, in particular, if it is permissible to take the
initial system state to be uncorrelated with the environment.
These equations are often non-local in time, though, in agreement with
causality, the change of the state at each point in time depends only on the
state evolution in the past. In this case, the evolution equation is called a
generalized master equation. It can be specified in terms of superoperator-
functionals, i.e., linear operators which take the density operator ρ with its
past time-evolution until time t and map it to the differential change of the
operator at that time,
∂t ρ = K [{ρτ : τ < t}] . (64)
An interpretation of this dependence on the system’s past is that the environ-
ment has a memory, since it affects the system in a way which depends on
the history of the system-environment interaction. One may hope that on a
coarse-grained time-scale, which is large compared to the inter-environmental
correlation times, these memory effects might become irrelevant. In this case, a
proper master equation might be appropriate, where the infinitesimal change
of ρ depends only on the instantaneous system state, through a Liouville
super-operator L,
∂t ρ = Lρ. (65)
Master equations of this type are also called Markovian, because of their re-
semblance to the differential Chapman-Kolmogorov equation for a classical
Markov process. However, since a number of approximations are involved in
their derivation, it is not clear whether the corresponding time evolution re-
spects important properties of a quantum state, such as its positivity. We will
see that these constraints restrict the possible form of L rather strongly.
20 Klaus Hornberger

3.1 Quantum dynamical semigroups

The notion a of quantum dynamical semigroup permits a rigorous formulation

of the Markov assumption in quantum theory. To introduce it we first need a
number of concepts from the theory of open quantum systems [11, 20, 21, 22].

(a) Dynamical maps

A dynamical map is a one-parameter family of trace-preserving, convex linear,

and completely positive maps (CPM)
Wt : ρ0 7→ ρt , for t ∈ R+
0 (66)
satisfying W0 = id. As such it yields the most general description of a time
evolution which maps an arbitrary initial state ρ0 to valid states at later times.
Specifically, the condition of trace preservation guarantees the normaliza-
tion of the state,
tr (ρt ) = 1,
and the convex linearity, i.e.,
Wt (λρ0 + (1 − λ) ρ′0 ) = λWt (ρ0 ) + (1 − λ) Wt (ρ′0 ) for all 0 6 λ 6 1
ensures that the transformation of mixed states is consistent with the classical
notion of ignorance. The final requirement of complete positivity is stronger
than mere positivity of Wt (ρ0 ). It means that in addition all the tensor prod-
uct extensions of Wt to spaces of higher dimension, defined with the identity
map idext , are positive,
Wt ⊗ idext > 0,
that is, the image of any positive operator in the higher dimensional space
is again a positive operator. This guarantees that the system state remains
positive even if it is the reduced part of a non-separable state evolving in a
higher dimensional space.

(b) Kraus representation

Any dynamical map admits a operator-sum representation of the form (8)

[22],
N
X
Wt (ρ) = Wk (t)ρWk† (t) , (67)
k=1
6
with the completeness relation
6
In case of a trace-decreasing, convex linear, completely positive map
P the condition
P
I I
(68) is replaced by k Wk† (t)Wk (t) < , i.e., the operator − k Wk† (t)Wk (t)
must be positive.
 Introduction to decoherence theory 21
N
X
Wk† (t)Wk (t) = I. (68)
k=1

The number of the required Kraus operators Wk (t) is limited by the dimension
2
of the system Hilbert space, N 6 dim (H) (and confined to a countable set
in case of an infinite-dimensional, separable Hilbert space), but their choice is
not unique.

(c) Semigroup assumption

We can now formulate the assumption that the {Wt : t ∈ R+

0 } form a
continuous dynamical semigroup 7 [20, 22]:
!
Wt2 (Wt1 (·)) = Wt1 +t2 for all t1 , t2 > 0 (69)

This statement is rather strong, and it is certainly violated for truly micro-
scopic times. But it seems not unreasonable on the level of a coarse-grained
time scale, which is long compared to the time it takes for the environment
to “forget” the past interactions with the system due to the dispersion of
correlations into the many environmental degrees of freedom.
For a given dynamical semigroup there exists, under rather weak condi-
tions, a generator, i.e., a superoperator L satisfying

Wt = eLt for t > 0. (70)

In this case Wt (ρ) is the formal solution of the Markovian master equation
(65).

(d) Dual maps

So far we used the Schrödinger picture, i.e., the notion that the state of an
open quantum system evolves in time, ρt = Wt (ρ0 ). Like in the description of
closed quantum systems, one can also take the Heisenberg point of view, where
the state does not evolve, while the operators A describing observables acquire
a time dependence. The corresponding map Wt♯ : A0 7→ At is called the dual
map, and it is related to Wt by the requirement tr(AWt (ρ)) = tr(ρWt♯ (A)).


In case of a dynamical semigroup, Wt♯ = exp L♯ t , the equation of motion
takes the form ∂t A = L♯ A, with the dual Liouville operator determined by
tr(AL (ρ)) = tr(ρL♯ (A)). From a mathematical point of view, the Heisenberg
picture is much more convenient since the observables form an algebra, and
it is therefore preferred in the mathematical literature.

7
The inverse element required for a group structure is missing for general, irre-
versible CPMs.
22 Klaus Hornberger

3.2 The Lindblad form

We can now derive the general form of the generator of a dynamical semigroup,
taking dim (H) = d < ∞ for simplicity [2, 22]. The bounded operators on H
then form a d2 -dimensional vector space which turns into a Hilbert space, if
equipped with the Hilbert-Schmidt scalar product (A, B) := tr(A† B).
Given an orthonormal basis of operators {Ej : 1 6 j 6 d2 } ⊂ L (H),

(Ei , Ej ) := tr(E†i Ej ) = δij , (71)

any Hilbert-Schmidt operator Wk can be expanded as

2
d
X
Wk = (Ej , Wk )Ej . (72)
j=1

We can choose one of the basis operators, say the d2 -th, to be proportional
to the identity operator,
1
Ed2 = √ I (73)
d
so that all other basis elements are traceless,

0 for j = 1, . . . , d2 − 1
tr(Ej ) = √ (74)
d for j = d2 .

Representing the superoperator of the dynamical map (67) in the {Ej } basis
we have
d2
X
Wt (ρ) = cij (t)Ei ρE†j (75)
i,j=1

with a time dependent, hermitian and positive coefficient matrix,

N
X
cij (t) = (Ei , Wk (t))(Ej , Wk† (t)). (76)
k=1

(Positivity can be checked in a small calculation.) We can now calculate the

semigroup generator in terms of the differential quotient by writing the terms
including the element Ed2 separately.
 Introduction to decoherence theory 23

Wτ (ρ) − ρ
Lρ = lim
τ →0 τ
2 2
1 dX −1 dX−1
d cd d (τ )
2 2 −1 cjd2 (τ ) cd2 j (τ ) †
= lim ρ + lim √ Ej ρ + ρ lim √ Ej
{z τ dτ dτ
τ →0 τ →0 τ →0
| } j=1 j=1
c0 ∈ R | {z } | {z }
B∈L(H) B† ∈L(H)
2
dX −1
cij (τ )
+ lim Ei ρE†j
i,j=1 |
τ →0 τ
{z }
αij ∈ R
2
dX −1
= c0 ρ + Bρ + ρB + †
αij Ei ρE†j
i,j=1
2
dX −1
1 1
= [H, ρ] + (Gρ + ρG) + αij Ei ρE†j (77)
i~ ~ i,j=1

In the last equality the following hermitian operators with the dimension of
an energy were introduced:
~
G= (B + B† + c0 )
2
~
H = (B − B† )
2i
By observing that the conservation of the trace implies tr(Lρ) = 0 one can
relate the operator G to the matrix α = (αij ), since
 2
 
dX −1
2G †
0 = tr(Lρ) = 0 + tr  + αij Ej Ei  ρ
~ i,j=1

must hold for all ρ. It follows that

2
d −1
~ X
G=− αij E†j Ei .
2 i,j=1

This leads to the first standard form for the generator of a dynamical semi-
group:
dX2
−1  
1 1 1
Lρ = [H, ρ] + αij Ei ρE†j − E†j Ei ρ − ρE†j Ei (78)
|i~ {z } i,j=1 2 2
unitary part | {z }
incoherent part

The complex coefficients αij have dimensions of frequency and constitute a

positive matrix α.
24 Klaus Hornberger

The second standard form or Lindblad form is obtained by diagonaliz-

ing the coefficient matrix α. The corresponding unitary matrix U satisfy-
ing U αU † = diag(γ1 , . . . , γd2 −1 ) allows to define the dimensionless operators
Pd2 −1 † Pd2 −1
Lk := j=1 Ej Ujk so that Ej = k=1 Lk Ukj and therefore8

N
X  
1 1 1
Lρ = [H, ρ] + γk Lk ρL†k − L†k Lk ρ − ρL†k Lk (79)
i~ 2 2
k=1

with N 6 d2 − 1. This shows that the general form of a generator of a dy-

namical semigroup is specified by a single hermitian operator H, which is not
necessarily equal to the Hamiltonian of the isolated system, see below, and at
most d2 − 1 arbitrary operators Lk with attributed positive rates γk . These
are called Lindblad operators or jump operators, a name motivated in the
following section.
Lindblad showed in 1976 that this is also valid for infinite-dimensional
systems as long as the generator L is bounded (which is usually not the case).
The Lindblad operators for a given generator are not unique. In fact the
equation is invariant under the transformation

Lk → Lk + c k (80)
~ X ∗ 
H→H+ cj Lj − cj L†j + |c0 |, (81)
2i j

so that the Lk can be chosen traceless. In this case the only remaining freedom
is unitary rotation, X
√ √
γi Li −→ Uij γj Lj . (82)
j

If L shows an additional invariance (e.g. with respect to rotations or transla-

tions) the form of the Lindblad operators will be further restricted [23].

3.3 Unravelling of Markovian master equations

Generally, if we write the Liouville superoperator L as the sum of two parts,

L0 and S, then the formal solution (70) of the master equation (65) can be
expressed as

8
It is easy to see that the dual Liouville operator discussed in Sect. 3.1(d) reads,
in Lindblad form,

XN „ «
♯ 1 † 1 † 1 †
L (A) = [A, H] + γk Lk ALk − Lk Lk A − ALk Lk .
i~ k=1
2 2

I I
Note that this implies L♯ ( ) = 0, while L ( ) =
P
k γk [Lk , L†k ] and tr (LX) = 0.
 Introduction to decoherence theory 25
∞ n
X t
Wt = e(L0 +S)t = (L0 + S)n
n=0
n!
∞
X ∞
X
P
tn+ j kj
=  P  Lk0n SLk0n−1 S · · · SLk01 SLk00
n + | {z }
n=0 k0 ,...,kn =0 j j !
k
n times
X∞ Z t Z tn Z t2
= dtn dtn−1 · · · dt1
n=0 0 0 0
∞
X kn kn−1 k0
(t − tn ) (tn − tn−1 ) (t1 − 0)
× ···
kn ! kn−1 ! k0 !
k0 ,...,kn =0
k
×Lk0n SL0n−1 S · · · SLk01 SLk00
∞ Z t
X Z tn Z t2
= eL0 t + dtn dtn−1 · · · dt1
n=1 0 0 0
L0 (t−tn ) L0 (tn −tn−1 )
×e Se S · · · eL0 (t2 −t1 ) SeL0 t1 . (83)
P
The step from the second to the third line, where tn+ j kj /(n + j kj )! is
P

replaced by n time integrals can be checked by induction.

The form (83) is a generalized Dyson expansion, and the comparison with
the Dyson series for unitary evolutions suggests to view exp (L0 τ ) as an “un-
perturbed” evolution and S as a “perturbation”, such that the exact time
evolution Wt is obtained by an integration over all iterations of the perturba-
tion, separated by the unperturbed evolutions.
The particular Lindblad form (79) of the generator suggests to introduce
the completely positive jump superoperators

Lk ρ = γk Lk ρL†k , (84)

along with the non-hermitian operator

N
i~ X
HC = H − γk L†k Lk . (85)
2
k=1

The latter has a negative imaginary part, Im (HC ) < 0, and can be used to
construct
1  
L0 ρ = HC ρ − ρHC†
. (86)
i~
It follows that the sum of these superoperators yields the Liouville operator
(79)
N
X
L = L0 + Lk . (87)
k=1
26 Klaus Hornberger
PN
Of course, neither L0 nor S = k=1 Lk generate a dynamical semigroup.
Nonetheless, they are useful since the interpretation of Eq. (83) can now be
taken one step further. We can take the point of view that the Lk with k > 1
describe elementary transformation events due to the environment (“jumps”),
which occur at random times with a rate γk . A particular realization of n such
events is specified by a sequence of the form

Rnt = (t1 , k1 ; t2 , k2 ; . . . .; tn , kn ). (88)

The attributed times satisfy 0 < t1 < . . . . < tn < t, and the kj ∈ {1, . . . , N }
indicate which kind of event “took place”. We call Rnt a record of length n.
The general time evolution Wt can thus be written as an integration over
all possible realizations of the jumps, with the “free” evolution exp (L0 τ ) in
between.
X∞ Z t Z tn Z t2
Wt = eL0 t + dtn dtn−1 · · · dt1 (89)
n=1 0 0 0
X
L0 (t−tn ) L0 (tn −tn−1 )
× e Lkn e Lkn−1 · · · eL0 (t2 −t1 ) Lk1 eL0 t1
{Rn }
| {z }
K Rt
n

As a result of the negative imaginary part in (85) the exp (L0 τ ) are trace
decreasing 9 completely positive maps,
   
iτ iτ †
eL0 τ ρ = exp − HC ρ exp HC > 0 (90)
~ ~
X N
d


tr eL0 τ ρ = tr L0 eL0 τ ρ = − tr(Lk eL0 τ ρ) < 0 (91)
dτ | {z }
k=1 >0


It is now natural to interpret tr eL0 t ρ as the probability that no jump occurs
during the time interval t,



Prob R0t |ρ := tr eL0 t ρ . (92)

To see that this makes sense, we attribute to each record of length n a n-time
probability density. For a given record Rnt we define



prob Rnt |ρ := tr KRtn ρ , (93)

in terms of the superoperators from the second line in (89),

KRtn := eL0 (t−tn ) Lkn eL0 (tn −tn−1 ) Lkn−1 · · · eL0 (t2 −t1 ) Lk1 eL0 t1 . (94)

This is reasonable since the KRtn are completely positive maps that do not
preserve the trace. Indeed, the probability density for a record is thus de-
termined both by the corresponding jump operators, which involve the rates
9
See the note in Sect. 3.1(b).
 Introduction to decoherence theory 27

γk , and by the eL0 τ ρ, which account for the fact that the likelihood for the
absence of a jump decreases with the length of the time interval.
This notion of probabilities is consistent, as can be seen by adding the
probability (92) for no jump to occur during the interval (0; t) to the integral
over the probability densities (93) of all possible jump sequences. As required,
the result is unity,
∞ Z Z Z

X t tn t2 X

Prob R0t |ρ + dtn dtn−1 · · · dt1 prob Rnt |ρ = 1
n=1 0 0 0 {Rtn }

for all ρ and t > 0. This follows immediately from the trace preservation of
the map (89).
It is now natural to normalize the transformation defined by the KRtn .
Formally, this yields the state transformation conditioned to a certain record
Rnt . It is called a quantum trajectory,

KRtn ρ
T ρ|Rnt :=
. (95)
tr KRtn ρ

Note that this definition comprises the trajectory corresponding to a null-

record R0t , where KRt0 = exp (L0 t). These completely positive, trace-preserving,
nonlinear maps ρ 7→ T (ρ|Rnt ) are defined for all states ρ that yield a finite
probability (density) for the given record Rnt , i.e., if the denominator in (95)
does not vanish.
Using these notions the exact solution of a general Lindblad master equa-
tion (83) may thus be rewritten in the form



ρt = Prob R0t T ρ|R0t
∞ Z t
X Z tn Z t2 X


+ dtn dtn−1 · · · dt1 prob Rnt |ρ T ρ|Rnt .(96)
n=1 0 0 0 {Rn }

It shows that the general Markovian quantum dynamics can be understood

as a summation over all quantum trajectories T (ρ|Rnt ) weighted by their
probability (density). This is called a stochastic unravelling of the master
equation. The set of trajectories and their weights are labeled by the possible
records (88), and determined by the Lindblad operators of the master equation
(79).
The semigroup property described by the master equation shows up if a
record is formed by joining the records of adjoining time intervals, (0; t′ ) and
(t′ ; t),

(0;t) (0;t′ ) (t′ ;t)

Rn+m := (Rn ; Rm ). (97)

As one expects, the probabilities and trajectories satisfy

28 Klaus Hornberger

(0;t) (0;t′ ) (t′ ;t) (0;t′ )

prob(Rn+m |ρ) = prob(Rn |ρ) prob(Rm |T (ρ|Rn )) (98)

and
(0;t) (0;t′ ) (t′ ;t)
T (ρ|Rn+m ) = T (T (ρ|Rn )|Rm ). (99)

Note finally, that the concept of quantum trajectories fits seamlessly into
the framework of generalized measurements discussed in Sect. 1.3(a). In par-
ticular, the conditioned state transformation T (ρ|Rnt ) has the form (18) of an
efficient measurement transformation,

MRtn ρM†Rt
T ρ|Rnt =  n
 (100)
tr M†Rt MRtn ρ
n

with compound measurement operators

MRtn := e−iHC (t−tn )/~ Lkn · · · Lk2 e−iHC (t2 −t1 )/~ Lk1 e−iHC t1 /~ . (101)

This shows that we can legitimately view the open quantum dynamics gen-
erated by L as due to the continuous monitoring of the system by the en-
vironment. We just have to identify the (aptly named) record Rnt with the
total outcome of a hypothetical, continuous measurement during the interval
(0; t). The jump operators Lk then describe the effects of the corresponding
elementary measurement events10 (“clicks of counter k”). Since the absence
of any click during the “waiting time” τ may also confer information about
the system, this lack of an event constitutes a measurement, as well, which
is described by the non-unitary operators exp (−iHC τ /~). A hypothetical de-
mon, who has the full record Rnt available, would then be able to predict the
final state T (ρ|Rnt ). In the absence of this information we have to resort to
the probabilistic description (96) weighting each quantum trajectory with its
(Bayesian) probability.
We can thus conclude that the dynamics of open quantum dynamics, and
therefore decoherence, can in principle be understood in terms of an infor-
mation transfer to the environment. Apart from this conceptual insight, the
unravelling of a maser equation provides also an efficient stochastic simula-
tion method for its numerical integration. In these quantum jump approaches
[24, 25, 26], which are based on the observation that the quantum trajectory
(95) of a pure state remains pure, one generates a finite ensemble of trajec-
tories such that the ensemble mean approximates the solution of the master
equation.

10
Keep in mind that the Lk are not uniquely specified by a given generator L, see
Eqs. (80)-(82). Different choices of the Lindblad operators lead to different un-
ravellings of the master equation, so that these hypothetical measurement events
must not be viewed as “real”.
 Introduction to decoherence theory 29

3.4 Exemplary master equations

Let us take a look at a number of very simple Markovian master equations11 ,

which are characterized by a single Lindblad operator L (together with a her-
mitian operator H). The first example gives a general description of dephasing,
while the others are empirically known to describe dissipative phenomena re-
alistically. We may then ask what they predict about decoherence.

(a) Dephasing

The simplest choice is to take the Lindblad operator to be proportional12 to

the Hamiltonian of a discrete quantum system, i.e., to the generator of the
√
unitary dynamics, L = γH. The Lindblad equation
 
1 1 1
∂t ρt = [H, ρt ] + γ Hρt H − H2 ρt − ρt H2 (102)
i~ 2 2
P
is immediately solved in the energy eigenbasis, H = m Em |mihm| :
 
Em − En γ 2
ρmn (t) ≡ hm|ρt |ni = ρmn (0) exp −i t − (Em − En ) t (103)
~ 2
As we expect from the discussion of qubit dephasing in Sect. 2, the energy
eigenstates are unaffected by the non-unitary dynamics if the environmental
effect commutes with H. The coherences show the exponential decay that
we found in the “thermal regime” (of times t which are long compared to
the inverse Matsubara frequency). The comparison with (54) indicates that γ
should be proportional to the temperature of the environment.

(b) Amplitude damping of the harmonic oscillator

Next we choose H to be the Hamiltonian of a harmonic oscillator, H = ~ωa† a

and take as Lindblad operator the ladder operator, L = a. The resulting
Lindblad equation is known empirically to describe the quantum dynamics of
a damped harmonic oscillator.
Choosing as initial state a coherent state, see Sect. 2.1 and Eq. (114),

ρ0 = |α0 ihα0 | ≡ D(α0 )|0ih0|D† (α0 )

2
= e−|α0 | e xp(α0 a† )|0ih0| exp (α∗0 a) (104)

we are faced with the exceptional fact that the state remains pure during the
Lindblad time evolution. Indeed, the solution of the Lindblad equation reads
11
See also Sect. ?.1.3 in Cord Müller’s contribution for a discussion of the master
equation describing of spin relaxation.
12
As an exception, γ does not have the dimensions of a rate here (to avoid clumsy
notation).
30 Klaus Hornberger

ρt = |αt ihαt | (105)

with  γ 
αt = α0 exp −iωt − t (106)
2
as can be verified easily using (104). It describes how the coherent state spirals
in phase space towards the origin, approaching the ground state as t → ∞.
The rate γ is the dissipation rate since it quantifies the energy loss, as shown
by the time dependence of the energy expectation value,

hαt |H|αt i = e−γt hα0 |H|α0 i. (107)

Superposition of coherent states

What happens if we start out with a superposition of coherent states,

1
|ψ0 i = √ (|α0 i + |β0 i) (108)
N
with N = 2 + 2 Rehα0 |β0 i, in particular, if the separation in phase space is
large compared to the quantum uncertainties, |α0 − β0 | ≫ 1 ? The initial
density operator corresponding to (108) reads
1
ρ0 = (|α0 ihα0 | + |β0 ihβ0 | + c0 |α0 ihβ0 | + c∗0 |β0 ihα0 |) (109)
N
with c0 = 1. One finds that the ansatz
1
ρt = (|αt ihαt | + |βt ihβt | + ct |αt ihβt | + c∗t |βt ihαt |) (110)
N
solves the Lindblad equation with (106) provided
  
1 2

ct = c0 exp − |α0 − β0 | + i Im(α0 β0∗ ) 1 − e−γt . (111)
2
That is, while the coherent “basis” states have the same time evolution as in
(105), the initial coherence c0 gets additionally suppressed. For times that are
short compared to the dissipative time scale, t ≪ γ −1 , we have an exponential
decay  
 γ 2 
|ct | = |c0 | exp  
− 2 |α0 − β0 | t (112)
| {z }
γdeco

with a rate γdeco. For macroscopically distinct superpositions, where the phase
space distance of the quantum states is much larger than their uncertainties,
|α0 − β0 | ≫ 1, γdeco can be much greater than the dissipation rate,
γdeco 1 2
= |α0 − β0 | ≫ 1. (113)
γ 2
 Introduction to decoherence theory 31

This quadratic increase of the decoherence rate with the separation between
the coherent states has been confirmed experimentally in a series of beautiful
cavity QED experiments in Paris, using field states with an average of 5 – 9
photons [27, 28].
Given this empiric support we can ask about the prediction for a material,
macroscopic oscillator. As an example, we take a pendulum with a mass of
m =100 g and a period of 2π/ω = 1s, and assume that we can prepare it in
a superposition of coherent states with a separation of x =1 cm. The mode
variable α is related to position and momentum by
r
mω  p 
α= x+i (114)
2~ mω
so that we get the prediction

γdeco ≃ 1030 γ.

This purports that even with an oscillator of enormously low friction corre-
sponding to a dissipation rate of γ =1/year the coherence is lost on a timescale
of 10−22 sec—in which light travels the distance of about a nuclear diameter.
This observation is often evoked to explain the absence of macroscopic su-
perpositions. However, it seems unreasonable to assume that anything physi-
cally relevant takes place on a timescale at which a signal travels at most by
the diameter of an atomic nucleus. Rather, one expects that the decoherence
rate should saturate at a finite value if one increases the phase space distance
between the superposed states.

(c) Quantum Brownian motion

Next, let us consider a particle in one dimension. A possible choice for

the Lindblad operator is a linear combination of its position and momentum
operators,
pth i
L= x+ p. (115)
~ pth
Here pth is just a momentum scale, which will be related to the temperature
of the environment below. The hermitian operator is taken to be the Hamil-
tonian of a particle in a potential V (x), plus a term due to the environmental
coupling,
p2 γ
H= + V (x) + (xp + px). (116)
2m 2
This additional term will be justified by the fact that the resulting Lindblad
equation is almost equal to the Caldeira-Leggett master equation. The latter
is the high-temperature limit of the exact evolution equation following from a
harmonic bath model of the environment [29, 30], see Sect. 4.1. It is empirically
known to describe the frictional quantum dynamics of a Brownian particle,
32 Klaus Hornberger

and, in particular, it leads to the correct thermal state in case of quadratic

potentials.
The choices (115) and (116) yield the following Lindblad equation:
Caldeira − Leggett master equation
z }| {
1 p2 γ γ p2th
∂t ρt = [ + V (x), ρt ] + [x, pρt + ρt p]− 2
[x, [x, ρt ]]
i~ 2m |i~ {z }| 2 ~ {z }
dissipation position localization
γ 1
− 2 [p, [p, ρt ]] (117)
2 pth

The three terms in the upper line (with pth from Eq. (124)) constitute the
Caldeira-Leggett master equation. It is a Markovian, but not a completely
positive master equation. In a sense, the last term in (117) adds the mini-
mal modification required to bring the Caldeira-Leggett master equation into
Lindblad form [2].
To see the most important properties of (117) let us take a look at the time
evolution of the relevant observables in the Heisenberg picture. As discussed
in Sect. 3.1(d), the Heisenberg equations of motion are determined by the
dual Liouville operator L♯ . In the present case, it takes the form

1 p2 γ γ p2th
L♯ (A) = [A, + V (x)] − (p [x, A] + [x, A] p) − [x, [x, A]]
i~ 2m i~ 2 ~2
γ 1
− 2 [p, [p, A]] . (118)
2 pth

It is now easy to see that

p
L♯ (x) =
m
L♯ (p) = −V ′ (x) − 2γp. (119)

Hence, the force arising from the potential is complemented by a frictional

force which will drive the particle into thermal equilibrium. The fact that this
frictional component stems from the second term in (117) indicates that the
latter describes the dissipative effect of the environment.
In the absence of an external potential,
V = 0, the time evolution deter-
n
mined by (119) is easily obtained, since L♯ (p) = (−2γ)n p for n ∈ N.

X∞ n
♯ (−2γt)
pt = e L t p = p = e−2γt p [for V = 0] (120)
n=0
n!
∞
♯ 1 X tn ♯
n−1
xt = eL t x = x + L (p)
m n=1 n!
p − pt
= x+ [for V = 0] (121)
2γm
 Introduction to decoherence theory 33

Note that, unlike in closed systems, the Heisenberg operators do not retain

their commutator, [xt , pt ] 6= i~ for t > 0 (since the map Wt♯ = exp L♯ t is
non-unitary). Similarly, (p2 )t 6= (pt )2 for t > 0, so that the kinetic energy
operator T = p2 /2m has to be calculated separately. Noting

p2th
L♯ (T) = γ − 4γT [for V = 0] (122)
2m
we find similarly
 
p2th p2
Tt = + T − th e−4γt [for V = 0]. (123)
8m 8m
This shows how the kinetic energy approaches a constant determined by the
momentum scale pth . We can now relate pth to a temperature by equating the
stationary expectation value tr (ρT∞ ) = p2th /8m with the the average kinetic
energy in a one-dimensional thermal distribution. This leads to
p
pth = 2 mkB T . (124)

Not always is one able to state the operator evolution in closed form. In
those cases it may be helpful to take a look at the Ehrenfest equations for
their expectation values. For example, given hp2 it = 2mhTit , the other second
moments, hx2 it and hpx + xpit form a closed set of differential equations.
Their solutions, given in [2], yield the time evolution of the position variance
σx2 (t) = hx2 it − hxi2t . It has the asymptotic form
kB T
σx2 (t) ∼ t as t → ∞, (125)
mγ
which shows the diffusive behavior expected of a (classical) Brownian parti-
cle13 .
Let us finally take a closer look at the physical meaning of the third term in
(117), which is dominant if the state is in a superposition of spatially separated
states. Back in the Schrödinger picture we have in position representation,
ρt (x, x′ ) = hx|ρt |x′ i,

γ p2th
∂t ρt (x, x′ ) = − 2
(x − x′ )2 ρt (x, x′ ) + [the other terms]. (126)
2
| ~ {z }
γdeco

The “diagonal elements” ρ (x, x) are unaffected by this term, so that it leaves
the particle density invariant. The coherences in position representation, on
the other hand, get exponentially suppressed,

ρt (x, x′ ) = exp (−γdecot) ρ0 (x, x′ ). (127)

13
Note that the definition of γ differs by a factor of 2 in part of the literature.
34 Klaus Hornberger

Again the decoherence rate is determined by the square of the relevant dis-
tance |x − x′ |,
γdeco (x − x′ )2
= 4π . (128)
γ Λ2th
Like in Sect. 3.4(b), the rate γdeco will be much larger than the dissipative
rate, provided the distance is large on the quantum scale, here given by the
thermal de Broglie wavelength,

2π~2
Λ2th = . (129)
mkB T
In particular, one finds γdeco ≫ γ if the separation is truly macroscopic.
Again, it seems unphysical that the decoherence rate does not saturate as
|x − x′ | → ∞, but grows above all bounds. One might conclude from this
that non-Markovian master equation are more appropriate on these short
timescales. However, I will argue that (unless the environment has very special
properties) Markovian master equations are well suited to study decoherence
processes, provided they involve an appropriate description of the microscopic
dynamics.

4 Microscopic derivations
In this section we discuss two important and rather different strategies to
obtain Markovian master equations based on microscopic considerations.

4.1 The weak coupling formulation

The most widely used form of incorporating the environment is the weak
coupling approach. Here one assumes that the total Hamiltonian is “known”
microscopically, usually in terms of a simplified model,

Htot = H + HE + Hint

and takes the interaction part Hint to be “weak” so that a perturbative treat-
ment of the interaction is permissible.
The main assumption, called the Born approximation, states that Hint
is sufficiently small so that we can take the total state as factorized, both
initially, ρtot (0) = ρ(0) ⊗ ρE , and also at t > 0 in those terms which involve
Hint to second order.

Assumption 1 : ρtot (t) ≃ ρ(t) ⊗ ρE [to 2nd order in Hint ] (130)

Here ρE is the stationary state of the environment, [HE , ρE ] = 0. Like above,

the use of the interaction picture is indicated with a tilde, cf. Eq. (28), so that
the von Neumann equation for the total system reads
 Introduction to decoherence theory 35

1 e
∂t ρ̃tot = [Hint (t), ρ̃tot (t)]
i~
Z t
1 e 1 e int (t), [H
e int (s), ρ̃tot (s)]].(131)
= [H int (t), ρ̃ tot (0)] + ds [H
i~ (i~)2 0
In the second equation, which is still exact, the von Neumann equation in its
integral equation version was inserted into the differential equation version.
Using a basis of Hilbert-Schmidt operators of the product Hilbert space, see
Sect. 3.2, one can decompose the general H e int into the form
X
e int (t) =
H Ae k (t) ⊗ B e k (t) (132)
k

with Aek = Ae† , B e † . The first approximation is now to replace ρ̃tot (s) by
ek = B
k k
ρ̃(s) ⊗ ρE in the double commutator of (131), where H e int appears to second
order. Performing the trace over the environment one gets

∂t ρ̃(t) = trE (∂t ρ̃tot )

∼ 1 X e e k (t), ρ̃(0)]
= hBk (t)iρE [A
i~
k

Z
1 X t e k (t)B e k (t)A
e ℓ (s)iρ {A e ℓ (s)ρ̃(s) − A
e ℓ (s)ρ̃(s)A
e k (t)}
+ ds hB E
(i~)2 
 0 | {z }
kℓ
Ckℓ (t−s)
)
+ h.c. (133)

All the relevant properties of the environment are now expressed in terms of
the (complex) bath correlation functions Ckℓ (t − s). They depend only on the
time difference t − s, since [HE , ρE ] = 0,



Ckℓ (τ ) = tr eiHE τ Bk e−iHE τ Bℓ ρE ≡ eiHE τ Bk e−iHE τ Bℓ ρE . (134)

This function is determined by the environmental state alone, and it is typi-

cally appreciable only for a small range of τ around τ = 0.
Equation (133) has the closed form of a generalized master equation, but
it is non-local in time, i.e., non-Markovian. Viewing the second term as a
superoperator K, which depends essentially on t − s we have
Z t
1 e
∂t ρ̃(t) = [hH I (t)iρE , ρ̃(0)] + dsK(t − s)ρ̃(s) , (135)
|i~ {z } 0
disregarded

where K is a superoperator memory kernel of the form (64). We may disregard

the first term since the model Hamiltonian HE can always be reformulated
such that hBe k (t)iρE = 0.
36 Klaus Hornberger

A naive application of second order of perturbation theory would now

replace ρ̃(s) by the initial ρ̃(0). However, since the memory kernel is dominant
at the origin it is much more reasonable to replace ρ̃(s) by ρ̃(t). The resulting
master equation is local in time,
Z t 
∼
∂t ρ̃(t) = 0 + dsK(t − s) ρ̃(t).
0

It is called the Redfield equation and it is not Markovian, because the inte-
grated superoperator still depends on time. Since the kernel is appreciable
only at the origin it is reasonable to replace t in the upper integration limit
by ∞.
These steps are summarized by the Born-Markov approximation:
Z t Z ∞
Assumption 2 : ∼
dsK(t − s)ρ̃ (s) = dsK(s)ρ̃ (t) (136)
0 0

It leads from (135) to a Markovian master equation provided hHI iρE = 0.

However, by no means is such a master equation guaranteed to be com-
pletely positive. An example is the Caldeira-Leggett master equation discussed
in Sect. 3.4(c). It can be derived by taking the environment to be a bath of
bosonic field modes whose field amplitude is coupled linearly to the particle’s
position operator. A model assumption on the spectral density of the coupling
then leads to the frictional behavior of Eq. (119) [29, 17].
A completely positive master equation can be obtained by a further sim-
plification, the “secular ” approximation, which is applicable if the system
Hamiltonian H has a discrete, non-degenerate spectrum. The system opera-
tors Ak can then be decomposed in the system energy eigenbasis. Combining
the contributions with equal energy differences
X
Ak (ω) = hE|Ak |E ′ i|EihE ′ | = A†k (ω) (137)
E ′ −E=~ω

we have
X
Ak = Ak (ω). (138)
ω

The time dependence of the operators in the interaction picture is now par-
ticularly simple, X
e k (t) =
A e−iωt Ak (ω) . (139)
ω
Inserting this decomposition we find
XX ′
∂t ρ̃(t) = ei(ω−ω )t Γkℓ (ω ′ ){Aℓ (ω ′ )ρ̃(t)A†k (ω) − A†k (ω)Aℓ (ω ′ )ρ̃(t)}
kℓ ωω ′

+ h.c. (140)
 Introduction to decoherence theory 37

with Z ∞
1 e k (s)Bℓ (0)iρ .
Γkℓ (ω) = 2 dseiωs hB E (141)
~ 0
For times t which are large compared to the time scale given by the smallest
system energy spacings it is reasonable to expect that only equal pairs of
frequencies ω, ω ′ contribute appreciably to the sum in (140), since all other
contributions are averaged out by the wildly oscillating phase factor. This
constitutes the rotating wave approximation, our third assumption
X ′ X
Assumption 3 : ei(ω−ω )t f (ω, ω ′ ) ≃ f (ω, ω). (142)
ωω ′ ω

It is now useful to rewrite

1
Γkℓ (ω) = γkℓ (ω) + iSkℓ (ω) (143)
2
with γkℓ (ω) given by the full Fourier transform of the bath correlation func-
tion,
Z ∞ D E
∗ 1 e k (t)Bℓ (0)
γkℓ (ω) = Γkℓ (ω) + Γℓk (ω) = 2 dteiωt B , (144)
~ −∞ ρE

and the hermitian matrix Skℓ (ω) defined by

1 ∗
Skℓ (ω) = (Γkℓ (ω) − Γℓk (ω)). (145)
2i
The matrix γkℓ (ω) is positive 14 so that we end up with a master equation of
the first Lindblad form (78),
1 X 
∂t ρ̃(t) = [HLamb , ρ̃(t)] + γkℓ (ω) Aℓ (ω)ρ̃(t)A†k (ω)
i~
kℓω

1 1
− A†k (ω)Aℓ (ω)ρ̃(t) − ρ̃(t)A†k (ω)Aℓ (ω) .(147)
2 2
14
To see that the matrix γ (ω) ≡ (γkℓ (ω))k,ℓ is positive we write
X ∗ Z X D iH t/~ E
1
(v, γv) = vk γkℓ (ω)vℓ = 2 dteiωt e E Bk (0)vk∗ e−iHE t/~ Bℓ (0)vℓ
~ ρE
kℓ kℓ
Z D E
= dteiωt eiHE t/~ C† e−iHE t/~ C (146)
ρE

−1
P
with C := ~ ℓ vℓ Bℓ (0). One can now check that due to its particular form the
correlation function
D E
f (t) = eiHE t/~ C† e−iHE t/~ C
ρE

appearing in (146) is of positive type, meaning that the n × n-matrices

N
(f (ti − tj ))ij defined by an arbitrary choice of t1 , . . . ., tn and n ∈ are pos-
itive. According to Bochner’s theorem [31] the Fourier transform of a function
which is of positive type is positive, which proves the positivity of Eq. (146).
38 Klaus Hornberger

The hermitian operator

X
HLamb = ~ Skℓ (ω)A†k (ω)Aℓ (ω) (148)
kℓω

describes a renormalization of the system energies due to the coupling with

the environment, the Lamb shift . Indeed, one finds [H, HLamb ] = 0.
Reviewing the three approximations (130),(136),(142) in view of the de-
coherence problem one comes to the conclusion that they all seem to be well
justified if the environment is generic and the coupling is sufficiently weak.
Hence, the master equation should be alright for times beyond the short-time
transient which is introduced due to the choice of a product state as initial
state. Evidently, the problem of non-saturating decoherence rates encountered
in Sects. 3.4(b) and 3.4(c) is rather due to the linear coupling assumption,
corresponding to a “dipole approximation”, which is clearly invalid once the
system states are separated by a larger distance than the wavelength of the
environmental field modes.
This shows the need to incorporate realistic, nonlinear environmental cou-
plings with a finite range. A convenient way of deriving such master equations
is discussed in the next section.

4.2 The monitoring approach

The following method to derive microscopic master equations differs consid-

erably from the weak coupling treatment discussed above. It is not based on
postulating an approximate “total” Hamiltonian of system plus environment,
but on two operators, which can be characterized individually in an opera-
tional sense. This permits to describe the environmental coupling in a non-
perturbative fashion, and to incorporate the Markov assumption right from
the beginning, rather than introducing it in the course of the calculation.
The approach is motivated by the observation made in Sect. 1.3 and
Sect. 3.3 that environmental decoherence can be understood as due to the
information transfer from the system to the environment occurring in a se-
quence of indirect measurements. In accordance with this, we will picture the
environment as monitoring the system continuously by sending probe particles
which scatter off the system at random times. This notion will be applicable
whenever the interaction with the environment can reasonably be described in
terms of individual interaction events or “collisions”, and it suggests a formu-
lation in terms of scattering theory, like in Sect. 1.2. The Markov assumption
is then easily incorporated by disregarding the change of the environmental
state after each collision.
When setting up a differential equation, one would like to write the tem-
poral change of the system as the rate of collisions multiplied by the state
transformation due to an individual scattering. However, in general not only
the transformed state will depend on the original system state, but also the
 Introduction to decoherence theory 39

Fig. 1. (a) In the monitoring approach the system is taken to interact at most with
one environmental (quasi-)particle at a time, so that three-body collisions are ex-
cluded. Moreover, in agreement with the Markov assumption, it is assumed that the
environmental particles disperse their correlation with the system before scattering
again. (b) In order to consistently incorporate the state-dependence of the collision
rate into the dynamic description of the scattering process, we imagine that the
system is monitored continuously by a transit detector, which tells at a temporal
resolution ∆t whether a particle is going to scatter off the system, or not.

collision rate, so that such a naive ansatz would yield a nonlinear equation.
To account for this state dependence of the collision rate in a proper way
we will apply the concept of generalized measurements discussed in Sect. 1.3.
Specifically, we shall assume that the system is surrounded by a hypotheti-
cal, minimally invasive detector, which tells at any instant whether a probe
particle has passed by and is going to scatter off the system, see Fig. 1.
The rate of collisions is then described by a positive operator Γ acting in
the the system-probe Hilbert space. Given the uncorrelated state ̺tot = ρ⊗ρE
it determines the probability of a collision to occur in a small time interval
∆t,
Prob (C∆t |ρ ⊗ ρE ) = ∆t tr (Γ [ρ ⊗ ρE ]) . (149)
Here, ρE is the stationary reduced single particle state of the environment.
The microscopic definition of Γ will in general involve the current density
operator of the relative motion and a total scattering cross section, see below.
The important point to note is that the information that a collision is going
to take place changes our knowledge about the state, that is, the description
of the state according to the generalized measurement transformation (16).
At the same time, we have to keep in mind that the detector is not real, but
is introduced here only for enabling us to account for the state dependence
of the collision probability. It is therefore reasonable to take the detection
process as efficient , see Sect. 1.3(a), and minimally-invasive, i.e., Uα = I in
Eq. (20), so that neither unnecessary uncertainty nor a reversible back-action
40 Klaus Hornberger

is introduced. This implies that after a (hypothetical) detector click, but prior
to scattering, the system-probe state will have the form

Γ1/2 ̺tot Γ1/2

M (̺tot |C∆t ) = . (150)
tr (Γ̺tot )

This measurement transformation reflects our improved knowledge about the

incoming two-particle wave packet, and it may be viewed as enhancing those
parts which are heading towards a collision. Also the absence of a detec-
tion event
during ∆t changes the state. This occurs with the probability
Prob C∆t = 1 − Prob (C∆t ), and the state conditioned on a null-event is
given by15

̺tot − ∆tΓ1/2 ̺tot Γ1/2

M ̺tot |C∆t = . (151)
1 − ∆t tr (Γ̺tot )

We can now form the unconditioned system-probe state after time ∆t by tak-
ing into account that the detection outcomes are not really available. The
infinitesimally evolved state is then given by the mixture of the colliding
state transformed by the S-matrix and the untransformed non-colliding one,
weighted with their respective probabilities,



̺′tot (∆t) = Prob (C∆t |̺tot ) SM (̺tot |C∆t ) S† + Prob C∆t |̺tot M ̺tot |C∆t
= SΓ1/2 ̺tot Γ1/2 S† ∆t + ̺tot − Γ1/2 ̺tot Γ1/2 ∆t. (152)

Splitting off the nontrivial part of the two-particle S-matrix S = I + iT one

finds that the unitarity of S implies that


i T − T† = −T† T. (153)

Using this relation we can write the differential quotient as

̺′tot (∆t) − ̺tot 1

= TΓ1/2 ̺tot Γ1/2 T† − T† TΓ1/2 ̺tot Γ1/2
∆t 2
1 ih i
− Γ1/2 ̺tot Γ1/2 T† T + T + T† , Γ1/2 ̺tot Γ1/2 . (154)
2 2
It is now easy to arrive at a closed differential equation. We trace out the
environment, assuming, in accordance with the Markov approximation, that
the factorization ̺tot = ρ ⊗ ρE is valid prior to each monitoring interval ∆t.
Taking the limit of continuous monitoring ∆t → 0 and adding the generator
H of the free system evolution we arrive at [32]

15
Although more general transformations are conceivable, this one is distinguished
by the fact that it introduces no further (time-dependent) operators.
 Introduction to decoherence theory 41

d 1 i h i
ρ= [H, ρ] + TrE T + T† , Γ1/2 [ρ ⊗ ρE ] Γ1/2
dt i~  2 
+ TrE TΓ1/2 [ρ ⊗ ρE ] Γ1/2 T†
1  
− TrE Γ1/2 T† TΓ1/2 [ρ ⊗ ρE ]
2
1  
− TrE [ρ ⊗ ρE ] Γ1/2 T† TΓ1/2 . (155)
2
This general evolution equation generates a dynamical semigroup since the
transformation (152) is completely positive. It incorporates the state depen-
dence of the collision rate in terms of the operators Γ1/2 , while the operators
T describe the individual microscopic interaction process without approxima-
tion. We will see that the second term in (155), which involves a commutator,
accounts for the renormalization of the system energies due to the coupling
to the environment, just like (148).
So far, the discussion was very general. To obtain concrete master equa-
tions one has to specify system and environment, along with the operators
Γ and S describing their interaction. In the following applications, we will
assume the environment to be an ideal Maxwell gas, whose single particle
state
 
Λ3th p2
ρgas = exp −β (156)
Ω 2m
is characterized by the inverse temperature β, the normalization volume Ω,
and the thermal de Broglie wave length Λth defined in (129).

4.3 Collisional decoherence of a Brownian particle

As a first application of the monitoring approach, let us consider the “local-

ization” of a mesoscopic particle by a gaseous environment. Specifically, we
will assume that the mass M of the Brownian particle is much greater than
the mass m of the gas particles. In this limit the energy exchange during an
elastic collision vanishes, so that the mesoscopic particle will not thermalize,
but we expect that the off-diagonal of the position state will get reduced by
decoherence, as discussed in Sect. 3.4(c).
This can be seen by taking the S-matrix in the limit m/M → 0. In general,
a collision keeps the center-of-mass invariant, and only the relative coordinates
are affected. Writing S0 for the S-matrix in the center of mass frame and
denoting the momentum eigenstates of the Brownian and the gas particle by
|P i and |pi, respectively, we have [33]
Z
m∗ m∗ m∗ m∗
S|P i|pi = d3 Q|P − Qi|p + Qih p− P + Q|S0 | p− Pi,
m M m M
where m∗ = M m/ (M + m) is the reduced mass. In the limit of a large Brow-
nian mass we have m∗ /m → 1 and m∗ /M → 0, so that
42 Klaus Hornberger
Z
S|P i|pi → d3 Q|P − Qi|p + Qihp + Q|S0 |pi [for M ≫ m]. (157)

It follows that a position eigenstate |Xi of the Brownian particle remains

unaffected by a collision,
 
S|Xi|ψin iE = |Xi e−ip·X/~ S0 eip·X/~ |ψin iE , (158)
| {z }
(X)
|ψout iE

as can be seen by inserting identities in terms of the momentum eigenstates.

Here, |ψin iE denotes an arbitrary single-particle wave packet state of a gas
atom. The exponentials in (158) effect a translation of S0 from the origin to
(X)
the position X, so that the scattered state of the gas particle |ψout iE depends
on the location of the Brownian particle.
Just like in
Sect. 1.1, a single collision will thus reduce the spatial coher-
ences ρ X, X ′ = hX|ρ|X ′ i by the overlap of the gas states scattered at
positions X and X ′ ,


(X ′ ) (X)
ρ′ X, X ′ = ρ X, X ′ hψout |ψout iE . (159)

The reduction factor will be the smaller in magnitude the better the scattered
state of the gas particle can “resolve” between the positions X and X ′ .
In order to obtain the dynamic equation we need to specify the rate oper-
ator. Classically, the collision rate is determined by the product of the current
density j = ngas vrel and the total cross section σ (prel ), and therefore Γ should
be expressed in terms of the corresponding operators. This is particularly sim-
ple in the large mass limit M → ∞, where vrel = |p/m − P /M | → |p| /m, so
that the current density and the cross section depend only on the momentum
of the gas particle, leading to

|p|
Γ = ngas σ (p) . (160)
m
If the gas particle moves in a normalized wave packet heading towards the
origin then the expectation value of this operator will indeed determine the
collision probability. However, this expression depends only on the modulus of
the velocity so that it will yield a finite collision probability even if the particle
is heading away form the origin. Hence, for (155) to make sense either the S-
matrix should keep such a non-colliding state unaffected, or Γ should contain
in addition a projection to the subset of incoming states, see the discussion
below.

In momentum representation, ρ P , P ′ = hP |ρ|P ′ i, equation (155) as-
sumes the general structure16

16
The second term in (155) vanishes in the limit M ≫ m.
 Introduction to decoherence theory 43
2

1 P 2 − (P ′ )

∂t ρ P , P ′ = ρ P,P′
i~Z 2M



+ dP 0 dP ′0 ρ P 0 , P ′0 M P , P ′ ; P 0 , P ′0
Z Z
1

− dP 0 ρ P 0 , P ′ dP f M (P f , P f ; P 0 , P )
2
Z Z
1


− dP ′0 ρ P , P ′0 dP f M P f , P f ; P ′ , P ′0 . (161)
2
The dynamics is therefore characterized by a single complex function

   
M P , P ′ ; P 0 , P ′0 = hP | trgas TΓ1/2 |P 0 ihP ′0 | ⊗ ρgas Γ1/2 T† |P ′ i,

which has to be evaluated. Inserting the diagonal representation of the gas

state (156)
Z
ρgas = dp0 µ (p0 ) |p0 ihp0 | (162)

it reads, with the choices (157) and (160) for S and Γ,

Z
′ ′



M P , P ; P − Q, P − Q = dp1 dp0 µ (p0 ) δ (Q + p1 − p0 ) δ Q′ + p0 − p1

ngas (2π~)3
× |p0 | σ (p0 ) |hp1 |T0 |p0 i|2
m Z Ω

ngas
= δ Q − Q′ dp0 µ (p0 ) |p0 | σ (p0 )
m
3
(2π~)
× |hp0 − Q|T0 |p0 i|2
Ω

=: δ Q − Q′ Min (Q) . (163)

This shows that, apart form the unitary motion, the dynamics is simply char-
acterized by momentum exchanges described in terms of gain and loss terms,
2 Z
′

1 P 2 − (P ′ )


∂t ρ P , P = ρ P , P + dQ ρ P − Q, P ′ − Q Min (Q)
′
i~ 2M
Z


−ρ P , P ′ dQMin (Q) . (164)

We still have to evaluate the function Min (Q), which can be clearly interpreted
as the rate of collisions leading to a momentum gain Q of the Brownian
particle,
Z 3
ngas (2π~)
Min (Q) = dp0 µ (p0 ) |p0 | σ (p0 ) |hp0 − Q|T0 |p0 i|2 . (165)
m Ω
44 Klaus Hornberger

It involves the momentum matrix element of the on-shell T0 -matrix, which,

according to elastic scattering theory [33], is proportional to the scattering
amplitude f ,
!
f (pf , pi ) p2f p2i
hpf |T0 |pi i = δ − . (166)
2π~ 2 2

The delta function ensures the conservation of energy during the collision.
At first sight, this leads to an ill-defined expression since the matrix element
(166) appears as a squared modulus in (165), so that the tree-dimensional
integration is over a squared delta function.
The appearance of this problem can be traced back to our disregard of
the projection to the subset of incoming states in the definition (160) of Γ.
When evaluating Min we used the diagonal representation (162) for ρgas in
terms of (improper) momentum eigenstates, which comprise both incoming
and outgoing characteristics if viewed as the limiting form of a wave packet.
One way of implementing the missing projection to incoming states would be
to use a different convex decomposition of ρgas , which admits a separation
into incoming and outgoing contributions [34]. This way, Min can indeed be
calculated properly, albeit in a somewhat lengthy calculation. A shorter route
to the same result sticks to the diagonal representation, but modifies the
definition of S in a formal sense so that it keeps all outgoing state invariant17 .
The conservation of the probability current, which must still be guaranteed
by any such modification, then implies a simple rule how to deal with the
squared matrix element [34],
  !
3
(2π~)  2 f (pf , pi )2 p2f p 2
hpf |T0 |pi i −→ δ − i . (167)
Ω pi σ(pi ) 2 2
R 2
Here σ(p) = dΩ ′ |f (pn′ , pn)| is the total elastic cross section. With this
replacement we obtain immediately
Z !
2
ngas 2 p20 (p0 − Q)
Min (Q) = dp0 µ (p0 ) |f (p0 − Q, p0 )| δ − .(168)
m 2 2

As one would expects, the rate of momentum changing collisions is determined

2
by a thermal average over the differential cross section dσ/dΩ = |f | .
Also for finite mass ratios m/M a master equation can be obtained this
way, although the calculation is more complicated [35]. The resulting linear
quantum Boltzmann equation then describes on equal footing the decoherence
and dissipation effects of a gas on the quantum motion of a particle.
The “localizing” effect of a gas on the Brownian particle can now be seen,
after going into interaction picture in order to remove the unitary part of
17
In general, even a purely outgoing state gets transformed by S, since the definition
of the S-matrix involves a backward time-evolution [33].
 Introduction to decoherence theory 45

Fig. 2. The localization rate (171) describing the loss of wave-like behavior in a
Brownian particle state saturates for large distances at the average collision rate. In
contrast, the Caldeira-Leggett model predicts a quadratic increase beyond all bounds
(dashed line), see Eq. (127). This indicates that linear coupling models should be
taken with care if time scales are involved that differ strongly from the dissipation
time scale.

the evolution, and by stating the master equation in position representation.

From equation (164) one obtains

∂t ρ̃(X, X ′ ) = −F (X − X ′ )ρ̃(X, X ′ ) (169)

with localization rate [34]

Z ∞ Z
dΩ1 dΩ2  
F (x) = dv ν(v) ngas v 1 − eimv(n1 −n2 )·x/~
0 4π
2
×|f (mvn2 , mvn1 )| . (170)

Here, the unit vectors n1 , n2 are the directions of incoming and outgoing gas
particles associated to the elements of solid angle dΩ1 and dΩ2 and ν (v) is
the velocity distribution in the gas. Clearly, F (x) determines how fast the
spatial coherences corresponding to the distance x decay.
One angular integral in (170) can
be performed
in the case of isotropic
scattering, f (pf , pi ) = f cos pf , pi ; E = p2i /2m . In this case,
Z  Z  
∞ 1
 m 2
F (x) = dv ν(v) n gas v σ(mv) − 2π d (cos θ) f cos θ; E = v 2 
0 −1 2
   
θ mv |x|
× sinc 2 sin , (171)
2 ~

with sinc(x) = sin(x)/x and θ the (polar) scattering angle.

The argument of the sinc function is equal to the momentum
  exchange
during the collision times the distance in units of ~. As X − X ′  −→ 0 the
46 Klaus Hornberger

sinc approaches unity and the angular integral yields the total cross section
σ so that the localization rate vanishes, as required. At very small distances,
a second order expansion in the distance x is permissible and one obtains a
quadratic dependence [36], such as predicted by the Caldeira-Leggett model,
see Eq. (128). However, once the distance is sufficiently large so that the
scattered state can resolve whether the collision took place at position X or
X ′ the sinc function in (171) suppresses the integrand. It follows that in the
limit of large distances the localization rate saturates, at a value given by the
average collision rate F (∞) = hσvngas i, see Figure 2.
Decoherence in this saturated regime of large separations has been ob-
served, in good agreement with this theory, in molecular interference exper-
iments in the presence of various gases [37]. The intermediate regime be-
tween quadratic increase and saturation was also seen in such experiments
on momentum-exchange mediated decoherence, by studying the influence of
the heat radiation emitted by fullerene molecules on the visibility of their
interference pattern [38].
As a conclusion of this section, we see that the scattering approach per-
mits to incorporate realistic microscopic interactions transparently and with-
out approximation in the interaction strength. The results show clearly that
linear coupling models, which imply that decoherence rates grow above all
bounds, have a limited range of validity. They cannot be judged by their suc-
cess in describing dissipative phenomena. Frequent claims of “universality” in
decoherence behavior, which are based on these linear coupling models, are
therefore to be treated with care.

4.4 Decoherence of a quantum dot

As a second application of the monitoring approach, let us see how the dynam-
ics of an immobile object with discrete internal structure, such as an imple-
mentation of a quantum dot, gets affected by an environment of ideal gas par-
ticles. For simplicity, we take the gas again in the Maxwell state (156), though
different dispersion relations, e.g., in the case of phonon quasi-particles, could
be easily incorporated. The interaction between system and gas will be de-
scribed in terms of the in general inelastic scattering amplitudes determined
by the interaction potential.
In the language of scattering theory the energy eigenstates of the non-
motional degrees of freedom are called channels. In our case of a structureless
gas they form a discrete basis of the system Hilbert space. In the following,
the notation |αi, not to be confused with the coherent states of Sect. 3.4(b),
will be used to indicate the system eigenstates of energy Eα . In this channel
basis, ραβ = hα|ρ|βi, the equation of motion (155) takes on the form of a
general discrete master equation of Lindblad type,
 Introduction to decoherence theory 47

Eα + εα − Eβ − εβ X α0 β0
∂t ραβ = ραβ + ρα0 β0 Mαβ
i~
α0 β0
1X X
α0 α 1X X
ββ0
− ρα0 β Mγγ − ραβ0 Mγγ . (172)
2 α γ
2 γ
0 β0

The real energy shifts εα given below describe the coherent modification of the
system energies due to the presence of the environment. They are due to the
second term in (155) and are the analogue of the Lamb shift (148) encountered
in the weak coupling calculation. The incoherent effect of the environment,
on the other hand, is described by the set of complex rate coefficients
 
α0 β0
Mαβ = hα| TrE TΓ1/2 [|α0 ihβ0 | ⊗ ρgas ] Γ1/2 T† |βi. (173)

In order to calculate these quantities we need again to specify the rate operator
Γ. In the present case, it is naturally given in terms of the current density
operator j = ngas p/m of the impinging gas particles multiplied by the channel-
specific total scattering cross sections σ (p, α),
X |p|
Γ= |αihα| ⊗ ngas σ (p, α) . (174)
α
m

Like in Sect. 4.3, this operator should in principle contain a projection to the
subset of incoming states of the gas particle. Again, this can be accounted
for in two different ways in the calculation of the rates (173). By using a
non-diagonal decomposition of ρgas , which permits to disregard the outgoing
states, one obtains18
Z
α0 β0 α0 β0 ngas
Mαβ = χαβ dp dp0 µ (p0 ) fαα0 (p, p0 )
m2
 2 
∗ p − p20
×fββ0 (p, p0 ) δ + Eα − Eα0 , (175)
2m
with the Kronecker-like factor

1 if Eα − Eα0 = Eβ − Eβ0
χααβ
0 β0
:= (176)
0 otherwise.

The energy shifts are determined the real parts of the forward scattering
amplitude,
Z
2 ngas
εα = −2π~ dp0 µ (p0 ) Re [fαα (p0 , p0 )] . (177)
m
18
For the special case of factorizing interactions, Hint = A ⊗ BE , and for times
large compared to all system time scales this result can be obtained rigorously
in a standard approach [39], by means of the “low density limit” scaling method
[22, 2].
48 Klaus Hornberger

Some details of this calculation can be found in [32]. Rather than repeating
them here we note that the result (175) can be obtained directly by using the
diagonal representation (162) of ρgas and the multichannel-generalization of
the replacement rule (167),
3 α0 β0 ∗
(2π~) χαβ fαα0 (p, p0 ) fββ (p, p0 )
hαp1 |T0 |α0 p0 ihβ0 p0 |T†0 |βp1 i → p 0

Ω p0 m σ (p0 , α0 ) σ (p0 , β0 )
 2 
p − p20
×δ + Eα − Eα0 .(178)
2m

The expression for the complex rates simplifies further if the scattering

am-
plitudes are rotationally invariant, fαα0 cos (p, p0 ) ; E = p20 /2m . In this case
we have
Z ∞ Z 1
α0 β0 α0 β0
Mαβ = χαβ dvν (v) ngas vout (v) 2π d (cos θ)
0 −1
 m   m 2
×fαα0 cos θ; E = v 2 fββ ∗
cos θ; E = v . (179)
2 0
2
with ν (v)the velocity distribution like in (171) and
r
2
vout (v) = v 2 − (Eα − Eα0 ) (180)
m
the velocity of a gas particle after a possibly inelastic collision.
This shows that limiting cases of (172) display the expected dynamics. For
the populations R ραα it reduces to a rate2 equation, where the cross sections
σαα0 (E) = 2π d (cos θ) |fαα0 (cos θ; E)| for scattering from channel α0 to α
determine the transition rates,
Z m 
α0 α0
Mαα = dvν (v) ngas vout (v) σαα0 v2 . (181)
2
α0 β0
In the case of purely elastic scattering, on the other hand, i.e., for Mαβ =
αβ
Mαβ δαα0 δββ0 , the coherences are found to decay exponentially,

elastic
∂t |ραβ | = −γαβ |ραβ | . (182)

The corresponding pure dephasing rates are determined by the difference of

the scattering amplitudes,
Z Z 1
elastic
γαβ =π dvν (v) ngas vout (v) d (cos θ)
−1
  m   m 2

× fαα cos θ; v 2 − fββ cos θ; v 2  . (183)
2 2
 Introduction to decoherence theory 49

As one expects in this case, the better the scattering environment can dis-
tinguish between system states |αi and |βi the more coherence is lost in this
elastic process.
In the general case, the decay of off-diagonal elements will be due to a
combination of elastic and inelastic processes. Although little can be said
without specifying the interaction, it is clear that the integral over |fαα − fββ |2
in (183), a “decoherence cross section” without classical interpretation, is not
related to the inelastic cross sections characterizing the population transfer,
and may be much larger. In this case, the resulting decoherence will be again
much faster than the corresponding relaxation time scales.

5 Robust states and the pointer basis

We have seen that, even though the decoherence predictions of linear coupling
models has to be taken with great care, the general observation remains valid
−1
that the loss of coherence may occur on a time scale γdeco that is much shorter
−1
the relaxation time γ . Let us therefore return to the general description of
open systems in terms of a semigroup generator L, and ask what we can
say about a general state after a time t which is still small compared to
the relaxation time, but much larger than the decoherence time scale. From a
classical point of view, which knows only about relaxation, the state has barely
changed, but in the quantum description it may now be well approximated
by a mixture determined by particular projectors Pℓ ,
γ −1 ≪t≪γ −1 X
eLt : ρ −→−→−→
deco
ρt ≃ ρ′ = tr(ρPℓ ) Pℓ . (184)
ℓ

This set of projectors {Pℓ }, which depend at most weakly on t, is called

pointer basis [40] or set of robust states [41]. It is distinguished by the fact
that a system prepared in such a state is hardly affected by the environment,
while a superposition of two distinct pointer states decoheres so rapidly that
it is never observed in practice.
We encountered this behavior with the damped harmonic oscillator dis-
cussed in Sect. 3.4(b). There the coherent oscillator states remained pure
under Markovian dynamics, while superpositions between (macroscopically
distinct) coherent states decayed rapidly. Hence, in this case the coherent
states Pα = |αihα| can be said to form an (over-complete) set of robust states,
leading to the mixture
Z
′
ρ = dµ (α) tr(ρPα )Pα , (185)

with appropriate measure µ.

The name pointer basis is well-fitting because the existence of such robust
states is a prerequisite for the description of an ideal measurement device in a
50 Klaus Hornberger

quantum framework. A macroscopic–and therefore decohering–apparatus im-

plementing the measurement of an observable A is ideally constructed in such
a way that macroscopically distinct positions of the “pointer” are obtained for
the different eigenstates of A. Provided these pointer states of the device are
robust the correct values are observed with certainty if the quantum system is
in an eigenstate of the observable. Conversely, if the quantum system is not in
an eigenstate of A, the apparatus will not end up in a superposition of pointer
positions, but be found at a definite position, albeit probabilistically, with a
probability given by the Born rule.
The main question regarding pointer states is, given the environmental
coupling or the generator L, what determines whether a state is robust or
not, and how can we determine the set of pointer states without solving the
master equation for all initial states. It is fair to say that this issue is not fully
understood, except for very simple model environments, nor is it even clear
how to quantify robustness.
An obvious ansatz, due to Zurek [6, 42], is to sort all pure states in the
Hilbert space according to their (linear) entropy production rate, or rate of
loss of purity,
∂t Slin [ρ] = −2 tr (ρL(ρ)) . (186)
It has been called “predictability sieve” since the least entropy–producing and
therefore most predictable states are candidate pointer states [6]. This formu-
lation is convenient due to its simplicity, though a time-integrated quantity
might be more appropriate.
In the following, a different approach will be described, following the pre-
sentation in [3, 43]. It is based on a time evolution equation for robust states.
Since such an equation must distinguish particular states from their linear
superpositions it is necessarily nonlinear.

5.1 Nonlinear equation for robust states

We seek a nonlinear time evolution equation for robust pure states Pt which,
on the one hand side, preserves their purity, and on the other, keeps them as
close as possible to the evolved state following the master equation.
The possibly simplest nonlinear equation keeping a pure state pure is a
nonlinear extension of the Heisenberg form for the infinitesimal time step,
 
1
Pt+δt = Pt + δt [At , Pt ] + [Pt , [Pt , Bt ]] , (187)
i

where A and B are hermitian operators. In fact, the unitary part can by
absorbed into the nonlinear part by introducing the hermitian operator
Xt = −i[At , Pt ] + Bt . It “generates” the infinitesimal time translation of the
projectors (and may be a function of Pt ),

Pt+δt = Pt + δt[Pt , [Pt , Xt ]] . (188)

 Introduction to decoherence theory 51

With this choice one confirms easily that the evolved operator has indeed the
properties of a projector, to leading order in δt,
P†t+δt = Pt+δt (189)
and
2
(Pt+δt ) = Pt+δt + O(δt2 ). (190)
The corresponding differential equation reads
Pt+δt − Pt
∂t Pt = = [Pt , [Pt , Xt ]] . (191)
δt
To determine the operator Xt one minimizes the distance between the time
derivatives of the truly evolved state and the projector. If we visualize the set
of all pure states as the extremal points in the convex set of states, then the
trajectory starting from a pure state will in general dive into the interior set
of mixed states, under the time evolution generated by L. With the help of
the minimization the operator Xt is chosen such that Pt remains extremal,
but as close as possible to the truly evolved state.
The (Hilbert-Schmidt) distance between the time derivatives can be cal-
culated as
h i
2
k L(Pt ) −∂t Pt k2HS = tr (Z − [Pt , [Pt , Xt ]])
| {z }
≡Z
 
2
= tr Z2 − 2(Z2 Pt − (ZPt ) )
 
+2 tr (Z − X)Pt − ((Z − X) Pt )2 . (192)
We note that the first term is independent of X, whereas the second one is
minimal for (Z − Xt )Pt = λPt . With the obvious choice Xt = Z ≡ L(Pt ) one
gets a nonlinear evolution equation for robust states Pt , which is trace and
purity preserving [43],
∂t Pt = [Pt , [Pt , L(Pt )]]. (193)
It is useful to write down the equation in terms of the vectors |ξi which
correspond to the pure state Pt = |ξihξ|,
∂t |ξi = [L(|ξihξ|) − hξ|L(|ξihξ|)|ξi]|ξi. (194)
| {z }
“decay rate”

If we take L to be of the Lindblad form (79) the equation reads

1 X  †   1
† †

∂t |ξi = H|ξi + γk hLk iξ Lk − hLk iξ − L L − hLk Lk i |ξi
i~ 2 k
k
1
− hHiξ |ξi. (195)
i~
Its last term is usually disregarded because it gives rise only to an additional
phase if hHiξ is constant. The meaning of this equation is best studied in
terms of concrete examples.
52 Klaus Hornberger

5.2 Applications

(a) Damped harmonic oscillator

Let us start with the damped harmonic oscillator discussed in Sect. 3.4(b).
By setting H = ~ωa† a and L = a Eq. (195) turns into
 
† † 1 † †


∂t |ξi = −iωa a|ξi + γ ha iξ (a − haiξ ) − a a − ha aiξ |ξi. (196)
2

Note that the first term of the non-unitary part vanishes if |ξi is a coherent
state, i.e, an eigenstate of a. This suggests the ansatz |ξi = |αi which leads to
h γ  † γ 2i
∂t |αi = −iω − αa + |α| |αi. (197)
2 2
It is easy to convince oneself that this equation is solved by
2
/2 αt a†
|αt i = |α0 e−iωt−γt/2 i = e−|αt | e |0i, (198)

with αt = α0 exp (−iωt − γt/2). It shows that the predicted robust states
are indeed given by the slowly decaying coherent states encountered in
Sect. 3.4(b).

(b) Quantum Brownian motion

A second example is given by the Brownian motion of a quantum particle.

The choice
√
p2 8π
H= and L = x (199)
2m Λth
yields a master equation of the form (117) but without the dissipation term.
Inserting these operators into (195) leads to

p2 4π
∂t |ξi = |ξi − γ 2 [(x − hxiξ )2 − h(x − hxiξ )2 iξ ]|ξi. (200)
2mi~ Λth | {z }
σξ2 (x)

The action of the non-unitary term is apparent in the position representation,

ξ(x) = hx|ξi. At positions x which are distant from mean position hxiξ as
 2
1/2
compared to the dispersion σξ (x) = h x − hxiξ iξ the term is negative and
the value ξ(x) gets suppressed. Conversely, the part of the wave function close
to the mean position gets enhanced,

suppressed if |x − hxiξ | > σξ (x)
hx|ξi =
enhanced if |x − hxiξ | < σξ (x)
 Introduction to decoherence theory 53

This localizing effect is countered by the first term in (200) which causes the
dispersive broadening of the wave function. Since both effects compete we
expect stationary, soliton-like solutions of the equation.
Indeed, a Gaussian ansatz for |ξi with ballistic motion, i.e., hpiξ = p0 ,
hxiξ = x0 + p0 t/m,
and a fixed width σξ (x) = σ0 solves (200) provided
s  1/2
2 1 kB T 2 ~3
σ0 = Λ = , (201)
4π ~γ th 4γm2 kB T

see (129). As an example, let us consider a dust particle with a mass of 10 µg

in the interstellar medium interacting only with the microwave background
of T = 2.7K. If we take a pretty small (an arbitrary) relaxation rate of γ =
10−25 s−1 , corresponding to the inverse lifetime of the universe, then the width
of this solitonic gaussian is σ0 = 1pm, indeed much smaller than its size.

Acknowledgments

Many thanks to Álvaro Tejero Cantero who provided me with his notes,
typed with the lovely TEXmacs program during the lecture. The present text
is based on his valuable input. I am also grateful to Marc Busse and Bassano
Vacchini for helpful comments on the manuscript.
This work was supported by the Emmy Noether program of the DFG.

References
1. E. Joos, H. D. Zeh, C. Kiefer, D. Giulini, J. Kupsch, and I.-O. Stamatescu, Deco-
herence and the Appearance of a Classical World in Quantum Theory, Springer,
Berlin, 2nd edition, 2003.
2. H.-P. Breuer and F. Petruccione, The Theory of Open Quantum Systems, Oxford
University Press, Oxford, 2002.
3. W. T. Strunz, Decoherence in Quantum Physics, in Coherent Evolution in Noisy
Environments, edited by A. Buchleitner and K. Hornberger, Lecture Notes in
Physics 611, Berlin, 2002, Springer.
4. G. Bacciagaluppi, The Role of Decoherence in Quantum Mechanics, in The
Stanford Encyclopedia of Philosophy, edited by E. N. Zalta, Stanford University,
2005, http://plato.stanford.edu.
5. M. Schlosshauer, Decoherence, the measurement problem, and interpretations
of quantum mechanics, Rev. Mod. Phys. 76, 1267–1305 (2004).
6. W. H. Zurek, Decoherence, Einselection, and the Quantum Origins of the Clas-
sical, Rev. Mod. Phys. 75, 715–775 (2003).
7. J. P. Paz and W. H. Zurek, Environment-Induced Decoherence and the Transi-
tion From Quantum to Classical, in Les Houches Summer School Series, edited
by R. Kaiser, C. Westbrook, and F. David, volume 72, page 533, Springer-Verlag,
2001.
8. A. Bassi and G. Ghirardi, Dynamical Reduction Models, Phys. Rev. 379, 257
(2003).
54 Klaus Hornberger

9. K. Kraus, States, Effects and Operations: Fundamental Notions of Quantum

Theory, Springer, Berlin, 1983.
10. P. Busch, P. J. Lahti, and P. Mittelstaed, The Quantum Theory of Measurement,
Springer-Verlag, Berlin, 1991.
11. A. S. Holevo, Statistical Structure of Quantum Theory, Springer, Berlin, 2001.
12. C. W. Helstrom, Quantum Detection and Estimation Theory, Academic Press,
New York, 1976.
13. A. Chefles, Quantum state discrimination, Contemp. Phys. 41, 401–424 (2000).
14. G. M. Palma, K.-A. Suominen, and A. K. Ekert, Quantum computers and
dissipation, Proc. R. Soc. Lond. A 452, 567 (1996).
15. D. F. Walls and G. J. Milburn, Quantum Optics, Springer, Berlin, 1994.
16. M. Hillery, R. F. O’Connell, M. O. Scully, and E. P. Wigner, Distribution
functions in physics: Fundamentals, Phys. Rep. 106, 121–167 (1984).
17. U. Weiss, Quantum Dissipative Systems, World Scientific, Singapore, 2nd edi-
tion, 1999.
18. P. Machnikowski, Change of decoherence scenario and appearance of localization
due to reservoir anharmonicity, Phys. Rev. Lett. 96, 140405 (2006).
19. R. Doll, M. Wubs, P. Hänggi, and S. Kohler, Limitation of entanglement due
to spatial qubit separation, Europhys. Lett. 76, 547–553 (2006).
20. E. B. Davies, Quantum Theory of Open Systems, Academic Press, London,
1976.
21. H. Spohn, Kinetic equations from Hamiltonian dynamics: Markovian limits,
Rev. Mod. Phys. 52, 569–615 (1980).
22. R. Alicki and K. Lendi, Quantum Dynamical Semigroups and Applications,
Springer, Berlin, 1987.
23. F. Petruccione and B. Vacchini, Quantum description of Einstein’s Brownian
motion, Phys. Rev. E 71, 046134 (2005).
24. H. Carmichael, An Open Systems Approach to Quantum Optics, Springer,
Berlin, 1993.
25. K. Mølmer, Y. Castin, and J. Dalibard, Monte Carlo wave-function method in
quantum optics, J. Opt. Soc. Am. B 10, 524–538 (1993).
26. M. B. Plenio and P. L. Knight, The quantum-jump approach to dissipative
dynamics in quantum optics, Rev. Mod. Phys. 70, 101–144 (1998).
27. J. M. Raimond, M. Brune, and S. Haroche, Colloquium: Manipulating Quantum
Entanglement with Atoms and Photons in a Cavity, Rev. Mod. Phys. 73, 565 –
582 (2001).
28. S. Haroche, Mesoscopic superpositions and decoherence in quantum optics, in
Quantum entanglement and information processing, edited by D. Estève, J.-M.
Raimond, and J. Dalibard, Les Houches 2003, Elsevier, Amsterdam, 2004.
29. A. O. Caldeira and A. J. Leggett, Path Integral Approach to Quantum Brownian
Motion, Physica A 121, 587–616 (1983).
30. W. G. Unruh and W. H. Zurek, Reduction of a Wave Packet in Quantum
Brownian Motion, Phys. Rev. D 40, 1071–1094 (1989).
31. L. Lukacs, Characteristic Functions, Griffin, London, 1966.
32. K. Hornberger, Monitoring approach to open quantum dynamics using scattering
theory, eprint quant-ph/0612078 (2006).
33. J. R. Taylor, Scattering Theory, John Wiley & Sons, New York, 1972.
34. K. Hornberger and J. E. Sipe, Collisional Decoherence Reexamined, Phys.
Rev. A 68, 012105 (2003).
 Introduction to decoherence theory 55

35. K. Hornberger, Master equation for a quantum particle in a gas, Phys. Rev.
Lett. 97, 060601 (2006).
36. E. Joos and H. D. Zeh, The Emergence of Classical Properties Through Inter-
action with the Environment, Z. Phys. B: Condens. Matter 59, 223–243 (1985).
37. K. Hornberger, S. Uttenthaler, B. Brezger, L. Hackermüller, M. Arndt, and
A. Zeilinger, Collisional Decoherence Observed in Matter Wave Interferometry,
Phys. Rev. Lett. 90, 160401 (2003).
38. L. Hackermüller, K. Hornberger, B. Brezger, A. Zeilinger, and M. Arndt, De-
coherence of Matter Waves by Thermal Emission of Radiation, Nature 427,
711–714 (2004).
39. R. Dümcke, The low density limit for an N-level system interacting with a free
bose or fermi gas, Commun. Math. Phys. 97, 331–359 (1985).
40. W. H. Zurek, Pointer Basis of Quantum Apparatus: Into What Mixture Does
the Wave Packet Collapse?, Phys. Rev. D 24, 1516–1525 (1981).
41. L. Diósi and C. Kiefer, Robustness and Diffusion of Pointer States, Phys. Rev.
Lett. 85, 3552–3555 (2000).
42. W. H. Zurek, S. Habib, and J. P. Paz, Coherent states via decoherence, Phys.
Rev. Lett. 70, 1187–1190 (1993).
43. N. Gisin and M. Rigo, Relevant and irrelevant nonlinear Schrödinger equations,
J. Phys. A: Math. Gen. 28, 7375–7390 (1995).
Contents

Introduction to decoherence theory

Klaus Hornberger . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1 The concept of decoherence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Decoherence in a nutshell . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 General scattering interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Decoherence as an environmental monitoring process . . . . . . . . . 5
1.4 A few words on nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2 Case study: Dephasing of qubits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.1 An exactly solvable model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2 The continuum limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3 Dephasing of N qubits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3 Markovian dynamics of open quantum systems . . . . . . . . . . . . . . . . . . . 19
3.1 Quantum dynamical semigroups . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.2 The Lindblad form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.3 Unravelling of Markovian master equations . . . . . . . . . . . . . . . . . . 24
3.4 Exemplary master equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4 Microscopic derivations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.1 The weak coupling formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.2 The monitoring approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.3 Collisional decoherence of a Brownian particle . . . . . . . . . . . . . . . 41
4.4 Decoherence of a quantum dot . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
5 Robust states and the pointer basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.1 Nonlinear equation for robust states . . . . . . . . . . . . . . . . . . . . . . . . 50
5.2 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53