Você está na página 1de 11

JOURNAL OF CHEMICAL PHYSICS

VOLUME 116, NUMBER 21

1 JUNE 2002

Theory of open quantum systems


Ruixue Xu and YiJing Yana)
Open Laboratory of Bond-Selective Chemistry, University of Science and Technology of China, Hefei,
China, Department of Chemistry, Hong Kong University of Science and Technology, Kowloon,
Hong Kong SAR, China, and Center for Theoretical Chemical Physics, Fudan University,
Shanghai 200433, China

Received 2 January 2002; accepted 8 March 2002


A quantum dissipation theory is constructed with the systembath interaction being treated
rigorously at the second-order cumulant level for both reduced dynamics and initial canonical
boundary condition. The theory is valid for arbitrary bath correlation functions and time-dependent
external driving fields, and satisfies correlated detailed-balance relation at any temperatures. The
general formulation assumes a particularly simple form in driven Brownian oscillator systems in
which the correlated driving-dissipation effects can be accounted for exactly in terms of local-field
correction. Remarks on a class of widely used phenomenological quantum master equations that
neglects the bath dispersion-induced dissipation are also made in contact with the present theory.
2002 American Institute of Physics. DOI: 10.1063/1.1474579

I. INTRODUCTION

mation, despite the fact that it could be treated in a selfconsistent manner together with the so-called Redfield approximation by neglecting the effects of bath dispersion, left
the resulting QME incomplete in the second order of
systembath interaction.3133 A generalized non-Markovian
QDT in the presence of external driving fields has also been
constructed in either the time-local3033 or the memory dissipation kernel prescriptions30 under the weak systembath
interaction and the initial factorization approximations.
In this paper, we shall use a similar algebraic
approach3032 to construct an improved QDT, in which the
second-order correlated systembath canonical state is used
explicitly as the initial condition. The resulting QDT is exact
up to the second-order cumulant of system-bath interaction
in the presence of arbitrary coherent driving fields. We shall
call the QDT to be developed CS-QDT complete secondorder QDT. Consequently, the CS-QDT also satisfies the
correlated detailed-balance relation up to the second-order
cumulant level, and, according to a recent work by Royer,34
can be of a dynamical semigroup as well in a time-coarsegrained limit. Note that a sort of CS-QDT in a memory kernel prescription has been constructed recently by Meier
and Tannor29 using the NakajimaZwanzig projection
technique.1 4 However, the bath reference state used by them
was neither canonical nor Hermitian.
The remainder of this paper is organized as follows. We
start with the background of quantum statistical mechanics
Sec. II A for the two-point correlation functions of generalized bath interaction operators. With the tedious derivation
presented in Appendix A, we summarize the final CS-QDT
formalism in terms of an integro-differential equation in Sec.
II B, and coupled differential equations of motion EOM in
Sec. II C. The possibility of recasting the CS-QDT in terms
of EOM is made via an extension of the MeierTannors
parameterization algorithm29 to the generalized spectral densities for cross-correlation functions cf. Appendix B. The
analysis of the reduced canonical density operator versus the

As a central topic in quantum statistical mechanics, dissipation plays important roles in almost all fields of modern
science. The concept of quantum dissipation is physically
related to a reduced description on a quantum open system in
contact with its quantum surroundings. The surroundings
consist of a practically infinite number of degrees of freedom
and act statistically as a whole identity, referred as bath, on
the open system. The key theoretical quantity is the reduced
density operator (t)trB T(t), i.e., the partial trace of the
total system-plus-bath density operator T(t) over the bath
degrees of freedom. Quantum dissipation theory QDT describes not only the evolution of (t), but also the equilibrium behavior of the reduced system as (t) eq(T)
trBe H T /(k B T) . The latter property is also referred as the
detailedbalance relation of the QDT.
Formally, an exact QDT can in principle be constructed
via
NakajimaZwanzigMori
projection
operator
techniques.19 However, dissipations in almost all practical
cases have to be treated with models and/or approximations.
As a result, there is a variety of nonequivalent formulations
of QDT, even in their asymptotic equilibrium behaviors and
within the framework of Markovian bath constructions.1024
Non-Markovian theories15,2530 are further distinct by their
memory versus time-local dissipation kernel constructions,
as long as certain practically inevitable approximations are
adopted.
Recently, we exploited a rather simple algebraic approach to construct a unified quantum master equation
QME3133 that bridges the BlochRedfield theory1018 and
a class of FokkerPlanck equations.1921 Analyzed there in
detail were the nature of approximations involved in these
two most commonly used QDTs. In particular, we pointed
out that the commonly adopted initial factorization approxia

Electronic mail: yyan@ust.hk

0021-9606/2002/116(21)/9196/11/$19.00

9196

2002 American Institute of Physics

J. Chem. Phys., Vol. 116, No. 21, 1 June 2002

Theory of open quantum systems

stationary solution to the CS-QDT is presented in Sec. III.


The fact that the CS-QDT satisfies the correlated detailedbalance at least to second order is further confirmed in Appendix C via the direct bath average of the total system-plusbath canonical ensembles. Provided in Appendix D is an
effective Hamiltonian to the second-order reduced canonical
density operator. We show that the effective Hamiltonian recovers the commonly used CaldeiraLeggetts renormalization energy21,35 in the high-temperature or Markovian limit.
In Sec. IV, the CS-QDT developed in Sec. II B is implemented in driven Brownian oscillator DBO systems in
which the aforementioned integro-differential equation can
be analytically reduced to a single differential equation. It is
found that the correlated driving-dissipation effects in DBO
systems can be accounted for by effective local fields. Numerical demonstrations on both the reduced canonical density operator and its laser-induced dynamics are also presented for the DBO systems. In Sec. V, we summarize CSQDT and conclude this paper with remarks on the class of
commonly used QME quantum master equations in contact
with the CS-QDT.

A. Prelude and bath statistics

Consider a molecular system (H s ) embedded in a dissipative bath (h B) and interacting via the molecular dipole
with an arbitrary external classical field (t).
operator
The total Hamiltonian assumes
H T t H t h BH ,

1a

t ,
H t H s

1b

a Q a F a .

1c

The systembath interaction H is assumed to be weak and


will be treated statistically to second order for its effects on
the reduced system density operator under coherent external
driving. The statistical average is defined via the unperturbed
canonical bath ensemble as

O trB O e h B /trB e h B .

Throughout this paper, we set 1 and 1/(k BT).


In Eq. 1c, H is decomposed in terms of Langevin
modes in which Q a and F a are Hermite operators in the
system and bath spaces, respectively. Without loss of generality, we choose F a (t) F a (0) 0. Here, F a (t)
e ih B t F a e ih B t . As a second-order theory to be developed,
the effects of bath interaction are completely determined via
their correlation functions36
ab t F a t F b .
C

ab ()0, and the


They satisfy the boundary condition, C
detailed-balance and the symmetry relations
ab
ab ti C
ba t .
* t C
C

1
2

D ab

1
2i

ab t e i t C
ba
* t ,
dt e i t C

The bath interaction spectral C ab ( ) and dispersion


D ab ( ) functions are defined as

5a

ab t e i t C
ba
* t ,
dt e i t C

5b

ab t e C ba ,
dt e i t C

6a

or equivalently
C ab

1
2

D ab iP

ab t
dt e i t C

C ab

6b

Here, P denotes the principal part. Obviously, we have

ab t C ab iD ab .
dt e i t C

The second identity in Eq. 6a is the detailed-balance relation in terms of spectral functions. The frequency-domain
symmetry relations read as cf. Eq. 5

* C ba ,
C ab

II. QUANTUM DISSIPATION THEORY

C ab

9197

* D ba .
D ab

We can easily show that C aa ( )0, or more generally, the


matrix C ab ( ) is positively defined. We shall also show
later that, as far as the reduced canonical density operator is
concerned, the bath dispersion is the primary source for energy renormalization cf. Appendices C and D.
We are now in the position to develop the CS-QDT that
governs the evolution of reduced density operator, (t)
trB T(t). Note that most existing QDT formulations invoke a so-called initial factorization approximation, in which
the initial total density operator T(t 0 ) assumes to be of a
factorized form. This is a questionable approximation as
far as a second-order theory is concerned. In this
work, the initial total density operator prior to external field
actions assumes to be canonical, T(t 0 )
e H T(t 0 ) /Tr e H T(t 0 ) . As a result, the initial time, which
in principle is t 0 , can be chosen at any moment prior
to the external field action; i.e.,

tt 0 0.

The second-order canonical initial T(t 0 ) will be used explicitly and the resulting theory is therefore complete at the
second-order level.
To conclude this subsection, let us introduce the Liouvillian L(t) and its associating propagator G(t, ):
L t A H t ,A ,

i G t, / tL t G t, .

10

The system-only field-free Liouvillian Ls is defined similarly. Its Liouville-space propagator is given by
Gs t, Gs t e iLs (t ) .

11

B. Final formulation in the time-local prescription

With the derivation detailed in Appendix A, the final


time-local form of CS-QDT, valid for arbitrary bath interaction spectra and external coherent driving fields, is summarized as follows:

9198

J. Chem. Phys., Vol. 116, No. 21, 1 June 2002

t iL t t R t t ,
R t t

a Q a ,Q a t t t Q a t ,

R. Xu and Y. J. Yan

12a
12b

C. Equations of motion for the correlated drivingdissipation dynamics

The two nonequivalent sets of EOM for CS-QDT to be


presented soon are constructed with the following form of
correlation functions cf. Eq. B5:

with
a t
Q

ab t
C
t

sa i
Q

ab t G t, Q b
d C

d C ab t

,Gs Q b .
G t,

13

The above second identity, which is obtained by using Eq.


A12, separates the dissipation effects into the field-free part
and the correlated driving-dissipation part. The field-free
sa is time-independent and given explicitly by
contribution Q
cf. Eqs. 7, 11, and 13
sa
Q

b
b

a,b,m

ab
m
Q a ,K
t t H.c. .

17

ab
m
(t) is the individual contribution of Eq. 16 to the
Here, K
a (t) the last term of Eq. 13. By changing the order of
Q
integration
variables
and
noting
that Lsf( )A

( ),A , we have

ab e iLs Q b
d C

ab
ab
m
K
t i m

14

Here, C ab (Ls ) and D ab (Ls ) are functions of the fieldfree Liouvillian Ls , defined via the bath interaction spectrum
Eq. 6a and dispersion Eq. 6b, respectively. The
sa -term results in the stationary contribution, Rs , to the
Q
total dissipation R(t) Eq. 12b. We have

16

t iL t t Rs t

b C ab Ls iD ab Ls Q b

Rs t

ab
m t

As C ab (t) must satisfy the detailed-balance and symmetry


ab
ab
,m
relations, Eq. 4, the parameters m
in Eq. 16
should be properly structured. Meier and Tannor29 proposed
a bath spectral density parametrization method to properly
aa (t) in the multiexpoexpress autocorrelation functions C
nential form. In Appendix B, we extend their method to
ab (t).
cross-correlation functions C
The time-local CS-QDT Eq. 12 with Eq. 16 can
now be recast as

ab t Gs t, Q b
d C

a.
a i

m mab e

a t t a i a .
Q a , a i
15

a (t)
The double-integration term in Eq. 13, denoted as Q
s

Q a (t)Q a , accounts for the correlated driving


dissipation effects on the reduced dynamics. Obviously,
a (t 0 )0 before external field action. In Sec. IV, we shall
Q
present a model harmonic system in which the double inte a (t) can be readily implemented. However, the
gration in Q
integro-differential form of CS-QDT Eq. 12 with Eq. 13
is often numerically expensive for anharmonic systems. To
facilitate this problem, there has been considerable effort on
arriving at a QDT in terms of differential equations of motion EOM.26 30 In the following subsection, we shall
present two sets of EOM. One is inhomogeneous and uncoupled, constructed directly for the above time-local
a (t). Another is homogeneous and coupled, amounting to
Q
a CS-QDT cf. Eq. 22 that differs from either Eq. 12 or
Eq. A13 only at high orders in systembath interaction.

ab

d e m (t )

G t, Lsf Gs , Q b .

18

It leads to
ab
ab
ab ab
m
dK
t /dt P m
t iL t m
Km t ,

19

with the inhomogeneous term


ab
ab
ab 1
Lsf t iLs m
Pm
Qb .
t i m

20

The initial conditions at t 0 before external driving are


ab
m
(t 0 )0. As the nature of the time-local form, Eq. 19 is
K
ab
m
(t) are obtained, we can
decoupled from Eq. 17. Once K
then solve Eq. 17 for (t) with the initial condition (t 0 )
eq . The uncoupled set of EOM Eqs. 17 and 19 that
is equivalent to Eq. 12 is often numerically inefficient due
to its inhomogeneous and nonlinear in nature.
Alternatively, we may construct a CS-QDT in terms of a
set of coupled linear EOM. To do that, let us consider
d
ab
ab
ab t t dK
m
m
K
t /dt t K
t d t /dt
dt m
ab
m
dK
t /dt t
ab
m
K
t iL t t
ab
ab
Pm
t t iL t m

ab
m
K
t t .

21

J. Chem. Phys., Vol. 116, No. 21, 1 June 2002

Theory of open quantum systems

Obviously, the above equation is identical to Eq. 19 up to


the second-order in systembath interaction. The resulting
CS-QDT is then cf. Eq. 17

t iL t Rs t

a,b,m

ab
Q a ,K m
t H.c. ,

22a
ab
ab
ab
ab
K m
t Pm
t t iL t m
Km
t .

22b

The initial conditions for the above set of coupled EOM are
(t 0 ) eq and K mab (t 0 )0.
III. STATIONARY SOLUTION VERSUS REDUCED
CANONICAL DENSITY OPERATOR

The two nonequivalent forms of CS-QDT, Eq. 12 or


17 and Eq. 22, share the same stationary solution in the
absence of continuous wave cw driving. This is about the
detailed-balance relation implied in the CS-QDT

t eq T .

23

Thus, the reduced canonical thermal equilibrium density


operator can be obtained via (t)0 in the absence of external fields, i.e.,
s eq T iLs Rs eq T 0.

24

Note that s is of a commutator form so that Tr (t) is conserved, leading to that det s 0 in Liouville space. This
confirms that zero must be an eigenvalue of s and eq(T) is
eq
;m,n
its associating Liouville-space eigenvector. Here, mn
0,...,N1, in a given Hilbert-space representation, is rearranged as eq ; mNn, and, accordingly, the Liouvilles
space operator as a matrix. Unless in pure-dephasing
cases, the eigenvalue zero is nondegenerate, thus eq(T) is
determined unambiguously in Eq. 24.
To show that the resulting eq(T) is generally different
from its zeroth-order counterpart, let us denote

eq T Z 1 e H s eq .

25

Here, Z is the normalization constant. Physically, eq is the


systembath-induced correction to the reduced canonical
density operator. By using the spectrum detailed-balance relation Eq. 6a, we can show that31
a e H s a e H s ,

26

and the a terms in R s e H s vanish cf. Eq. 15. Equation


24 can then be recast as
s eqi

a Q a , a e H e H a .
s

ebrated CaldeiraLeggetts renormalization Hamiltonian in


the high temperature and/or Markovian limit.
It is noticed in some literature reports, e.g., Ref. 37, that
the reduced thermal equilibrium was claimed to be the unperturbed, zeroth-order canonical reduced density operator.
This illusion may be caused by the fact that
s
s
R mm,nn
/R nn,mm
e mn in the H s representation, with
mn being the transition frequencies. However, the above
relation cannot lead to eq in Eq. 27 being zero. In fact,
Eq. 27 can be recast as ( s eq) mn j R smn, j j e j .
The spinboson system considered in Ref. 38 is an exception, in which R smn, j j 0; for mn, thus ( s eq) mn 0
and the resulting eq(T) from Eq. 24 does coincide with the
zeroth-order canonical density operator.

IV. DRIVEN BROWNIAN OSCILLATOR MODEL


A. Analytical local-field formulation

Exemplified in this section is a driven Brownian oscillator DBO system, whose Hamiltonian H s , dipole operator
, and dissipation coupling mode Q are modeled as
H s a a 1/2 ,

2 1/2 a a ,

Q2 1/2 q a q a .

28a
28b

* are complex numbers, and a (a ), which satHere, q q

isfies a ,a 1, is the annihilation creation operator for


the bare, uncorrelated oscillator of frequency . Note that as
a second-order cumulant formulation, the time-local CSQDT, Eq. 12, is exact for the DBO system. In Eq. 28 and
hereafter, we remove the Qs index (a or b) as only a single
Langevin mode Q is in study.
The simplicity of DBO Eq. 28 arises from that Ls a
a and L s a a . These relations result in that
f (Ls )a f ()a and f (Ls )a f ()a , for any given
function f ( ). Thus, we have cf. Eq. 14
s i 2 1/2 a a ,
Q

29

with

q C iD .

30

Rs Eq. 15 is therefore determined. Moreover, the commutator contained in the integrand of Eq. 13 now becomes
a c-number as
,Gs Q b 21 a a ,q e i( ) a

27

It implies that eq0 in general cases. In Appendix C, we


will confirm that the resulting eq is identical to that obtained via trB e (H s h B H ) /Z T , at least up to the second
order. Obviously, the second-order correction can be obtained via the above equation by replacing s with iLs cf.
Eq. C1. It thus concludes that the bath dispersion functions
D ab ( ), responsible for a cf. Eq. 14, are the primary
source for eq . We shall present in Appendix D an effective
Hamiltonian correction to eq , which recovers the cel-

9199

q e i( ) a
12 q e i( ) q e i( ) .
31
Substituting it into Eq. 13, followed by some minor algebra, we obtain
t Q
s
Q
where

i
2

d t ,

32

9200

J. Chem. Phys., Vol. 116, No. 21, 1 June 2002

R. Xu and Y. J. Yan

FIG. 1. The temperature dependence of relative x-broadening, x r


(x 2 /x 20 1) 1/2 at the specified value of systembath interaction strength,
B0.1. That for B0.001 not shown is similar but about 10 times
smaller. x 20 coth(/2)/2 is taken from the unperturbed, bare harmonic
system. In comparison with the bare system, eq for the Brownian oscillator
is only broadened along the x direction. The figure contains therefore all
information about the reduced canonical density operator.

.
d q e i( ) q e i( ) C
33

The second term in the right-hand side of Eq. 32 is a


c-number which contributes to the reduced dynamics Eq.
12 only via its imaginary part. Denoting

t Re

d t ,

34

thus, the final CS-QDT formulation for DBO Eq. 28 is

t i H s t Q t , t Rs t .

35

The above formulation leads to the following implications:


i (t) serves as an effective local-field correction, acting on
the system via Q cf. Eq. 35; ii (t) is the convolution
between the systembath coupling response (t) Eq. 33
and the driving field (t) cf. Eq. 34; iii (t) depends
only on the imaginary part of the bath correlation function
(t) cf. Eq. 33; iv the effective local-field correction
C
(t) at time t depends on the incident field (t ) at t t

FIG. 2. Evolution of the laser-induced wave packet dynamics, (t)


(t) eq , presented in the dimensionless coordinate space, for two indicated values of systembath coupling strength at T0 K. The excitation
field is a transform-limited resonance Gaussian-pulse, centered at t0, with
temporal width of F50/ and strength of F0.1. See the text for details. Before excitation, the system with each value of B is initially in
thermal equilibrium eq cf. Fig. 1.

s (t)
FIG. 3. Evolution of the laser-induced excess energy, H
Tr H s (t) , for the same system as Fig. 2. Note the curve for the strong
systembath coupling ( B0.1) is enlarged by factor of 5.

t cf. Eq. 34. Preliminary numerical results also indicate that (t) usually tends to effectively reduce the system
field coupling strength.
B. Numerical results

In the following calculation, the dissipation coupling


mode is chosen as Q2 1/2(a a ) i.e. Eq. 28b with
q 1. The interaction bath spectral density J( )C( )
C( ) cf. Eq. B1 assumes Ohmic
J B e / c .

36

The Ohmic cutoff frequency is fixed at c . The system


bath coupling strength is considered at two values, B0.1
and 0.001.
Numerical implementation is carried out in the H s representation using 30 eigenstates. The reduced canonical density operator eq(T) is determined via Eq. 24 as an eigenvalue problem, and also found to be identical to that obtained
in the time-domain method by propagating CS-QDT with
arbitrary initial density matrix in study. Propagation of Eq.
35 is via short-iterative-Arnoldi algorithm17,39,40 in single
precision with error of 107 .
Let us first study the reduced canonical density operator
eq(T), which is found to be Gaussian wave packet in phase
space. Thus, it is completely characterized by the first five
moments in the dimensionless phase space: x , p , x 2 , p 2 , and
xppx, where A Tr A eq(T) . The unperturbed, zerothorder values for these parameters are x 0 p 0 (xppx) 0
0, and x 20 p 20 coth(/2)/2, where A 0 Tr(Ae H s )/
Tr e H s . The calculated x , p , p 2 , and xppx are found to
be same as their unperturbed counterparts; however, x 2 is
enlarged even at T0 K. This is physically consistent with
the model in which the Langevin force acts only on the oscillators coordinate. Therefore, the calculated eq(T) is an
x-broadened Gaussian wave packet in comparison with the
(0)
unperturbed eq
. Figure 1 depicts the relative x broadening,
2 2
x r (x /x 0 1) 1/2, as a function of temperature, with the
systembath interaction strength B0.1. That for B
0.001 not shown behaves similar but is about 10 times
smaller. In fact, this figure indicates also the x-broadened
Gaussian wave packet eq(T) for the system in study. Thus,
it determines the initial condition (t 0 ) eq(T) to Eq. 35
for the reduced density operator evolution.

J. Chem. Phys., Vol. 116, No. 21, 1 June 2002

The dynamic behavior of (t) is studied at T0 K


with a transform-limited Gaussian-pulse driving, (t)
2
2
Fe t /(2 F) cos(Ft). The field parameters are chosen as
F0.1 for the systemfield coupling strength, F50/
for its temporal width, and F for resonance condition.
Demonstrated in Fig. 2 is the induced dynamics, (t)
(t)eq , for the two specified values of systembath coupling strength B . The competition between excitation and
dissipation is further depicted in Fig. 3 in terms of induced
s (t)Tr H s (t) . Under the excitation condienergy H
tion in study, the induced energy reaches over 15 quanta
when the systembath interaction is weak at B0.001, and
only about 2 when B0.1. Note that the local-field correction (t) cf. Eqs. 3335 effectively reduces the driving
field strength. However, in the present cases of study, their
effects are too small to be very meaningful. The dissipation
Rs -induced suppression of excitation is evident in both Fig.
2 and Fig. 3.
V. SUMMARY AND CONCLUDING REMARKS

In summary, we have developed the complete secondorder QDT CS-QDT in which the initial correlated systemplus-bath canonical distribution is used explicitly in construction. Not only are the time-local CS-QDT Eq. 12 and
its time-ordered memory counterpart Eq. A13 presented,
but also a partially ordered memory-like CS-QDT in terms of
a set of coupled equations of motion EOM Eq. 22. The
latter can be considered as a direct variation of the time-local
CS-QDT.
The CS-QDT-EOM, either of the time-local prescription
Eq. 17 with 19 or its variation Eq. 22, is constructed
via the extended MeierTannor method29 Appendix B.
With the field-free Rs contribution separated out, the auxilab
ab
m
(t) or K m
(t) in the CS-QDT-EOM are
iary operators K
induced only for the effects of direct driving-dissipation correlation. As a result, the present CS-QDT-EOM can be rather
easily implemented in arbitrary temperatures, even including
T0 K if the effects of directly correlated driving dissipation are modest. Furthermore, according to a recent work by
Royer,34 the CS-QDT-EOM also tends to be completely positive in a time-coarse-grained scheme, as long as the consistent eq(T) Eq. 24 is used for the initial state before external driving.41 In fact, for the testing systems we have tried
so far, (t) is always positively defined. Moreover, the timelocal CS-QDT Eq. 12 or its variation Eq. 22 constitutes a convenient starting point to construct a dynamical
mean-field formulation for dissipative many-particle boson
or fermion systems.42
The time-local CS-QDT Eq. 12 may enjoy some advantage for its being simple but exact for the DBO system
Sec. IV A. In this case, the correlated driving-dissipation
contribution can be carried out analytically, leading to an
interesting picture of an effective local-field correction. The
resulting time-local CS-QDT formulation Eq. 35 can thus
be easily implemented at arbitrary temperatures including
0 K cf. the figures in Sec. IV B. The physical implication of
this result may be significant as the correlated driving dissipation in anharmonic systems may also be well approxi-

Theory of open quantum systems

9201

mated in terms of effective local-field corrections. Further


study along this line will be conducted in the future.
Due to its rigorous construction, the present CS-QDT
formulation has the corrected eq as its asymptotic solution
cf. Sec. III and Appendix C. The effective Hamiltonian
H eff , defined via eq(T)e H eff, is in general temperature
dependent Appendix D. The resulting H eff(T) contains second order exactly, but also higher orders in partial resummation. In general, H eff(T)H eff(T)H s cf. Eq. 27 depends on both bath spectrum and dispersion functions.
(2)
(T) is completely deterHowever, its second-order H eff
mined by bath dispersion functions cf. Eq. D7. The commonly used CaldeiraLeggetts renormalization Hamiltonian
H CL is identified as the high-temperature or Markovian
(2)
(T) cf. Eq. D16, but, H CL can equal
limit of H eff
H eff(T), which also includes a partial sum for higher-order
contributions, at a certain finite temperature.
To conclude this paper, we would like to make some
comments on the widely used QME quantum master equations, such as the BlochRedfield theory1018 and a class of
FokkerPlanck equations.1921 A unified QME assumes the
following algebraic form:3133

i H eff t ,

eff
Q
a

a Q a ,Q effa H.c. ,

b C ab Leff Q b .

37a

37b

Here, is treated as an adjusting or empirical parameter that


will be explained shortly. QME satisfies (t 0 ) ()
e H eff as its detailed-balance relation. Therefore, H eff is
used here to allow QME to share the commonly renormalized field-free stationary state as in CS-QDT Eq. 12.
However, in comparison with CS-QDT Eq. 14, QME invokes the so-called Redfield approximation3133 that apparently does not take the dispersion-induced dissipation into
account Eq. 37b. In order to make a quantitative comparison meaningful, the parameter in Eq. 37b is adjusted to
minimize the difference in an experimentally measurable
eff , between CS-QDT and QME.
quantity, chosen here as H
Depicted in Fig. 4 is a representative comparison between
the QMEs results thin-solid and those of CS-QDT thickdash. The testing model is the same as Fig. 2 and Fig. 3 for
B0.1, but at the specified temperature at which the
CaldeiraLeggett Hamiltonian H CL cf. Eq. D15 is coincident with H eff . The CS-QDT calculation in Fig. 4 is made
by setting (t)0 in Eq. 35, so that the direct dispersioninduced dissipation effects are isolated out. With the aforementioned adjusting parameter in Eq. 37 found to be
about 1/3, the QME, as its approximations involved are concerned, agrees fairly well with CS-QDT in the present example system. This may be largely accounted for by the fact
eff
that H eff depends on bath dispersion. As a result, Q
a Eq.
37b, which depends on Leff H eff , accounts for the
bath dispersion contribution to a certain degree.

9202

J. Chem. Phys., Vol. 116, No. 21, 1 June 2002

R. Xu and Y. J. Yan

W T / e H 0 /2H e H 0 /2W TH.c. /2


e L0 /2H W TH.c. /2
H W T .

A5

The initial condition to the above equation is that W T(


0)I , the unit operator. The superoperator H ( ) defined
in the last identity of Eq. A5 maps a Hermitian operator to
another Hermitian operator. The formal solution to Eq. A5
is given by

W Texp

FIG. 4. Comparison between QME Eq. 37 and CS-QDT Eq. 12 for


the same DBO oscillator as Figs. 2 and 3 with B0.1, but at a finite
temperature as described in the text. The local-field correction (t) in CSQDT is set to be zero such that the difference is purely due to the apparent
eff of QME cf. Eq. 37b.
neglect of bath dispersion from Q

Support from the Research Grants Council of the Hong


Kong Government and the National Science Foundation of
China is gratefully acknowledged.

APPENDIX A: DERIVATION OF EQ. 12


AND ITS MEMORY-KERNEL COUNTERPART

Denote the uncorrelated system-plus-bath Hamiltonian


in Eq. 1 as H(t)h BH t , and its corresponding Liouvillian as Lt H t , . Denote similarly L as that for systembath interaction H Eq. 1c. Define the interaction picture
via the Lt propagator, Gt (t,t 0 ), as

TI t G t t,t 0 T t ,

A1a

LI t G t t,t 0 L Gt t,t 0 .

A1b

By using the standard formulation in quantum mechanics,


the reduced density operator in the interaction picture,
I (t)trB TI (t), can be expressed as

I t trB exp i

t0

d LI T t 0 .

A2

The initial time t 0 shall be chosen as any moment before


external field actions. Physically, it implies that

T t 0 Teq T ;

as t 0 ,

A3

being the canonical density operator at temperature T in the


absence of external fields.
Conventionally, Teq is constructed via the imaginary time
propagation formalism. Here, we propose a rather simple algebraic approach based on a Hermitian-symmetrized, called
the canonical, interaction picture. Denote H 0 H s h B and

Teqe (H 0 H ) e H 0 /2W Te H 0 /2.


The above identity leads to

A4

d H I .

A6

Obviously, every expansion term of Teq defined via Eqs. A4


and A6 remains Hermitian.
By using a second-order cumulant expansion method exploited in our previous work30 but now for the combined
dynamics and initial condition, Eq. A2 can be approximated as

I t I

ACKNOWLEDGMENTS

t0

d RI t 0

exp

t0

d RI t 0 .

A7

The above equation is equivalent to I (t)RI (t) I (t).


Note that (t 0 )trB T(t 0 ) containing the static second-order
cumulant systembath correction.
The key quantity is the dissipation superoperator, RI (t),
which is formally given as
RI t

t0

LI t LI R t ,

A8a

LI t H .

A8b

R t i

The first term in Eq. A8a is the same as the conventional


second-order local-time dissipative superoperator,30 while
the second term R (t) Eq. A8b is additional. In the following, we shall show that this additional term vanishes as
the initial time t 0 .
By using the definitions of LI (t) and H ( ) Eq. A1b
and the last identity of Eq. A5, followed by taking the
form of H Eq. 1c, we can obtain the following identity:
R t A


i
2

ab

d C ab tt 0 i /2

G t,t 0 Q a G t,t 0 ,e H s /2Q b e H s /2A


H.c.

A9

ab ()0 and further, by the analytical continuNote that C

ation, C ab (tt 0 i /2)0 for t 0 . We have, therefore,


R t 0;

as t 0 .

A10

J. Chem. Phys., Vol. 116, No. 21, 1 June 2002

Theory of open quantum systems

Thus, Eq. A8a is the same as Eq. 2.22 of Ref. 30, assuming the initial time there changed to t 0

I t

d LI t LI I t .

A11

Royer34 has pointed out recently that Eq. A11 tends to


Lindblad form24 in time-coarse-grained limit. Following the
same procedure in deriving Eq. 2.31 of Ref. 30, we can
a (t) given by the first identity
finally obtain Eq. 12 with Q
of Eq. 13. The equivalence between the two expressions in
Eq. 13 can be verified via the following identity:
G t, Gs t, i

d G t, Lsf Gs , ,
A12

(t).
where Lsf( ) H sf(t), with H sf(t)
To end this Appendix, let us also present a corresponding
memory-kernel CS-QDT in the chronological-ordering prescription COP cf. Eq. 2.30 of Ref. 30

t iL t

cop

cop
Q
a t

cop

cop
Q a ,Q
a t H.c. ,
A13a

ab t G t, Q b cop .
d C
A13b

Note that the time-local-QDT Eq. 12 and memory-kernelQDT Eq. A13 are identical up to second order, regardless
of the nature of dissipation. They differ at the resummation
schemes that account for higher orders corrections. In Sec.
II C, we present an alternative, non-COP, memory-kernel
CS-QDT Eq. 22 that is identical to Eqs. 12 and A13
up to second order.

B1

It is easy to show that in the commonly used independent


harmonic bath model, J aa ( ) defined in above is coincident
with its conventional definition becoming temperature independent. The symmetry relations implied in Eqs. B1 and
8 lead to

* .
J ab J ba J ba

D ab

1
2

D ba

1
2

B2

We can now recast Eq. 6 in terms of the generalized spectral density functions as cf. Eq. 6a

B3a

J ab

d 1coth /2

d 1coth /2

,
B3b

J ab

B3c

In particular, D ab ( )D ba ( ) will not depend on temperature if J ab ( ) does not.


The MT method29 was originally constructed for fitting a
conventional spectral density J aa ( ) that is a real and odd
function of cf. Eq. B2. In this paper, the MT method is
generalized for J ab ( ), which satisfies the symmetry relation
of Eq. B2, in terms of the following expression:
J ab

ab
ab
k i k

k 2 ab i ab 2 2 ,
k

B4

ab
with all fitting parameters being real ( ab
k and k positive
as well and symmetric about the ab swap, except for
ab
ba being antisymmetric. Thus, aa
k k
k 0 and Eq.
B4 with ab amounts to the original MT expression.29
The bath interaction correlation function can now be obtained via the contour integration algorithm as denoting
ab
ab
z ab
k k i k )

ab t0
C

e i t J ab
1e

ab ab
ab
k i k /z k

k 4 ab ab e z
k

The original MeierTannor MT method29 is to parametrize the conventional spectral density function via which
the autocorrelation function C aa (t) can be expressed in the
form of Eq. 16 as a sum of exponential functions. In this
Appendix, we shall extend the MT method to the crosscorrelation function C ab (t) Eq. 16. The EOM-form of
QDT can then be obtained readily as indicated in Sec. II C.
To show the possibility of expressing C ab (t) in the form
of Eq. 16, let us consider the generalized bath interaction
spectral density functions
J ab C ab C ba .

C ab J ab / 1e e C ba ,

APPENDIX B: EXTENDED MEIERTANNOR


PARAMETERIZATION METHOD

9203

ab
k

ab

e iz k

ab ab
ab
k i k /z k
ab
4 ab
k k 1e

n1

ab

ab
zk

J ab n
e n t
n

e iz k

m mab e

ab
m t

B5
Here, n 2 n/ the Matsubara frequency, and J ab ( )
i J ab (i ) is a real function cf. Eq. B4.
The last identity in Eq. B5 is identical to Eq. 16. It
highlights the fact that the extended MT method Eq. B4
ab (t) in terms of a sum of N expoamounts to expressing C
nential functions, where N equals twice the number of MT
terms plus the number of Matsubara terms needed. The main
disadvantage of the MT method is that the number (N) of
EOM can become huge at low temperature regime due to the
high density of the Matsubara frequency n 2 n/ . In
Sec. II C, the MT method is applied only to treat the correlated driving-dissipation contribution, without altering the
equilibrium or stationary property of CS-QDT.

9204

J. Chem. Phys., Vol. 116, No. 21, 1 June 2002

R. Xu and Y. J. Yan

APPENDIX C: EQUIVALENCE OF SECOND-ORDER


REDUCED CANONICAL DENSITY OPERATOR
IN TWO APPROACHES
(2)
The second-order correction eq
from the stationary
solution of CS-QDT Eq. 12 or Eq. 22 is given by Eq.
27 with removing Rs from s iLs Rs there. That is,
(2)

Ls eq

a Q a , a e H e H a .
s

C1

(2)
An alternative approach to eq
is to start with the secondeq
order expansion of eqtrB T (T); i.e., Eqs. A4A6. The
result there can be recast in contact with Eq. C1 as
(2)
e H s /2 Ls W T(2) e H s /2,
Ls eq

C2

with
W T(2)

d H H I .

C3

Here, is defined in Eq. 2 and H ( ) is in the last


identity of Eq. A5. We shall show that Eq. C2 is equivalent to Eq. C1.
To simplify the notation, let us denote (L0 Ls LB)
V e

L0 /2

H L0 V .

C4

It immediately follows that


V L 1
0 V iP
V e

L0 /2

dt e iL0 t V ,

C5a

Therefore cf. Eqs. C4 and C5b,


1
V V L 1
0 V L 0 V

L 1
0 V V H.c.,

and Eq. C9 becomes

U L 1
0 V V V V H.c.,

H L 0 V ,

which leads to U(0)0. By using Eqs. C8 and C11,


together with the identity L0 W T(2) Ls W T(2) , we obtain
Ls W T(2) V V V V H.c. C12
To proceed, we shall use the explicit form of H Eq. 1c
with the Heisenberg picture, Q a (t)e iLs t Q a and F a (t)
e iLBt F a , for Eq. C4 and Eq. C5a; i.e.,
V

V V e L0 /2 V 0 V 0 ,

C5d

d
1
V
V V
V V .
V

2
d

C5e

The integrand in Eq. C3 can now be recast as cf. Eq. A5


H H I 41 V V V H.c.
C6

V iP

C13a

dt e iLs t Q a i /2 F a ti /2 .
C13b

We have then

V V iP
a,b

dt Q a i /2

F a i /2 F b ti /2
ie Ls /2P

a,b

ab t .
dt Q a e iLs t Q b C
C14

By using the first identity of Eq. 6b and the definition of a


cf. Eq. 14, the above equation can be recast as

V V e Ls /2 Q a D ab Ls Q b
a,b

e H s /2

Substituting it into Eq. C3 and then performing the


integration, we obtain cf. Eq. C5c

a Q a i /2 F a i /2 ,

C5b
C5c

C11

e iLs t Q b i /2

/d V /2,
dV

W T(2) 21

C10

a Q a a e H /2.
s

C15

Similarly, we have
H.c.
d V V V

C7

Now, by using Eqs. C5d and C5e, the above equation can
be formally integrated, resulting in
W T(2) U U 0 ,

a,b

C8

a e H /2 a e H Q a e H /2.
s

C16

Substituting Eqs. C15 and C16 into Eq. C12, we obtain

with

U L 1
0 V V H.c. V V .

V V D ab Ls Q b i /2 Q a i /2

C9

Lemma: If C C, as the commutator action of C, then


C (C 1 A)(C 1 B) A(C 1 B)(C 1 A)B.
B )(CA
)B
A
(CB
), then
This can be proved via C(A
1
1

setting A C A and B C B.

e H s /2 L s W T(2) e H s /2

a Q a , a e H e H a .
s

It thus proves that Eq. C2 is equivalent to Eq. C1.

C17

J. Chem. Phys., Vol. 116, No. 21, 1 June 2002

Theory of open quantum systems

APPENDIX D: EFFECTIVE HAMILTONIAN VERSUS


CALDEIRALEGGETTS RENORMALIZATION
HAMILTONIAN

Having knowledge of eq(T) cf. Sec. III, we may


evaluate the associating effective Hamiltonian such that

eq T e H eff/Tr e H eff,

(2)
H eff
D 0 Q 2 O .

(2)
H eff
D 0 Q 2 ,

denote that H effH s H eff , and


D2

if D D 0 .

H TCLH CL

D3

1
2

We have then
W I
I

sinh Ls /2
H eff .
Ls /2

eqe

H s /2

sinh Ls /2
H eff e H s /2.
Ls /2

D7

D 21 e /2D e /2D

D 0

1
2

sinh /2 J
sinh /2

. D8

Note that

d J /.

D9

In the following, we shall show that in the high(2)


Eq. D7 reduces to
temperature or Markovian limit, H eff
the well-known CaldeiraLeggetts renormalization correction that will be specified soon.
Let us start with the high-temperature limit. We have
sinh /2
sinh /2

m j j

.
D13

D14

O 2 .

c 2j

j 2m 2 Q 2 H CLD 0 Q 2 .

D10

It leads Eq. D7 to have the following high temperature


value:

D15

This is to say that the CaldeiraLeggetts renormalization


correction is given by H CLH CLH s D(0)Q 2 . We have
thus shown that cf. Eqs. D11 and D12
(2)
H eff
H CL ;

D6

with cf. Eqs. B3b and B3c

Q
2

c 2j

1
e Ls /2Q D Ls Q H.c.,
sinh Ls /2

1
2

cj

j m j j j .

H s H CL

D5

Ls /2
W T(2) .
sinh Ls /2

mj

m j 2j x j

Comparing to Eq. 1, we have cf. Eq. D9

Here, W T(2) was defined in Appendix C cf. Eq. C17.


In the case of the single dissipation mode in which H
QF, Eq. D6 assumes
(2)
H eff

J
2

D4

The second term in the above square bracket is


(2) H s /2
e
cf. Eq. 25. In comparison with Eq.
e H s /2 eq
C2, we have therefore
(2)
H eff

p 2j

For this model, the spectral density is

d cosh Ls /2 H eff

We have cf. Eq. D2

D12

Let us now turn to the widely used CaldeiraLeggetts


system-plus-bath Hamiltonian21

which satisfies W(0)I and

W/ e Ls /2 H eff WH.c. /2.

D11

In the Markovian limit, which assumes a broad bath interaction spectrum, D( )D(0). In this case, D ( )
cosh(/2)D(0). Substituting this into Eq. D7, we can
obtain

D1

W e H s /2e H effe H s /2,

9205

in high-T or Markovian limits.


D16

S. Nakajima, Prog. Theor. Phys. 20, 948 1958.


R. W. Zwanzig, J. Chem. Phys. 33, 423 1960.
3
R. W. Zwanzig, Statistical Mechanics of Irreversibility: Lectures in Theoretical Physics Wiley, New York, 1961, Vol. III.
4
R. W. Zwanzig, Annu. Rev. Phys. Chem. 16, 67 1965.
5
H. Mori, Prog. Theor. Phys. 33, 423 1965.
6
U. Fano, Phys. Rev. 131, 259 1963.
7
B. J. Berne, in Physical Chemistry, An Advanced Treatise, edited by H.
Eyring, W. Jost, and D. Henderson Academic, New York, 1971, Vol. 8B,
p. 539.
8
K. Blum, Density Matrix Theory and Applications Plenum, New York,
1981.
9
S. Mukamel, Adv. Chem. Phys. 47, 509 1981.
10
R. K. Wangsness and F. Bloch, Phys. Rev. 89, 728 1953.
11
F. Bloch, Phys. Rev. 105, 1206 1957.
12
A. G. Redfieldl, Adv. Magn. Reson. 1, 1 1965.
13
W. H. Louisell, Quantum Statistical Properties of Radiation Wiley, New
York, 1973.
14
Quantum Statistics in Optics and Solid State Physics. Statistical Treatment
of Open Systems by Generalized Master Equations, edited by F. Haake,
Springer Tracts in Modern Physics, Vol. 66 Springer, Berlin, 1973.
15
H. Haken, Laser Theory Springer, Berlin, 1970.
16
M. Sargent, III, M. O. Scully, and J. W. E. Lamb, Laser Physics AddisonWesley, Reading, MA, 1974.
17
W. T. Pollard, A. K. Felts, and R. A. Friesner, Adv. Chem. Phys. 93, 77
1996.
18
D. Kohen, C. C. Marson, and D. J. Tannor, J. Chem. Phys. 107, 5236
1997.
19
H. Dekker, Phys. Rep. 80, 1 1981.
20
A. O. Caldeira and A. J. Leggett, Physica A 121, 587 1983.
21
A. O. Caldeira and A. J. Leggett, Ann. Phys. San Diego 149, 374 1983.
22
G. Lindblad, Chem. Phys. 48, 119 1976.
23
V. Gorini, A. Kossakowski, and E. C. G. Sudarshan, J. Math. Phys. 17,
821 1976.
1
2

9206
24

J. Chem. Phys., Vol. 116, No. 21, 1 June 2002

R. Alicki and K. Lendi, Quantum Dynamical Semigroups and Applications


Springer, New York, 1987, Lecture Notes in Physics 286.
25
Y. Tanimura and R. Kubo, J. Phys. Soc. Jpn. 58, 101 1989.
26
Y. Tanimura and P. G. Wolynes, Phys. Rev. A 43, 4131 1991.
27
Y. Tanimura and P. G. Wolynes, J. Chem. Phys. 96, 8485 1992.
28
J. Cao, J. Chem. Phys. 107, 3204 1997.
29
C. Meier and D. J. Tannor, J. Chem. Phys. 111, 3365 1999.
30
Y. J. Yan, Phys. Rev. A 58, 2721 1998.
31
Y. J. Yan, F. Shuang, R. X. Xu, J. X. Cheng, X. Q. Li, C. Yang, and H. Y.
Zhang, J. Chem. Phys. 113, 2068 2000.
32
R. X. Xu, Y. J. Yan, and X.-Q. Li, Phys. Rev. A 65, 023807 2002.
33
F. Shuang, C. Yang, and Y. J. Yan, J. Chem. Phys. 114, 3868 2001.
34
A. Royer, Phys. Rev. Lett. 77, 3272 1996.

R. Xu and Y. J. Yan
35

U. Weiss, Quantum Dissipative Systems, 2nd ed. Series in Modern Condensed Matter Physics, Vol. 10 World Scientific, Singapore, 1999.
36
D. Forster, Hydrodynamics Fluctuations, Broken Symmetry, and Correlation Functions Benjamin, London, 1975.
37
R. Kubo, M. Toda, and N. Hashitsume, Statistical Physics II: Nonequilibrium Statistical Mechanics, 2nd ed. Springer, Berlin, 1985, p. 88.
38
N. G. van Kampen, Stochastic Processes in Physics and Chemistry
North-Holland, Amsterdam, 1992, Chap 17.6.
39
W. E. Arnoldi, Q. Appl. Math. 9, 17 1951.
40
T. J. Park and J. C. Light, J. Chem. Phys. 85, 5870 1986.
41
A. Suarez, R. Silbey, and I. Oppenheim, J. Chem. Phys. 97, 5101 1992.
42
S. Yokojima, G. H. Chen, R. X. Xu, and Y. J. Yan, J. Chem. Phys. submitted.

Você também pode gostar