Home

Search

Collections

Journals

About

Contact us

My IOPscience

Finite quantum dissipation: the challenge of obtaining specific heat

This content has been downloaded from IOPscience. Please scroll down to see the full text.
2008 New J. Phys. 10 115008
(http://iopscience.iop.org/1367-2630/10/11/115008)
View the table of contents for this issue, or go to the journal homepage for more

Download details:
IP Address: 14.139.204.210
This content was downloaded on 23/10/2016 at 16:57

Please note that terms and conditions apply.

You may also be interested in:

Anomalies in the specific heat of a free damped particle: the role of the cutoff in the spectral
density of the coupling
Benjamin Spreng, Gert-Ludwig Ingold and Ulrich Weiss
A semiclassical approach to explore the bistable kinetics of a Brownian particle in anonequilibrium
environment
Pradipta Ghosh, Anindita Shit, Sudip Chattopadhyay et al.
Reentrant classicality of a damped system
Benjamin Spreng, Gert-Ludwig Ingold and Ulrich Weiss
Entropy production in quantum Brownian motion
Lorenzo Pucci, Massimiliano Esposito and Luca Peliti
Entropy of a quantum oscillator in a structured environment
Jishad Kumar
Quantum equilibrium factor in the decay of a metastable state at low temperature
Debashis Barik and Deb Shankar Ray

New Journal of Physics

The openaccess journal for physics

Finite quantum dissipation: the challenge

of obtaining specific heat
Peter Hnggi1 , Gert-Ludwig Ingold and Peter Talkner
Institut fr Physik, Universitt Augsburg, D-86135 Augsburg, Germany
E-mail: [email protected]
New Journal of Physics 10 (2008) 115008 (17pp)

Received 26 May 2008

Published 20 November 2008
Online at http://www.njp.org/
doi:10.1088/1367-2630/10/11/115008

Abstract. We consider a free particle coupled with finite strength to a bath

and investigate the evaluation of its specific heat. A harmonic oscillator bath of
Drude type with cutoff frequency D is employed to model an ohmic friction
force with dissipation strength . Two scenarios for obtaining specific heat
are presented. The first one uses the measurement of the kinetic energy of the
free particle, while the second one is based on the reduced partition function.
Both descriptions yield results which are consistent with the Third Law of
thermodynamics. Nevertheless, the two methods produce different results that
disagree even in their leading quantum corrections at high temperatures. We
also consider the regime where the cutoff frequency is smaller than the friction
strength, i.e. D < . There, we encounter puzzling results at low temperatures
where the specific heat based on the thermodynamic prescription becomes
negative. This anomaly is rooted in an ill-defined density of states of the damped
free particle, which assumes unphysical negative values when /D > 1.

Author to whom any correspondence should be addressed.

New Journal of Physics 10 (2008) 115008

1367-2630/08/115008+17$30.00

IOP Publishing Ltd and Deutsche Physikalische Gesellschaft

2
Contents

1. Introduction
2. Coupling a free particle to a heat bath
2.1. Dissipation and the corresponding quantum Langevin equation . . . . . . . . .
2.2. Thermodynamics of a dissipative free particle . . . . . . . . . . . . . . . . . .
3. Free quantum Brownian motion: two routes for calculating specific heat
4. Follow the route using energy
5. Follow the route using the thermodynamic partition function
6. Partition function and density of states
7. Conclusions
Acknowledgments
References

2
4
4
6
7
8
10
13
15
16
16

1. Introduction

While some physical disciplines such as classical mechanics and electrodynamics underwent
profound changes with the birth of quantum mechanics and relativity, thermodynamics
proved impressively robust over the last century. The main reason is that the formulation of
thermodynamics rests on a few pillars only, such as entropy, temperature and the three Laws
relating these state variables. The grandness of thermodynamics is that these concepts hold
independently of the details of the corresponding total system dynamics. Nonetheless, the
statistical mechanical foundation of thermodynamics strongly relies on the quantum mechanical
properties of matter, in particular what concerns the low-temperature behaviour.
Statistical mechanics gives rise to some subtle issues when going from a closed description
of all degrees of freedom, including those of large environments, to a reduced description of an
open system where all bath degrees of freedom are traced over. Generally, no problems arise in
the weak-coupling limit where the systembath interaction tends to zero. This is the situation
typically assumed explicitly, or at least implicitly, in the majority of textbooks. Even in this
limit, however, pitfalls can arise in the quantum case, as recently elucidated in [1]. Generally,
care must also be taken in defining correctly the expression for (Gibbs) work in the First Law,
as recent debates on the validity of nonlinear fluctuation theorems have shown [2, 3].
In this work, we emphasize the finiteness of the coupling between the open quantum
system and an environment of temperature kB T = 1/. The finite coupling will be modelled in
terms of a generalized quantum Langevin equation (GLE) with a velocity-proportional memory
friction, i.e. ohmic damping. An equivalent microscopic approach is provided by a bilinear
coordinatecoordinate coupling between system degrees of freedom and environmental degrees
of freedom, see e.g. in [1] and references cited therein.
The ohmic friction kernel involves two parameters that characterize the role of the
environment. These are the frictional coupling strength and the (Drude)-frequency D of the
largest bath frequency. In typical situations, this Drude frequency markedly exceeds the friction
strength; i.e. D  . Quantum dissipation with an unconventional environment is obtained,
however, when D < . This situation arises, for example, for suitably tailored thermal baths or
for molecular electron transfer systems which are exposed to polar solvents [4].
New Journal of Physics 10 (2008) 115008 (http://www.njp.org/)

3
The presence of an ohmic friction term leads to an irreversible motion with a unique
stationary state. Already at this level, prominent differences with the classical situation emerge.
As is well known, the classical canonical thermal equilibrium probability assumes the familiar
GibbsBoltzmann expression exp( HS ), with HS being the bare system Hamiltonian in
the absence of interaction. Most remarkably, this classical result holds true independently of
friction strength. In striking contrast, the equilibrium density operator of an open quantum
system becomes a function of the friction strength [1, 5], thus exhibiting a dependence on the
coupling to the environmental degrees of freedom. This feature can explicitly be inspected for
an ohmic-like, damped harmonic quantum oscillator [1, 6, 7]: only in the weak-coupling limit
does the canonical density operator reduce to the common Gibbs state, where HS then represents
the operator-valued system Hamiltonian.
Given this observation in the preceding paragraph it should not come as too big a
surprise that the evaluation of thermodynamic quantities for an open quantum system, such
as its specific heat, is also plagued with subtleties. These difficulties all originate from the
assumed finite systembath interaction. For the case of specific heat of a linear quantum
oscillator of finite friction strength, this was explicitly demonstrated in a recent work [8]. While
in the absence of damping, zero entropy is approached exponentially fast, this approach is
weakened with finite friction to a power law dependence in friction strength and temperature
[8, 9]. The Third Law has also been validated for the quantum dissipative oscillator in the
presence of velocitycoordinate and velocityvelocity couplings [10] as well as for a charged
oscillator in a magnetic field [11]. The main finding is that the thermodynamic entropy of
the open system vanishes according to a power law in temperature with the same exponent
that characterizes the frequency dependence of the memory friction in the limit of vanishing
frequency.
It was remarked in [8], however, that the definition of the specific heat is ambiguous and
may lead to pronouncedly different values. Therefore, the relation to experimental observations
needs scrutiny, in particular in the study of nanosystems whose behaviour is quantum in nature
but that are still coupled with finite strength to an environment. Another surprising observation
is that the familiar von-Neumann entropy for a quantum dissipative oscillator fails to approach
zero for vanishing bath temperature [12].
In the following, we elucidate in detail the complications that arise when evaluating
thermodynamic quantities of open quantum systems whose dissipation strength is finite. A most
suitable test bed is the case of a free quantum particle [5, 8]. It is our working hypothesis that
this simple system better be understood first before daring to embark on more complex physical
situations. As is well known, a classical free particle does not obey the Third Law. Surprisingly,
the coupling to a heat bath renders the system more quantum thereby helping to restore the
Third Law [8].
The work is organized as follows. In the next section, we couple a free quantum particle to
a heat bath which exerts a finite dissipation on the free particle. We then focus on the evaluation
of thermal equilibrium quantities, in particular the specific heat. In section 3, we introduce
two definitions of the specific heat which a priori both seem physically well motivated. Their
properties are investigated in sections 4 and 5. The emerging results cause worrisome baffling,
which we attempt to resolve by inspecting more closely the underlying density of states of a
dissipative free particle. Some conclusions and consequences are given in our final section.

New Journal of Physics 10 (2008) 115008 (http://www.njp.org/)

HS

HSB

HB

T
Figure 1. The setup of an open nanosystem with Hamiltonian HS which is

coupled with finite strength through an interaction HSB to a bath described

by a Hamiltonian HB . The total system composed of system plus bath is weakly
coupled to a super-bath which provides the working temperature T in thermal
equilibrium.
2. Coupling a free particle to a heat bath

2.1. Dissipation and the corresponding quantum Langevin equation

As mentioned in the introduction, a free particle, or more generally, an ensemble of noninteracting, distinguishable particles fails to approach zero entropy at accessible temperatures.
This classical behaviour is modified when the particle is able to exchange energy with a heat
bath. We therefore focus on the simplest situation of a free quantum particle of mass M that is
coupled to a heat bath made up of harmonic oscillators. The total Hamiltonian of particle plus
bath thus reads (see figure 1)
H = HS + HB + HSB ,

(1)

where
p2
2M
describes the free particle with momentum operator p, and

N 
X
pi2 m i 2 2
HB =
+ x
2m i 2 i i
i=1
HS =

(2)

(3)

represents a set of harmonic oscillators constituting the heat bath which is bilinearly coupled to
the free particle via its position operator q by
HSB =

N
X
m i 2
i

i=1

2q xi + q 2 .

(4)

Here, the coupling constants have been chosen without loss of generality in such a way that the
ensemble of free particles and heat bath is translationally invariant so that the damped particle
can still be considered as free [5].
Elimination of the heat bath leads to an effective equation of motion for the position
operator of the damped free particle, the quantum Langevin equation of a free particle, reading
Z t
d
d2
M 2q + M
ds (t s) q = (t).
(5)
dt
ds
t0
New Journal of Physics 10 (2008) 115008 (http://www.njp.org/)

5
The coupling to the heat bath results in a damping kernel
N
1 X
m i i2 cos(i t)
(t) =
M i=1

(6)

and an operator-valued noise term

(t) = M (t t0 )q(t0 ) +

N
X



m i i2 xi (t0 ) cos (i (t t0 )) + i pi (t0 ) sin(i (t t0 )) ,

(7)

i=1

which depends on the initial conditions of free particle and heat bath. The sum of the second
and third contributions of this expression, i.e.
(t) = (t) + M (t t0 )q(t0 ),

(8)

only depends on the initial positions and momenta of the heat bath and is characterized by a
vanishing first moment,
h(t)iB = 0,

(9)

while the symmetrized correlation function obeys



N
1
h X
h i
3
h(t)(s) + (s)(t)iB =
m i i cos (i (t s)) coth
.
2
2 i=1
2kB T

(10)

Here, h...i B denotes an expectation value with respect to the equilibrium density matrix
exp( HB )/Tr[exp( HB )] of the isolated bath. The temperature T is imposed on the heat
bath via the weak coupling to a super-bath (see figure 1). The commutator of the noise is
non-vanishing, reading
[(t), (s)] = ih

N
X

m i i3 sin (i (t s)),

(11)

i=1

and guarantees that the familiar commutator-relation between position q and momentum p is
obeyed at all times, as required by quantum mechanics.
In the following, we will restrict ourselves to the so-called Drude model of quantum
dissipation where the damping kernel describes exponential memory on the timescale D1 .
For positive arguments t > 0, the damping kernel assumes the form:
(t) = D exp(D t),

(12)

while for negative times we formally obtain (t < 0) = (|t|). This in turn assures that the
symmetrized correlation function of the time-homogeneous noise correlation in (10) obeys a
generalized Einstein relation [1]. The value yields the damping strength according to
Z
dt (t) = .
(13)
0

For later use, we will also need the Laplace transform of the damping kernel, which is given by
(z) =

Z
0

dt (t) exp(zt) =

D
.
z + D

New Journal of Physics 10 (2008) 115008 (http://www.njp.org/)

(14)

6
The total mass of the heat bath can be expressed via the appealing formula [5]
N
X
i=1

(z)
.
z0 z

m i = M lim

(15)

As a consequence of the fact that the Drude model behaves ohmically for low frequencies, i.e.
(z) goes to a positive constant for z 0, this heat bath behaves non-ballistically in the sense
that it assumes an infinite mass.
The Drude model is typically employed for regularization purposes. As an example,
we mention that in the case of strict ohmic dissipation, i.e. memoryless damping where
(t) = 2 (t), the second moment h p 2 i of the momentum p with respect to the equilibrium
density matrix of system plus bath exhibits an ultraviolet divergence. For the purposes of the
regularization of such a quantity, it is assumed that D1 is by far the shortest timescale in the
problem apart possibly from the thermal timescale h . For a free particle subject to Drude
damping, one therefore typically assumes  D . In this paper, we drop the requirement
of a high-frequency cutoff, which in turn enables us to study also structured, uncommon
environments that may contain a low-frequency cutoff D , even smaller than the damping
strength .
The Drude model represents the simplest reservoir with memory in the sense that the
memory effect can be described by a single additional degree of freedom. In fact, the
deterministic equation of motion (5), i.e. where (t) is set to zero, is equivalent to
q = v,
v = z,
z = D z D v.

(16)

This system of differential equations contains a zero-frequency mode because system plus bath
is translationally invariant. The remaining two eigenfrequencies are obtained from the second
and third equations which describe a damped harmonic oscillator with damping strength D
and oscillator frequency ( D )1/2 . For D > 4 , one finds exponentially damped motion which
for very large cutoff frequencies contains the timescales and D as expected. However, for
sufficiently small cutoff frequency, where D < 4 , the eigenvalues become complex and one
observes a damped oscillation. This makes the Drude model more interesting than one might
initially expect.
2.2. Thermodynamics of a dissipative free particle
We note that a thermodynamic description of a free particle can be meaningful only if the
particle is confined to stay within a box of finite size. The presence of the box leads to the
quantization of the energy levels with the typical excitation energy 1E = h 2 2 /2M L 2 between
the first excited state and the ground state for a one-dimensional box of length L with reflecting
walls. This system in isolation (i.e. in the absence of dissipation) approaches zero entropy only
at extremely low temperatures when the thermal energy is comparable with or below the energy
difference between the first excited and the ground state.
Remarkably, although the case of a free particle is of comparably simple structure as
that of a harmonic oscillator, the thermodynamic properties of the free particle are more
intriguing and cannot be obtained as the limiting case of a harmonic oscillator with a vanishing
oscillation frequency, 0 [8]. This limiting procedure would yield a free particle dynamics
New Journal of Physics 10 (2008) 115008 (http://www.njp.org/)

7
without a spatially confining box. Such a confinement is intrinsically essential, however, for the
existence of a thermodynamic description in terms of a partition function and its corresponding
thermodynamic functions, see below.
We shall restrict our investigations to temperatures which are high enough such that
the discreteness of the spectrum of a single free particle can be ignored even in the weakcoupling limit, i.e. we will always assume that 1E  1. For a helium atom at a temperature
of approximately 500 pK, which is the lowest temperature that can be reached today in
the laboratory, the thermal energy and the excitation energy are equal for a cavity of linear
size of the order of 30 m. Even for such extremely low temperatures, a handy box of linear
size of, say, 1 cm would suffice to meet the required condition with 1E 105 . In order
that any quantum effects can survive under thermal conditions rendering the discreteness of
the particle spectrum practically invisible, the bath must provide relevant energy scales that
are large compared with the thermal energy. In the case of a bath with a Drude cutoff, these
relevant energies are given by the damping constant and the cutoff frequency. Hence, a regime
of low temperatures with 1E  kB T  h , h D exists where quantum effects can be expected
to become relevant.
3. Free quantum Brownian motion: two routes for calculating specific heat

We are interested in the specific heat of a free damped particle. The volume in which the particle
can move, will be assumed very large but fixed. In this case, the specific heat is obtained from
the internal energy U by taking the derivative with respect to the temperature T , i.e.
U
C=
.
(17)
T
Usually, it is supposed that the coupling of the system to the heat bath defining the
temperature can be treated in the limit of vanishing coupling strength. However, here we are
interested in the case of finite coupling where the meaning of the systems internal energy is no
longer obvious. In the following, we will study two different approaches.
One possibility is to replace U by the energy E defined as expectation value of the system
Hamiltonian HS
E = hHS i,
(18)
where
TrS+B [HS exp( H )]
hHS i =
,
(19)
TrS+B [exp( H )]
with = 1/kB T being the inverse temperature. This leads to our first definition of a specific
heat
E
CE =
.
(20)
T
This definition is based on the system Hamiltonian HS and includes the interaction of the system
with the bath only via the density matrix of the total system.
Alternatively, one can start from the well-known and widely used expression for the
partition function of the reduced system: It is defined in terms of the partition functions of
the coupled system and of the uncoupled bath as [5, 6, 8, 9], [13][20]
TrS+B [exp( H )]
Z=
,
(21)
TrB [exp( HB )]
New Journal of Physics 10 (2008) 115008 (http://www.njp.org/)

8
where the total Hamiltonian H consists of contributions from the system, the bath and the
coupling according to (1). Employing the standard relation between partition function and
internal energy

U = ln(Z ),
(22)

equation (21) implies that the internal energy is defined as

U = hH i hHB iB
= E + [hHSB i + hHB i hHB iB ].

(23)

While the internal energy U and the system energy E agree in the absence of a coupling between
system and bath, this is no longer the case at finite coupling. Particularly disturbing is the
observation that this difference is not solely given by the expectation value of the interaction
Hamiltonian HSB , but still contains the difference of the bath energies caused by the interaction
with the system.
We can now proceed and define a second specific heat, reading
U
CZ =
.
(24)
T
The partition function (21) also allows us to define an entropy



ln Z .
S = kB ln(Z )
(25)

This results from the thermodynamic relation

F
S=
(26)
T
with the free energy F = (1/)ln(Z ).
Alternatively, the second specific heat (24) can as well be derived from the entropy (25);
i.e. C Z is equivalently obtained as
S
CZ = T
(27)
.
T
Our main concern is that these two routes, (20) and (24), of obtaining a specific heat
generally yield different results. This will be elucidated further by studying the dissipation
model of the Drude form when applied to the simple quantum dynamics of a free particle.
4. Follow the route using energy

We start our exploration of the specific heat by employing the definition (20) based on the
expectation value of the system energy
1
E=
h p 2 i.
(28)
2M
Here, M denotes the mass of the particle and p the momentum operator. For a general heat bath
leading to an equation of motion of the form (5) and for thermal energies 1 that are large
compared with the energy scale h 2 2 /2M L 2 , one finds [5]
"
#

X
1
(n )
E=
1+2
(29)
2
+ (n )
n=1 n
New Journal of Physics 10 (2008) 115008 (http://www.njp.org/)

9
0.5

C E /kB

0.4
0.3
0.2

D / =
5
1
0.2

0.1
0
0

0.2

0.4

0.6

0.8

kB T /
h

Figure 2. Energy route: the specific heat C E is depicted as a function of

dimensionless temperature for different ratios of D / . The approach towards

the classical limit C E = 0.5 at low temperature is inversely proportional to
friction strength ; it notably becomes enhanced for decreasing values of cutoff
frequencies D . The specific heat itself is always positive and monotonically
increases with increasing temperature T .
with the Matsubara frequencies n = 2 n/h . In the high-temperature limit, the sum vanishes
and one recovers E = kB T /2, in agreement with the classical equipartition theorem. For strictly
ohmic damping, (z) = , the sum diverges as mentioned above and one is obliged to introduce
a cutoff in the spectral density of the heat bath. From (29), we obtain by means of (20) the
specific heat

CE
1 X 2 (n ) + n2 0 (n )
= +
,
kB
2 n=1 (n + (n ))2

(30)

where the prime denotes a derivative with respect to the argument.

For strictly ohmic damping one has (z) = and therefore 0 (z) = 0. It is tempting to drop
the second term in the numerator of the sum and to evaluate the resulting sum which converges
even for strict ohmic damping. However, this would yield a specific heat which diverges as the
temperature approaches zero (see also the discussion after (35)). The reason for this unphysical
behaviour is that for constant (z) the sum in (29) does not converge and therefore derivatives
should not be taken term-by-term.
For the Drude model, one obtains with (14) for the specific heat C E :
 1
CE
x1 x2 
=
x2 0 (x2 ) x1 0 (x1 ) ,
kB
x1 x2
2
0
where (x) denotes the trigamma function and
s
!
h D
4
h 1,2
x1,2 =
1 1

.
4
D
2

(31)

(32)

The behaviour of this result is depicted in figure 2 for differently structured ohmic baths as
characterized by various ratios of cutoff D over damping strength .
New Journal of Physics 10 (2008) 115008 (http://www.njp.org/)

10
We next study some asymptotic limits more closely. For high temperatures T much larger
than h /kB and h D /kB , the specific heat approaches the expected classical result with the
leading quantum corrections reading
CE
1
h 2 D
=
+ O(T 3 ).
kB
2 24(kB T )2

(33)

At low temperatures, i.e. for kB T  h , h D but still kB T  h 2 2 /2M L 2 , the specific heat
goes to zero linearly according to

 

CE

kB T 4 3 kB T 3

12
+ O(T 5 ).
=
(34)
kB
3 h
15
h
D
This behaviour is in agreement with the Third Law: Finite quantum dissipation thus
restores the validity of the Third Law of a free particle already at temperatures for which
the discreteness of the particle spectrum is still irrelevant. The limiting behaviour is inversely
proportional to damping strength. Put differently, strong dissipation diminishes the prefactor
and thereby yields a turnover to the classical behaviour at higher temperatures only. The leading
low-temperature behaviour of the specific heat only depends on the low-frequency behaviour
of (z). Notably, only the next-to-leading order also depends on the cutoff frequency D .
In the limit of an infinite cutoff frequency D the complete temperature dependence
is obtained from (31) as

2 

h
CE
h
1
h
0

.
(35)
kB
2 kB T
2 kB T
2kB T 2
The second term would have been missed in a naive calculation where one simply sets 0 (z) = 0
and subsequently evaluates the sum in (30). This term is important because it ensures that the
specific heat does not diverge for T 0.
5. Follow the route using the thermodynamic partition function

In order to obtain the specific heat from the partition function (21), one must first determine the
latter quantity for the damped free particle.
We again assume that the volume available to the particle is sufficiently large so that the
discreteness of the energy levels can be neglected. In the limit of vanishing coupling to the heat
bath, the partition function is given by


L 2 m 1/2
Z0 =
.
(36)
h

This result is a consequence of the density of states (E) E 1/2 of a free particle in one
dimension.
In the presence of a finite coupling to the heat bath, the partition function is modified by
the ratio of the two fluctuation determinants for the free particle in the absence and presence of
the dissipative coupling [5, 13, 14, 17]. We then obtain for this so defined partition function


L 2 m 1/2 Y
n
Z=
,
(37)
h

(
)
n
n
n=1
New Journal of Physics 10 (2008) 115008 (http://www.njp.org/)

11
which will represent the starting point for all following calculations discussed in this section.
We recall that the modifications due to the environmental coupling do not take into account
the presence of confining walls. In using this expression for the partition function, we therefore
assume a sufficiently large confining box so that boundary effects, which could show up in the
partition function, can safely be neglected.
The specific heat can be calculated from the partition function either by means of the
internal energy (24) or the entropy (27). Both routes lead to identical results. In order to stay
close to the reasoning of the previous section, we choose the first alternative.
The internal energy of a damped free particle is obtained from (37) by means of (22),
reading
"
#

X
1
(n ) n 0 (n )
U=
1+2
.
(38)
2
n + (n )
n=1
This internal energy differs from the energy (29) by the second term of the numerator appearing
in the sum.
For the Drude model, the sum can be evaluated and the internal energy is obtained as


h D
h D
x1
x2
1
(x1 ) (x2 )
U=

,
(39)
2
2

2
where (x) is the digamma function and x1,2 are defined in (32). This result should be
contrasted with the energy E for which one finds from (29) for the Drude model
x1 x2
1
E=
.
(40)
((x1 ) (x2 ))
(x1 x2 )
2
It is interesting to note that the internal energy U and the energy E already differ in the
high-temperature limit, i.e. h D /2  1; consequently leading to differing high-temperature
quantum corrections to the specific heat as shown below. This disagreement even persists in
the case of strict ohmic damping. Note, however, that in this limit both quantities diverge
logarithmically in the cutoff frequency D , so that E and U will contain an infinite energy
contribution independent of temperature.
From (39) it is straightforward to evaluate the specific heat C Z by means of (24); i.e.




CZ
h D 2 0 h D
1
2 0
2 0
= x1 (x1 ) + x2 (x2 )

.
(41)
kB
2
2
2
In the strictly ohmic limit D , this result agrees with the corresponding expression (35)
for C E . Some results obtained from (41) are depicted in figure 3.
For high temperatures but finite cutoff, i.e. if h /2, h D /2  1, one obtains
1
h 2 D
CZ
=
+ O(T 3 ).
(42)
kB
2 12(kB T )2
While this reproduces the correct classical result, the leading correction differs by a factor of
two from the high-temperature result of the specific heat (33) derived from the system energy E.
Very puzzling is the behaviour at low temperatures. In this limit, the specific heat
behaves like



 "
 3 #
CZ
kB T

4 3 kB T 3

=
1

13

+ O(T 5 ).
(43)
kB
3 h
D
15
h
D
D
New Journal of Physics 10 (2008) 115008 (http://www.njp.org/)

12
0.6

C Z /kB

0.4
0.2
D / =
5
1
0.2

0
0.2

0.2

0.4

0.6

0.8

h
kB T /

Figure 3. Partition function route: the specific heat C Z is depicted as a function of

dimensionless temperature for different ratios of D / . In the strict ohmic limit

D this result agrees with C E , as depicted in figure 2. The behaviour as
a function of finite cutoff is intriguing: while for D > , C Z is monotonically
growing with increasing temperature, as expected from thermodynamics, this
fails to be the case when the friction strength surpasses the cutoff strength.
Although still vanishing at absolute zero temperature we find at low temperatures
formally negative-valued specific heats for this dissipative quantum system.
Notably this result does again obey the Third Law in the same sense as C E does. Moreover,
equation (43) agrees with (34) for D  . The big surprise surfaces, however, for small cutoff
frequencies. For D < , the specific heat C Z becomes negative at very low temperatures. This
indicates a serious defect which must be hidden somewhere in the theory leading to (43).
The appearance of a negative specific heat is not restricted to the Drude model. This can be
seen by expanding the general expression (38) for the internal energy in powers of temperature.
By means of the EulerMacLaurin summation formula, one obtains
 

X


B2n 2 2n1
U = U0 +
(kB T )2n f (2n1) () f (2n1) (0) ,
(44)
(2n)!
h

n=1
where B2n are the Bernoulli numbers, f (k) denotes the kth derivative of f and the ground-state
energy is given by
Z
h
dx f (x).
U0 =
(45)
2 0
For the free damped particle, one finds, by means of (38),
(x) x 0 (x)
f (x) =
.
(46)
x + (x)
From (44), one readily obtains the specific heat




CZ X
B2n
2 kB T 2n1  (2n1)
=
f
() f (2n1) (0) .
(47)
kB
(2n

1)!
h

n=1
The leading term of this expansion yields
CZ
1 + 0 (0) kB T
=
+ O(T 3 )
kB
3 (0)
h
New Journal of Physics 10 (2008) 115008 (http://www.njp.org/)

(48)

13
in agreement with the specific heat (43) for the Drude model. This result implies that a negative
specific heat will be found within the approach based on the partition function for every damping
kernel with 0 (0) < 1 .
In contrast, for a damped harmonic oscillator with frequency 0 , one obtains
f (x) =

202 + x (x) x 2 0 (x)

,
02 + x (x) + x 2

(49)

resulting in the leading low-temperature behaviour of the specific heat

CZ
(0) kB T
+ O(T 3 ).
=
2
kB
3 0 h

(50)

Here, the specific heat remains positive independently of the damping kernel.
6. Partition function and density of states

The theory leading to the unexpected appearance of a negative specific heat, cf (43), is based
on an assumption and an approximation. The assumption is that the thermodynamic behaviour
can be described by the partition function (21). The approximation concerns the calculation
of this partition function and limits the validity of the specific heat (42) to sufficiently high
temperatures for a fixed size of the box confining the particle. By choosing a sufficiently large
box, one though can approach such low temperatures for which the specific heat C E becomes
negative but for which the approximation is still fully justified.
It is therefore the form of the partition function as the ratio of the two partition functions
of the total system and the bath that deserves further scrutiny. We will analyse the situation
within the Drude model. Inserting the Laplace transform (14) of the damping kernel into the
general expression (37) of the partition function, one finds


L 2 m 1/2 0(1 + x1 )0(1 + x2 )
Z=
,
(51)
h

0(1 + h D /2)
where 0(z) is the gamma function and x1,2 are defined in (32).
Formally, the partition function can be related to a density of states (E) of a damped
system by means of a Laplace transform [18]
Z
Z () =
dE (E) exp( E).
(52)
0

By its very definition as the number of states at energy E per unit energy, the density of
states (E) must not assume negative values. This basic property of the density of states also
restricts the admissible form of physically meaningful partition functions. In the remainder of
this section, we demonstrate that the unphysical behaviour of the specific heat as given by (43)
comes along with negative regions of the density of states.
In order to discuss the partition function and the density of states it is useful to shift the
origin of the energy scale to the ground-state energy of the damped free particle
 
 
h 1
D
h 2
D
U0 =
ln
+
ln
,
(53)
2
1
2
2
where the frequencies 1,2 have been defined in (32). This expression is obtained from (39) in
the limit of zero temperature. Instead of the partition function Z , we thus consider Z exp(U0 ).
New Journal of Physics 10 (2008) 115008 (http://www.njp.org/)

14
2

hD LD /L

1
0
D / =
=5
=1
= 0.2

1
2

4
(E U0 )/hD

Figure 4. Energy dependence of the density of states for D / = 0.2, 1 and 5.

The solid line corresponds to the density of states of an undamped particle in one
dimension.
As the specific heat (43) may become negative at low temperatures, we focus on the
behaviour of the density of states at low energies. For h D  1, one finds


L 1/2
Z exp(U0 ) =
L D D
#
"



2
D
2
D

1+
1 +
1 + O( 3 ) ,
(54)
6h D
72(h D )2
where


h
LD =
2mD

1/2
(55)

is a characteristic length related to the Drude cutoff frequency D . The inverse Laplace
transform of (54) yields the low-energy behaviour of the density of states



 " 

E U0
L
1/2
D
(E) =

1
+
h D L D D
h D
6

2

 !#
2 D
E U0
E U0 2
+
(56)
1
+O
.
72
h D
h D
Except for the delta function at the ground-state energy U0 , the low-energy behaviour of the
density of states is dominated by a constant which changes sign at = D . If the cutoff
frequency D is smaller than the damping constant , the density of states is no longer positive
everywhere. This readily explains the surprising behaviour of the specific heat found in the
previous section. In figure 4, the energy dependence of the density of states is shown for
D / = 0.2, 1 and 5. For D < , the density of states starts out with negative values as
expected. At larger energies, peaks of the density of states are observed which can be related
to the characteristic frequencies of the equation of motion (16), which, for D < , give rise to
New Journal of Physics 10 (2008) 115008 (http://www.njp.org/)

15
underdamped oscillations. For a sufficiently large damping constant , there may exist several
energy regions displaying a negative density of states. For D  , one recovers the limit of
vanishing damping where
L
1
0 (E) =
.
(57)
L D (h D E)1/2
In the previous section, we have seen that the specific heat of a damped harmonic oscillator
at low temperatures is always positive. An analysis along the lines presented above for the free
damped particle indeed shows that the density of states, apart from a delta-function contribution,
starts out at low temperatures with a positive value. However, the positivity condition of a
density of states defined by (52) is by no means obvious as we will discuss next.
By its statistical mechanical definition, the partition function is the sum over the Boltzmann
factors counted with the multiplicity of the respective eigenenergies and therefore coincides with
the Laplace transform of the density of states as a function of the energy. The corresponding
Laplace variable is the inverse temperature . This in particular implies that the inverse Laplace
transform of any proper partition function must not assume any negative values.
Apparently, the positivity of the density of states is not automatically guaranteed by the
above definition (21) as can be seen in the case of a free particle. Another example is provided
by a system that is coupled to a rather trivial bath consisting of a single oscillator with angular
frequency . In this case, one obtains for the partition function of the reduced system, according
to the definition (21)


TrS+osc [exp( H )] X
Z=
=
gn e En eh /2 eh /2 ,
(58)
Trosc [exp( Hosc )]
n
where E n denotes the eigenenergies of the system coupled to the oscillator and the positive
weights gn are the corresponding degeneracies. For a finite, spatially confined system, these
eigenenergies constitute a discrete spectrum. The first factor of the right-hand side coincides
with the partition function of the total system, Z S+B , and the second term represents the inverse
of the partition function of the bath consisting of a single oscillator. The inverse Laplace
transform (E) of Z becomes
X
X
(E) =
gn (E E n + h /2)
gn (E E n h /2).
(59)
n

It consists of two sets of delta functions, one with positive weights at the energies E n h /2,
whereas the second set, which is located at the shifted energies E n + h /2, possesses negative
weights. Therefore, neither doesR (E) represent a physically meaningful density of states nor
must its Laplace transform Z = dEexp( E)(E) be interpreted as the partition function of
the reduced system.
For reservoirs that represent true thermal baths, both the total system and the reservoir
alone will have a continuous spectrum such that a more subtle compensation of the different
contributions to the inverse Laplace transform may indeed lead to a positive density of states
and a corresponding, physically meaningful partition function.
7. Conclusions

In this work, we compared two definitions of specific heat for a free quantum particle
that interacts at a finite strength with an oscillator heat bath possessing a Drude cutoff.
New Journal of Physics 10 (2008) 115008 (http://www.njp.org/)

16
The first definition was based on the assumption that the equilibrium expectation value
of the Hamiltonian HS of the system can be identified with the thermodynamic internal
energy of the particle. The resulting specific heat approaches the classical result for high
temperatures and goes to zero for vanishing temperature in accordance with the Third Law.
The second definition is based on the ratio of the partition functions of the total system and
the heat bath, which traditionally is identified as the partition function of an open quantum
system [5, 6, 8, 9], [13][20].
The two definitions yield different results which even disagree in the high-temperature
regime in terms of their quantum corrections. If one associates a free energy with each of the
three partition functions of the total system, the system and the bath (as done commonly), the
free energy of the open system is expressed as the difference of the free energies of the total
system and the heat bath, in accordance with the most remarkable formula of Ford et al [17].
We found that this definition may lead to unphysical results for the damped free particle, such as
negative specific heat and negative regions of the density of states. A corresponding analysis of
the respective quantities for a damped harmonic quantum oscillator though does not show any
violations of their formal properties [8, 9]. This latter fact alone does not resolve the encountered
baffling: the finding of a thermodynamic consistent result which is in no apparent violation with
formal properties, such as a truly positive-valued density of states, does not necessarily imply
that the so obtained values are physically meaningful.
The definition of the specific heat via the expectation value of the system Hamiltonian HS
leads to a quantity that can be measured in principle. In contrast, no clear scheme seems to be
available for a measurement of the specific heat that derives from the partition function of the
reduced system.
The conundrum of the physical meaning of a partition function of a reduced system that
is coupled with finite strength to an environment thus remains. We emphasize once more
that the detected difficulties all evaporate when the dissipation can be described within the
weak-coupling limit. However, the impressive experimental progress on tailored, well-defined
nanosystems whose behaviour is quantum in nature but that are intrinsically in interaction with
omnipresent macroscopic environments, requires us to go beyond weak coupling. It thus calls
for an urgent clarification of the thermodynamics of nanosystems and related issues.
Acknowledgments

We acknowledge financial support from the German Excellence Initiative via the Nanosystems
Initiative Munich (NIM).
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]

Hnggi P and Ingold G-L 2005 Chaos 15 026105

Jarzynski C 2007 C. R. Phys. 8 495
Horowitz J and Jarzynski C 2007 J. Stat. Mech. P11002
Goychuk I and Hnggi P 2005 Adv. Phys. 54 525
Grabert H, Schramm P and Ingold G-L 1988 Phys. Rep. 168 115
Grabert H, Weiss U and Talkner P 1984 Z. Phys. B 55 87
Riseborough P S, Hnggi P and Weiss U 1985 Phys. Rev. A 31 471
Hnggi P and Ingold G-L 2006 Acta Phys. Pol. B 37 1537

New Journal of Physics 10 (2008) 115008 (http://www.njp.org/)

17
Ford G W and OConnell R F 2007 Phys. Rev. B 75 134301
Wang C-Y and Bao J-D 2008 Chin. Phys. Lett. 25 429
Bandyopadhyay M 2008 Preprint 0804.0290v1
Hrhammer C and Bttner H 2007 Preprint 0710.1716v2
Feynman R P and Vernon F L 1963 Ann. Phys., NY 24 118
Feynman R P 1972 Statistical Mechanics (Redwood City, CA: Addison-Wesley) p 82
Caldeira A O and Leggett A J 1983 Ann. Phys., NY 149 374
Leggett A J, Chakravarty S, Dorsey A T, Fisher M P A, Garg A and Zwerger W 1987 Rev. Mod. Phys. 59 1
Ford G W, Lewis J T and OConnell R F 1988 Ann. Phys., NY 185 270
Hanke A and Zwerger W 1995 Phys. Rev. E 52 6875
Dittrich T, Hnggi P, Ingold G L, Kramer B, Schn G and Zwerger W 1998 Quantum Transport and
Dissipation (Weinheim: Wiley-VCH) chapter 4 pp 21348
[20] Ingold G L 2002 Lect. Notes Phys. 611 1

[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]

New Journal of Physics 10 (2008) 115008 (http://www.njp.org/)