Sie sind auf Seite 1von 43

Equilibrium Statistical Thermodynamics of a

Many-Particle System Coupled to an External


Scalar Field
Published in Physical Review A, 41, 4236 (1990)
R. E. Salvino
Sverdrup Technology, Inc..
Lewis Research Center Group
Mail Stop 500/217
Cleveland, Ohio 44135
Abstract
The equilibrium thermodynamics of a many-particle assembly in
the presence of an external scalar eld is examined. Two types of
scalar coupling are considered: an external eld coupled to the parti-
cle density and an external scalar eld coupled to the energy density.
It is shown that the broken translational and rotational invariance of
the system due to the external eld is reected in the macroscopic
physics by loss of the usual extensivity property of the system and by
means of anisotropy in the response of the system to changes in the
system lengths or to the system shape. In addition, the assumptions
used in local equilibrium analyses are shown to be incorrect in princi-
ple. Nonlocal eects due to the external eld must be included in the
determination of the equation of state. Simple model calculations for
a system in an external gravitational eld and an externally imposed
temperature eld are presented as illustrations.
I Introduction
The eects of an external scalar eld on the equilibrium thermodynam-
ics of a many-particle system have never been fully developed. The usual
statistical-mechanical approach examines the eect of an external magnetic
or electric eld on a system of magnetic or electric dipoles [1], especially
with regard to phase transitions [2], or it examines small perturbations from
equilibrium [35]. The density-functional approach [6, 7] is a local theory
1
that makes a number of assumptions about local thermodynamic behavior.
This approach has also given much focus to phase transitions [68]. Such
analyses do not address all the implications that these elds have on the
formal development of the thermodynamics of many-particle systems. The
standard many-body theory approach buries the external eld in the single-
particie states used in the quantization process [9, 10]. This procedure also
hides the eects of the external eld and the implications for thermodynam-
ics are not very transparent. The more common approach labeled local
equilibrium [11] requires some further discussion.
The equilibrium conditions of a body or a system of particles in an
external eld requires the uniformity of both temperature and chemical po-
tential [11]. That is,
T(r) = T = constant (1.1)
(r) = = constant (1.2)
where T(r) and (r) are the local temperature and chemical potential. The
constant parameters T and are the parameters that are identied with
the Lagrange multipliers in the entropy maximization process that gives the
statistical density operator. At this point, one now makes an assumption
such as the chemical potential has the form
=
0
[P
loc
(r, T)] +U
ext
(r) (1.3)
U
ext
(r) is assumed to depend on the coordinates of the center of mass of
the system of particles composing the body and not on any internal co-
ordinates of the particles. The function
0
is the chemical potential in the
absence of the external eld as a function of the local pressure and the tem-
perature of the system. Assuming the external eld is a gravitational eld
of the mgz type and assuming a weak eld, (1.3) then gives
P
loc
(z) = constant mgz (1.4)
The constant is presumed to be independent of g. The problematic ther-
modynamic character, however, of (1.4) and its starting point (1.3) will be
2
shown below. It is important to note that (1.3) and (1.4) do not follow
from (1.1) and (1.2). The rejection of assumptions such as (1.3) does not
imply rejection of local equilibrium. That is, Eqs. (1.1) and (1.2) are not
questioned, but the form of the local equilibrium conditions on the pressure,
for instance, will not be given by an assumption such as (1.3).
Wolf and Tang [12] have shown the dierence between the negative of
the grand canonical potential per unit volume and the momentum current
for a system of particles in an external periodic potential. They, however,
make no distinction between the thermodynamic pressure and the negative
grand potential density. The thermodynamic pressure is dened in such a
way as to provide an alternative to the microscopic route to the pressure,
that is, to the momentum delivered per unit area per unit time to a surface.
As will be shown below, the connection between the two points of view can
always be made but the thermodynamic pressure will not be the negative
grand potential per unit volume. This distinction between the thermody-
namic pressure and the grand canonical potential density for a system in an
external eld is the basis of the dierence found by Wolf and Tang.
To establish the notation and to provide the usual reference system re-
sults for purposes of comparison, consider a single component collection of
particles which interact with a two-body central potential with no external
elds present. The Hamiltonian of such a system is of the form [9, 10]

H =
_
d
3
r

(r, t)

2
2m

(r, t)
+
1
2
_ _
d
3
r
1
d
3
r
2
V (r
12
)

(r
1
, t)

(r
2
, t)

(r
2
, t)

(r
1
, t) (1.5)
where

(r, t) is the many-particle eld operator, V (r
12
) is the two-body
interparticle potential, r
12
= |r
1
r
2
|, and m is the particle mass. The
process of maximizing the entropy S, dened by
S = k
B
Tr ln , (1.6)
subject to the constraints that
3

H = Tr

H = E,

N = Tr

N = N, (1.7)
1 = Tr = 1,
leads to the usual results [1, 9]
=
1
Z
e
(

H

N)
,
Z = Tr e
(

H

N)
. (1.8)
The parameters = k
B
T and appear by virtue of comparing the Lagrange
multipliers used to incorporate the constraints (1.7) in the maximization
process with the rst law of thermodynamics
dE = T dS + dN P
th
dV. (1.9)
Equation (1.9), of course, is the thermodynamic statement of the conserva-
tion of energy. P
th
is the thermodynamically dened pressure.
In addition, for the Hamiltonian (1.5), there are a set of microscopic
conservation laws which are simply derived from the general equation of
motion [4, 10]
i


A(r, t)
t
= [

A(r, t),

H], (1.10)
where the time dependence in

A is assumed to enter only through the elds.
The conservation laws are [4]
(r, t)
t
+

j(r, t) = 0
m

j(r, t)
t
+

T(r, t) = 0 (1.11)
(r, t)
t
+

(r, t) = 0
4
The particle density, momentum density, momentum density current, energy
density, and energy-density current are given by [4]
(r, t) =

(r, t)

(r, t)
m

j(r, t) =
i
2
_
_

(r, t)
_

(r, t)

(r, t)
_

(r, t)
_
_

T
kl
(r, t) =

2
4m
_
_

k
__

l
_

(r

, t)

(r, t)
_
r

=r

1
4
_
d
3
r

k
r

l
r

dV (r

)
dr

_
1
0
d

(X

, t) (Y

, t)

(X

, t) (1.12)
(r, t) =

2
2m
_

(r, t)
_

(r, t)
_
+
1
2
_
d
3
r

V (|r r

|)

(r, t) (r

, t)

(r, t)

k
(r, t) =
1
i
_

2
2m
_
2_
_

k
_

(r

, t)

(r, t)
_
r

=r
+
1
2
_
d
3
r

V (|r r

|)

(r

, t)

j
k
(r, t)

(r

, t)

1
4
3

l=1
_
d
3
r

k
r

l
r

dV (r

)
dr

_
1
0
d

(X

, t)

j
l
(Y

, t)

(X

, t)
where the variables X

and Y

are dened to be
X

r +
1
2
(1 +)r

Y

r
1
2
(1 )r

The currents are determined to within the curl of a vector and an additive
constant. The symbols ,

j,

T, , and

j

will be reserved for the above


dened operators.
5
Physical walls may be included as external constraints to conne the
system of particles to a nite volume. These walls can be represented as a
scalar wall potential and the wall Hamiltonian,
H
wall
=
_
d
3
r (r, t) U
wall
(xe

1
, ye

2
, ze

3
) (1.13)
should then be added to the Hamiltonian (1.5) [4,13]. Note that if U
ext
(x, y, z)
connes the system to the volume V
0
= L
x,0
L
y,0
L
z,0
then the wall potential
in (1.13) connes the system to the volume V = V
0
e

1
+
2
+
3
. This, however,
is usually treated as an artice that is used to relate the pressure to the work
done by the external wall potential in conning the system to the specied
volume,
P
th
=
1
3V
_
d
3
r (r)
wall
r U
wall
(r) (1.14)
where (r, t)
wall
is the local particle density in the conned system. As
Eq. (1.14) requires the solution of the inhomogeneous many-body problem,
at least in the vicinity of the walls, this form for the pressure is usually
not regarded as a computationally useful starting point for obtaining the
pressure.
Although the familiar alternative to (1.14), the virial theorem, has been
derived many times [4, 13], a procedure that will be used in later sections
will be presented here [4]. Since the adiabatic variation of the energy is the
so-called isolated variation [4] and since it is recognized that any unitary
transformation can be applied before taking the trace to obtain the results
(1.7), the energy variation at constant entropy and number of particles is

H
wall
= Tr

H = Tr

= Tr G

G. (1.15)
G is the unitary transformation, (

,

H

) = G( ,

H)G

, and the cyclic prop-


erty of the trace was used. Since dV = V d(
1
+
2
+
3
), a direct application of
the thermodynamic relation P = (E/V )
S,N
(that is, the transformation
G = 1) gives (1.14). However, using the unitary transformation
G

(r, t)G

= e
(
1
+
2
+
3
)/2

(xe

1
, ye

2
, ze

3
) (1.16)
6
eliminates the
i
dependence from the wall part of the Hamiltonian and in-
serts it in the kinetic and interparticle interaction pieces of the Hamiltonian.
The thermodynamic derivative for the pressure then gives

T
ii
= 2
_
d
3
r

(r, t)

2
i
2m

(r, t)

1
2
_
d
3
r
1
_
d
3
r
2
(r
12
)
2
i
r
12
dV (r
12
)
dr
12

(r
1
, t) (r
2
, t)

(r
1
, t)
P =
1
3V
3

i=1

T
ii

wall
(1.17)
P =
2

KE
wall
3V

1
6V
_ _
d
3
r
1
d
3
r
2
r
12

12
V (r
12
)S
wall
(r
1
, r
2
)
where

T
ii
are the diagonal elements of the volume-integrated momentum-
density current, r
12
is the relative coordinate r
1
r
2
, and S
wall
(r, r

) is
the two-body distribution function for the wall-conned system. In the end,
the walls are pushed out to innity and the system is translationally and
rotationally invariant so that (1.17) becomes
P =
2
3
_
d
3
k
(2)
3

2
k
2
2m
n(k)

2
6
_
d
3
r r V

(r) g(r) (1.18)


In (1.18), n(k) is the bulk momentum distribution function, g(r) is the
bulk radial distribution function, and the prime means dierentiation with
respect to the argument, that is, V

(r) = dV (r)/dr. The wall potential also


enters into the momentum and energy conservation laws
m

j(r, t)
t
+

T(r, t) = (r, t)U
wall
(r)
(1.19)
(r, t)
t
+

(r, t) = j(r, t) U
wall
(r)
7
but as the walls are pushed out to innity, Eqs. (1.11) are recovered. Al-
though formally present, the net result is to set U
wall
(r) = 0, that is, an
actual wall potential is not usually implemented.
Now the eect of a scalar eld similar to but distinct from U
wall
will
be developed. The present analysis will examine the eects of a scalar cou-
pling to an external eld on the equilibrium thermodynamics. Two types of
coupling will be considered, a particle-density-eld coupling and an energy-
density-eld coupling [4]

H
ext
=
_
d
3
r (r, t)U
ext
(r) (1.20)

H
ext
=
_
d
3
r (r, t)U
ext
(r) (1.21)
The microscopic forms for and are dened in (1.12). Clearly, the coupling
described by (1.20) is of the same form as (1.13), but U
ext
(r) is not assumed
to be localized in the region of the walls of the system and does not depend on
any of the wall parameters since the source of the eld U
ext
(r) is physically
distinct from the source of U
wall
(r). The particle density and the energy
density are the two signicant scalar microscopic quantities that can be
constructed from the elds, although (1.21) could be decomposed into a
kinetic energy coupling and a potential-energy coupling

H
ext
=
_
d
3
r

E
kin
(r, t)U
KE
ext
(r) (1.22)

H
ext
=
_
d
3
r

V
pot
(r, t)U
PE
ext
(r) (1.23)
Equation (1.20) has the simplest mechanical interpretation: the external
eld gives each particle an additional energy U
ext
(r) at the point r. Cou-
plings (1.21) (1.23) are less straightforward in part because the external
eld is dimensionless and in some sense requires extra mechanical considera-
tions [3,4]. From the microscopic point of view, (1.22) may be interpreted as
8
a local renormalization of the particle mass, m
ren
(r) = m/[1+U
KE
ext
(r)], and
(1.23) may be interpreted as a local renormalization of the two-body poten-
tial, V
ren
(r, r

) = V (|r r

|)[1 +U
PE
ext
(r)]. The interpretation of the energy
density-eld coupling will be based, however, on (1.21) and an attempt to
make a connection with the analyses of linear response theory [3, 4].
Since the couplings (1.20) and (1.21) will have eects on the thermo-
dynamic description of the system of particles that are independent of the
form of the coupling, a coupling-independent thermodynamic analysis will
be given. Section II contains a discussion of these thermodynamic eects
and shows the dierences from the usual thermodynamic description. The
implications for assumption (1.3) are developed. The eects of the particle-
density coupling (1.20) are presented in Sec. III and the eects of the energy-
density coupling (1.21) are developed in Sec. IV. Particular attention is paid
to the conservation laws and the implications for the assumption (1.3) are
further developed. Section V illustrates some of the results by a simple cal-
culation for noninteractinc particles in an external gravitational eld and
Sec. VI uses a perturbative approach to explore some eects of a gravi-
tational eld on a liquid-gas phase transition of the van der Waals type.
Section VII illustrates the energy-density coupling for the case of nonin-
teracting particles and explores the interpretation of an externally imposed
temperature gradient. Concluding remarks are contained in Sec. VIII.
II Thermodynamic Considerations
The standard thermodynamic relations for a single-phase, one-component
system may be generalized by including a coupling constant in the external
Hamiltonians (1.20) and (1.21)

H
ext
=
_
d
3
r (r, t)U
ext
(r)
(2.1)

H
ext
=
_
d
3
r (r, t)U
ext
(r)
to allow for changes in the external coupling. may be considered as a
dimensionless variable or as the strength parameter in the external potential.
9
Using the theorem in Sec. I [Eq. (1.15)], that the adiabatic variation is the
isolated variation,
_

_
S,N
=
_

H
ext

(2.2)
Here,

H
ext
is either of the external Hamiltomans in (2.1) and

means
that the thermal average is calculated with the Hamiltonian for coupling
constant . Consequently, the rst law of thermodynamics is now
dE = TdS +dN
2

i=1
V
ii
dL
i
L
i
+

H
ext

T =
_
E
S
_
N,L
i
,
=
_
E
N
_
S,L
i
,
(2.3)

ii
=
L
i
V
_
E
L
i
_
S,N,

H
ext

=
_
E

_
S,N,L
i
where L
i
means L
1
, L
2
, and L
3
or L
x
, L
y
, and L
z
, which are the system
lengths dening the geometrical volume of the system of particles (here
assumed rectangular for simplicity).
ii
is the thermodynamically dened
pressure matrix. The generalization of the pressure P
th
= (E/V )
S,N,
requires the shape of the volume to be held xed
P
th
=
_
E
V
_
S,N,,Shape
(2.4)
The condition of xed shape means the fractional changes of lengths in each
direction are the same, that is, dL
1
/L
1
= dL
2
/L
2
= dL
3
/L
3
. Consequently,
the thermodynamic pressure is a scalar, isotropic response of the system to
volume changes and is given by the trace of
ii
10
P
th
=
1
3
3

i=1

ii
(2.5)
Not only are the temperature T and chemical potential uniform through-
out the system, but so is the thermodynamic pressure in Eq. (2.5) as well.
Consequently, P
th
= constant should be appended to (1.1) and (1.2). This is
in contrast to the local momentum density current which is not uniform [see
below, Eqs. (3.11) and (4.12)]. Although a local pressure that is not uniform
throughout the system may be dened front the microscopic viewpoint [see
Eqs. (3.16) and (4.8)], the thermodynamic pressure obtained from (2.3)
(2.5) is not a local quantity and represents an average response of the system
as a whole to changes in the system volume. It will also be shown explicitly
below that
ii
=

T
ii

ext
/V , that is, the pressure is no longer given by (1.17)
if external elds are present. The microscopic momentum current requires
modication if the correct thermodynamic pressure is to be obtained.
The relations for the Helmholtz free energy F = E TS and the grand
potential = F N = E TS N follow in the usual way:
dF = SdT +dN
3

i=1
V
ii
dL
i
L
i
+

H
ext

(2.6)
d = SdT Nd
3

i=1
V
ii
dL
i
L
i
+

H
ext

(2.7)
In the standard thermodynamic system, the grand canonical potential is
extensive and is given by = P
th
(T, )V [1, 4, 9, 11, 13], where the thermo-
dynamic pressure is a function of the two intensive variables T and only.
Dening, by analogy with the standard system, a quantity P by
= PV (2.8)
(2.7) becomes
11
dP = sdT d +
3

i=1
V (
ii
P)
dL
i
L
i
h
ext
d

(2.9)
s =
S
V
, =
N
V
, h
ext
=

H
ext

V
and, consequently, the thermodynamics for the Helmholtz free energy per
unit volume f = F/V and the energy per unit volume = E/V are
df = Tds +d
3

i=1
V (
ii
P)
dL
i
L
i
+h
ext
d

(2.10)
d = sdT +d
3

i=1
V (
ii
P)
dL
i
L
i
+h
ext
d

(2.11)
P can no longer be identied with the thermodynamic pressure. Sections V
VII give explicit calculations showing that P = P
th
for systems in an
external gravitational eld and externally imposed temperature eld. As
noted above, the thermodynamic pressure P is
1
3
of the trace of
ii
so that
if

H
ext
= 0, then P = P
th
=
ii
and the standard thermodynamic results
are obtained. However, for

H
ext
= 0, then P = P
th
and the
ii
are, in
general, distinct so that (2.9) (2.11) indicate an anisotropy in the ther-
modynamics. This means that the system responds dierently to changes
in the length L
i
than to changes in the length L
k
due to the presence of

H
ext
, that is. the system may change its shape while keeping the volume
xed. Furthermore, as noted above, the usual extensivity property appears
to be absent. Equations (2.9) (2.11) indicate that the partial derivative of
a thermodynamic potential per unit volume with respect to the lengths of
the system while holding intensive parameters xed is not necessarily zero.
The thermodynamic relations indicate that, for a chosen pair of intensive
variables (for example, T and ), not all extensivity properties hold. Which
extensivity properties hold and which do not will depend upon the external
potential and which intensive variables are chosen as independent variables.
For instance, if the quantity
ii
P as a function of T, , L
i
, and is indepen-
dent of the L
i
, then f(T, , L
i
, ) will have a logarithmic dependence on the
12
system length parameters. As mentioned previously, this will be illustrated
explicitly in Sec. V. Of course, the size dependence of

H
ext

/V is not at
all apparent and will, in general, require potential-dependent statements.
Equations (2.9) (2.11) may be used to characterize the extensivity
property of the system in the following way. If the limit V of

H
ext

/V
exists and if (
ii
P) is of the order L

k
for > 0, then the extensivity
breaking may be ignored since they will introduce terms of the order L

k
which vanish as the lengths tend to innity. If the limit V of

H
ext

/V
does not exist and diverges more strongly than logarithmically, and/or if
(
ii
P) is of the order L

i
(again, with > 0), then the extensivity property
will be considered to be strongly broken. If the limit V of

H
ext

/V
does not exist and diverges with a logarithmic dependence on the L
i
and/or
if
ii
P) is independent of the L
i
, then extensivity will be considered
to be weakly broken. These properties, however, will also be dependent
on the properties of the external potential and require potential-dependent
statements.
It should be apparent from the thermodynamic relations (2.8) and (2.9),
and the denition of that P is actually the quantity which should appear in
the local equilibrium assumption (1.3) and not the thermodynamic pressure
P. If it is assumed that a local entropy density exists, then the Legendre
transformation = E TS N implies the local relation
(r)
ext
= Ts
loc
(r) + (r)
ext
P
loc
(r) (2.12)
where =
_
d
3
rP
loc
(r) = PV . Furthermore, = (P/)
T,L
i
,
=
(P
th
/)
T,L
i
,
follows from the basic thermodynamic relation (2.9). It will
be shown below that the usual microscopic expression for the thermody-
namic pressure (1.17) is not correct if there are additional external elds
present which are distinct from wall potentials. A purely nonlocal contri-
bution from the external eld must be included in the momentum current
if the microscopic approach is to give the correct thermodynamic pressure.
It would be tempting to identify P with the calculation of (1.17) for the
case of external elds present so that the distinction between the thermo-
dynamic pressure matrix and the momentum current exclusive of nonlocal
contributions would then be reected in the distinction between the pressure
P = (E/V )
S,N,shape
and the negative of the grand canonical potential
per unit volume PV = TS +N E. At present, however, there is no rm
theoretical basis for making such an identication. In any case, assumption
(1.3) may be correct but the interpretation P
loc
(r) as the local pressure is
13
not: P
loc
(r) must be replaced by the local grand canonical potential P
loc
(r).
The assumption (1.3) would then be equivalent to identifying P with the
calculation of the usual integrated momentum current (1.17). Of course, in
the absence of additional external elds, these quantities are identical.
The thermodynamic relations (2.3), (2.6), and (2.7) imply
E(S, N, L
i
, ) = E(S, N, L
i
, = 0) +
_

0
d

H
ext

(2.13)
F(T, N, L
i
, ) = F(T, N, L
i
, = 0) +
_

0
d

H
ext

(2.14)
(T, , L
i
, ) = (T, , L
i
, = 0) +
_

0
d

H
ext

(2.15)
In (2.13), the integration is done at constant S, N, and L
i
; in (2.14),
at constant T, N, and L
i
; and in (2.15), at constant T, , and L
i
. These
equations allow a complete formal separation of the external eld eects from
the standard thermodynamic properties of the system under consideration
(the = 0 terms). They are most useful, from the calculational point
of view, in perturbation theory. Equation (2.14) will be the basis of the
discussion in Sec. VI.
The additional thermodynamic parameter enriches and increases the
number of Maxwell relations as well as introducing some further exibility in
the thermodynamic description of the system. For instance, for the particle-
density-external-eld coupling (1.20), a generalized susceptibility [2, 5, 11]
distinct from the isothermal compressibility may be dened by
_

_
T,L
i
,
=
_

2
P

_
=
1

_
h
ext

_
T,L
i
,
(2.16)
This is analogous to the magnetic susceptibility [2, 5, 11] since it contains a
density-density uctuation. But because this uctuation is integrated with
the external potential, the zero-eld value of (2.16) (that is, the = 0
value) may or may not be zero depending upon the symmetry properties of
the external eld. A similar denition for the energy-density - external-eld
coupling (1.21) would involve the derivative (/) and would contain an
energy-densityenergy-density uctuation. This would more closely resem-
ble the specic heat at constant volume and number of particles.
14
Finally, it should be noted that a true maximization of the entropy
would require, in addition to the conditions given in Sec. I, maximization
with respect to the coupling constant . This adds the requirement that
(S/)
T,N,E,L
i
= 0 or

H
ext

= 0. The external eld decreases the en-


tropy by breaking some elements of the randomness of the system, that is,
by adding (in some sense) order to the system. The external eld gives the
degree to which the system is out of its most equilibrium state. The en-
tropy will always be higher by setting equal to zero, that is, by decoupling
from the external eld.
III Microscopic Considerations: Particle-DensityField
Coupling
Consider a Hamiltonian of the form (1.5), that is, a many-particle system
with pairwise additive interactions, with a particle-density coupling to an
external scalar eld of the form (1.20). The total Hamiltonian of the system
of particles is then

H =

E
kin
+

V
pot
+

H
ext
+

H
wall
(3.1)

E
kin
=
_
d
3
r

(r, t)

2
2m

(r, t) (3.2)

V
pot
=
1
2
_ _
d
3
r
1
d
3
r
2
V (r
12
)

(r
1
, t) (r
2
, t),

(r
1
, t) (3.3)

H
ext
=
_
d
3
r (r, t)U
ext
(r) (3.4)

H
wall
=
_
d
3
r (r, t)U
wall
(xe

1
, ye

2
, ze

3
) (3.5)
Eventually the walls will be pushed out to innity,

H
wall
0 and the system
will ll all space. Consequently, all reference to U
wall
(r) could be suppressed
as in Sec. I. However, to show the formal dierence between U
wall
(r) and
U
ext
(r), all references U
wall
(r) will be kept. The innite system can then be
obtained by simply setting U
wall
(r) = 0. The scalar eld U
ext
(r), however,
will be kept and all translational and rotational symmetry breaking terms
15
will he due to U
ext
(r). The lengths L
i
of the system container are wall
potential parameters and appear in the wall Hamiltonian. They do not
appear in the distinct external eld U
ext
(r).
It is important to note that the full Hamiltonian must be used in the con-
straint (1.7), otherwise

H
ext
will never appear in the statistical-mechanical
description of this system. That is, if the constraint (1.7) used

H
int
= E
where

H
int
=

E
kin
+

V
pot
, then only

H
int
will appear in the density operator
and

H
ext
will not appear. The constraint

H = E is required in order to
introduce

H
ext
in the density operator .
Following the procedure outlined in Sec. I for the pressure calculation,
P
th
=
_
E
V
_
S,N,,shape
=
1
3V
3

i=1
_
E

i
_
S,N,,shape
=
1
3V
3

i=1
_
G

i
G
_
(3.6)
the generalization of the usual virial theorem for the Hamiltonian (3.1) is
P
th
=
1
3V
3

i=1

T
ii

ext

1
3
_
d
3
r (r)
ext
r U
ext
(r) (3.7)

T
ii
is the volume-integrated momentum-density current dened in (1.17).
The unitary transformation (1.16) introduces
i
dependence in

H
ext
in the
same manner as it introduces
i
dependence in the kinetic and interparticle
potential-energy pieces of the Hamiltonian. Clearly, in the case U
ext
(r) = 0
the usual virial theorem result is obtained. The implicit appearance of
U
ext
(r) in
ext
means that the more general form for the integrated mo-
mentum current must be used since translational and rotational invariance
are broken by U
ext
(r). It should also be noted that (3.7) is not the volume
integral of the standard local momentum-density current (1.12) except in
the case U
ext
(r) = 0. The usual integrated local momentum current term in
(3.7) comes from

E
kin
+

V
pot
while the external force term comes from

H
ext
.
16
The local conservation laws for this system are [4]
(r, t)
t
+

j(r, t) = 0 (3.8)
m

j(r, t)
t
+

T(r, t) = (r, t)U
ext
(r) (r, t)U
wall
(r) (3.9)
(r, t)
t
+

(r, t) =

j(r, t) U
ext
(r)

j(r, t) U
wall
(r) (3.10)
(r, t),

j(r, t),

T(r, t), (r, t), and

j

(r, t) are the same operators dened


in (1.12). In equilibrium the average value of the time derivatives vanish so
that (3.9) becomes

T(r, t)
ext
= (r, t)
ext
U
ext
(r) (r, t)
ext
U
wall
(r) (3.11)
or, in component form,
3

j=1

T
ji
(r, t)
ext
= (r, t)
ext

i
U
ext
(r) (r, t)
ext

i
U
wall
(r) (3.12)
Clearly,

T
ji
(r)
ext
is not uniform in contrast to the thermodynamic pressure
(3.7). Multiplying (3.12) by r
k
, integrating over the volume of the system,
and dening the pressure matrix
kl
by

ki

1
V
_
d
3
r
3

j=1

j
_
r
k

T
ji
(r)
ext
_
+
1
V
_
d
3
r (r)
ext
r
k

i
U
wall
(r) (3.13)
17
leads to the result

ki
=
1
V

T
ki

ext

1
V
_
d
3
r (r)
ext
r
k

i
U
ext
(r) (3.14)
This is the microcopic expression for the pressure matrix whose diagonal
elements are dened thermodynamically in Sec. II, Eq. (2.3). Taking
1
3
of
the trace of
kl
and comparing with (3.7) shows that
P
th
=
1
3
3

j=1

jj
(3.15)
which agrees with the thermodynamic result as it must. The thermodynamic
pressure is determined by the quantity
kl
and not by the volume average
of the standard momentum-density current (1.17). The general relation be-
tween the thermodynamic pressure and the usual local momentum-density
current is Eq. (3.13). Most pressure measurements actually refer to the
momentum delivered per unit time per unit area of a surface, that is, a
force per unit area. This gives some conceptual priority to the mechanical
considerations for pressure, or rather to the momentum current. The lo-
cal momentum current is measured and the local pressure tensor, the local
force per unit area on the surfaces of interest, is identied with that local
momentum current. But if the local momentum current is to be interpreted
as the local pressure tensor, if it is to have thermodynamic signicance as a
pressure tensor, then the volume average of the trace of the local momentum
current must give the thermodynamic pressure as it does in the case of no
external elds. This requires the addition of a purely nonlocal contribution
from the external eld to the standard momentum current. The correct
local pressure tensor, the pressure tensor whose volume average yields the
thermodynamic pressure matrix and pressure, is
P
ki
(r)

T
ki
(r)
ext

1
V
_
d
3
r (r)
ext
r
k

i
U
ext
(r) (3.16)
The volume average of (3.16) gives (3.14). The denition of the local pres-
sure is the trace of the pressure tensor P
kl
(r),
18
P
loc
(r) =
1
3
3

i=1

T
ii
(r)
ext

1
3V
_
d
3
r (r)
ext
r U
ext
(r). (3.17)
The volume average of (3.17) gives (3.7). It is clear that P
loc
(r) is the trace
of P
kl
(r) and not T
kl
(r)
ext
, the dierence being a nonlocal contribution
from the external eld. The local pressure tensor dened in (3.16) obeys the
same equations, (3.9) and (3.11), as T
kl
(r)
ext
, but will dier from the usual
local momentum-density current by a function of thermodynamic variables
and external eld parameters. The freedom to choose an arbitrary operator
and/or function which is independent of the space coordinates to be added
to the usual local momentum-density current operator must be exploited if
the microscopic approach is to give the correct thermodynamic quantities.
In eect, a boundary condition, the volume average of the local momentum
current must give the thermodynamic pressure matrix, has to be supplied.
It is important to recognize that such nonlocal eects do not contribute to
dierences in pressure at two distinct points, but they must be included if
the correct equation of state is to be obtained. In particular, the purely local
equation of state implied by the local equilibrium assumption (1.3) is not
correct. The local pressure in (3.17) has the form P
loc
(r) = P
0
(
loc
(r, T)) +
f
ext
(, T, L
x
, L
y
, L
z
), where the P
0
term is the local equilibrium equation of
state and the f
ext
term is the purely nonlocal contribution from the external
eld. It is this nonlocal contribution which invalidates the local equilibrium
assumption (1.3) for the equation of state.
The assumption leading to (1.4) must be incorrect in principle or, at
least, reinterpreted. This point, discussed above, will be clearly illustrated
in Sec. V. Since the thermodynamic relation (2.9) shows that P is the
fundamental thermodynamic potential and not P
th
, the local equilibrium
assumption (1.3) will not give the correct equation for equilibrium for the
local pressure, although it may give the correct equation for the local grand
potential P
loc
(r). This gives added incentive for identifying the trace of
T
kl
(r)
ext
, that is, the usual momentum current, with the local grand canon-
ical potential. Only in the absence of an external eld can this be the local
pressure.
Finally, it should be noted that since the equilibrium value of the time
derivatives are zero, the divergence of the equilibrium particle current is also
zero so that the energy conservation equation (3.10) gives the equilibrium
equation
19

(r)
ext
+

j(r)
ext
U
ext
(r) +

j(r)
ext
U
wall
(r)

= 0 (3.18)
Furthermore, since the total energy density of the system of panicles is

tot
= +U
ext
+U
wall
, (3.10) can be written as
(r, t)
t
+
_

(r) +

j(r)U
ext
(r) +

j(r)U
wall
(r)

= 0 (3.19)
Thus the total energy of the system is conserved as expected.
In summary, the scalar potentials U
ext
(r) and U
wall
(r) enter the micro-
scopic formalism in similar ways. However, due to the physically distinct
sources of these elds and the independence of the eld parameters, U
ext
(r)
and U
wall
(r) enter into the pressure and the grand canonical potential in dif-
ferent ways, which yields a formal distinction between the thermodynamic
pressure and the negative of the grand potential per unit volume. At the
microscopic level, it is this dierence which invalidates assumption (1.3)
used to augment local equilibrium arguments and requires purely nonlocal
contributions to the momentum density current.
IV Microscopic Considerations: Energy-DensityField
Coupling
Now consider a Hamiltonian of the form (3.1) except that

H
ext
is not
given by (3.4) but by

H
ext
=
_
d
3
r (r, t)U
ext
(r) (4.1)
The energy-density operator (r, t) is dened in (1.12). Since it was shown
in the preceding section how the wall potential enters into the analysis, it will
be suppressed here in the manner of Sec. I for simplicity. As mentioned in
the introduction, the external eld U
ext
(r) is a dimensionless quantity and
will require extra mechanical considerations to give a physical interpretation
to such an external coupling (see Sec. VII below).
20
The pressure calculation (3.6) now gives
P
th
=
1
3V
3

i=1
_
d
3
r

T
ii
(r)
ext
_
1 +U
ext
(r)

1
3V
_
d
3
r (r)
ext
r U
ext
(r) (4.2)
Once again, the usual results are obtained in the limit of zero external eld
and there is an implicit dependence on U
ext
(r) in the thermal average as
well as the explicit dependence on U
ext
(r). Again, it should be noted that
(4.2) is not the volume average of the usual local momentum-density current
(1.12) except in the case that U
ext
(r) = 0. The usual integrated momentum
current term comes from the

E
kin
and

V
pot
terms in the Hamiltonian, while
the external potential and force terms come from the

H
ext
term as in Sec. III.
The local conservation laws for the coupling (4.1) are [14]
(r, t)
t
+

j(r, t) =
_

j(r, t)U
ext
(r)

(4.3)
m

j(r, t)
t
+

T(r, t) = (r, t)U
ext
(r)
_

T(r, t)U
ext
(r)

(4.4)
(r, t)
t
+

(r, t) =
_

(r, t)U
ext
(r)

(r, t) U(r) (4.5)


Again, the particle density, momentum density, momentum-density current,
energy density, and energy-density current are the same operators given in
(1.12). Following the same line of development as in Sec. III for Eq. (4.4)
shows that

ki
=
1
V
_
d
3
r

T
ki
(r)
ext
_
1 +U
ext
(r)

1
V
_
d
3
r (r)
ext
r
k

i
U
ext
(r) (4.6)
21
This is the microscopic denition of the pressure matrix
kl
for the energy-
density coupling; the diagonal elements are dened thermodynamically in
Eq. (2.3). Taking
1
3
of the trace of (4.6) and comparing with (4.2) shows
that
P
th
=
1
3
3

l=1

ll
(4.7)
as in (3.15) and (2.5).
Dening a local pressure in a way similar to (3.17), so that the volume
average of the local pressure gives the thermodynamic pressure, requires
P
loc
(r) =
1
3
3

i=1

T
ii
(r)
ext
_
1 +U
ext
(r)

1
3V
_
d
3
r (r)
ext
r U
ext
(r) (4.8)
As for (3.17), P
loc
(r) is not the trace of the usual local momentum-density
current. In other words, the additional local and nonlocal terms due to
the external eld are required in order for the volume average of the local
pressure to be equal to the thermodynamic pressure. The points made in
the discussion after Eq. (3.17) are also relevant here. It may be noted that
(4.8) indicates that the local equation of state is given by
P
loc
(r) = P(
loc
(r), T)
_
1 +U
ext
(r)

+f
ext
(T, , L
x
, L
y
, L
z
)
where T is the constant thermodynamically dened temperature. The con-
nection with a local temperature is discussed at the end of this section and
is illustrated in Sec. VII.
Note that U
ext
(r) gives an additional particle current since (4.3) can be
written as
(r, t)
t
+
_

j(r, t)
_
1 +U
ext
(r)

_
= 0 (4.9)
22
Consequently, the full particle-density current is no longer the momentum
density per unit mass as in (1.12). There is also an additional energy current.
Since the total energy density is (r, t)[1 +U
ext
(r)], (4.5) can be written as

tot
(r, t)
t
+
_

(r, t)
_
1 +U
ext
(r)

2
_
= 0 (4.10)
As for the external coupling in Sec. III, the total energy of the system is
conserved as should be expected. A comparison of the set of equations for
the two types of external coupling indicates that a main formal distinction
between the scalar coupling (r, t)U
ext
(r) and (r, t)U(r) is that the U
ext
(r)
term enhances the particle current density whereas the U
ext
(r) term does
not.
The equilibrium, time-independent equations are

j(r)
ext
_
1 +U
ext
(r)

_
= 0 (4.11)

T(r)
ext
_
1 +U
ext
(r)

_
= (r)
ext
U
ext
(r) (4.12)

(r)
ext
_
1 +U
ext
(r)

2
_
= 0 (4.13)
In the nonequilibrium case, the energy-density current would be modeled
by [3, 4]

tot
(r, t) =

(r, t) T(r, t) (4.14)


where T(r, t) is the local temperature eld and is the thermal conductivity.
There cannot, however, be a direct comparison of the linear response model
with (4.13) since the equilibrium, time-independent quantities appear as co-
ecients in the zero-frequency () contributions to the response functions.
That is, the temperature gradient term does not enter in the zero-frequency
contribution to the linear response functions [4].
However, (4.12) shows that, to lowest order in the external eld, the
divergence of the momentum current density is proportional to the gradient
of the external eld. If it is assumed that the energy current is proportional
to U
ext
(r) as well, then to lowest order (4.13) shows that
2
U
ext
(r) = 0.
23
If U
ext
(r) is interpreted as a temperature gradient, then this gives the
time-independent or steady-state heat equation. Furthermore, if the particle
current is also assumed to be proportional to the external eld gradient, the
same steady-state equation for the external eld is obtained. In addition, if
the particle current is modeled as a density gradient, then the conventional
wisdom which states that a density gradient produces an energy current
by establishing a temperature gradient [15] is obtained. Consequently the
interpretation that the external eld coupled to the energy density is an
externally imposed and dimensionless temperature eld leads to no obvious
internal inconsistencies. However, the temperature dened in (2.3) is a con-
stant and cannot be identied with the local temperature, although it could
be identied with the volume average of the local temperature. Microscop-
ically, this may pose problems of denition, but it may be fruitful to dene
a local temperature by means of the local kinetic energy density in analogy
with the classical equipartition result E
kin
=
3
2
Nk
B
T. This is similar to
dening a local pressure in terms of the local momentum current such that
its volume average is the thermodynamic pressure. This classical local tem-
perature, however, must be dened with some care so it will have a volume
average given by the thermodynamic temperature (2.3). An illustration of
these ideas will be given in Sec. VII.
V Noninteracting Particles in an External Gravitational Field:
Classical Case
The canonical partition function for noninteraciing classical particles
(Maxwellons) in an external eld U
ext
(r) is given by
Z
can
(T, N, L
i
) = Z
FP
(T, N, V )
_
1
V
_
d
3
r e
Uext(r)
_
N
(5.1)
where L
x
, L
y
, and L
z
appear in the limits of integration and Z
FP
is the
Maxwellon free-particle partition function. Consequently, the thermody-
namics for this system is given by the following equations:
F = F
FP
Nk
B
T ln
_
1
V
_
d
3
r e
Uext(r)
_
(5.2)
24
=
FP
k
B
T ln
_
1
V
_
d
3
r e
Uext(r)
_
(5.3)
= PV = F N =
FP
= Nk
B
T (5.4)
S = S
FP
+Nk
B
ln
_
1
V
_
d
3
r e
Uext(r)
_
+Nk
B
_
d
3
r U
ext
(r)e
Uext(r)
_
d
3
r e
Uext(r)
(5.5)
E = F +TS = E
FP
+Nk
B
T
_
d
3
r U
ext
(r)e
Uext(r)
_
d
3
r e
Uext(r)
(5.6)

ii
= k
B
T k
B
T
_
d
3
r r
i

i
U
ext
(r)e
Uext(r)
_
d
3
r e
Uext(r)
(5.7)
P
th
=
1
3
3

i=1

ii
= k
B
T k
B
T
_
d
3
r r U
ext
(r)e
Uext(r)
_
d
3
r e
Uext(r)
(5.8)
Again, it should be noted that P
th
= P. Specializing to the case of an
external potential of the mgz type, and taking the bottom of the system
container to be at z = z
bot
= 0 and the zero point of the potential z
0
to be
z
0
= z
bot
, Eqs. (5.1) - (5.8) become
Z
can
(T, N, L
i
, g) = Z
FP
(T, N, V )
_
1 e
mgLz
mgL
z
_
N
(5.9)
F = Nk
B
T
_
ln(
3
t
) 1

Nk
B
T ln
_
1 e
mgLz
mgL
z
_
(5.10)
= k
B
T ln(
3
t
) k
B
T ln
_
1 e
mgLz
mgL
z
_
(5.11)
S = Nk
B
_
5
2
ln(
3
t
) + ln
_
1 e
mgLz
mgL
z
_

_
mgL
z
e
mgLz
1
1
__
(5.12)
25
E =
3
2
Nk
B
T Nk
B
T
_
mgL
z
e
mgLz
1
1
_
(5.13)
= Nk
B
T = PV (5.14)

xx
= k
B
T (5.15)

yy
= k
B
T (5.16)

zz
= k
B
T +k
B
T
_
mgL
z
e
mgLz
1
1
_
(5.17)
P
th
= k
B
T +
1
3
k
B
T
_
mgL
z
e
mgLz
1
1
_
(5.18)

t
is the thermal de Broglie wavelength h/(2mk
B
T)
1/2
. It is instructive
to examine these equations in two distinct limits: for mgL
z
>> 1 and
mgL
z
<< 1. The usual bulk limit requires L
x
, L
y
, L
z
and N , but
in such a way that the thermodynamic parameters per unit volume (, ,
etc.) are nite. With , m, and g xed, the L
z
limit corresponds to
mgL
z
>> 1:
f = k
B
T
_
ln(
3
t
) 1

+k
B
T ln(mgL
z
) (5.19)
= k
B
T ln(
3
t
) +k
B
T ln(mgL
z
) (5.20)
s = k
B
_
7
2
ln(
3
t
)

k
B
ln(mgL
z
)
_
(5.21)
=
5
2
k
B
T (5.22)
P = k
B
T (5.23)

xx
=
yy
= k
B
T (5.24)
26

zz
= 0 (5.25)
P
th
=
2
3
k
B
T (5.26)
The grand potential (5.23) was previously identied as the thermodynamic
pressure [16]. Note the logarithmic divergence in the free energy density,
chemical potential, and entropy density. Note also that the external gravi-
tational contribution to the energy density is k
B
T [16], the z component of
the volume-averaged pressure tensor is zero, and the free Maxwellon pres-
sure has been lowered from k
B
T to
2
3
k
B
T. The pressure indicates that
the system of noninteracting Maxwellons has been compressed by the ex-
ternal gravitational eld into a quasi-two-dimensional system. Since s 0,
the usual free-particle result requires 12.18
3
t
. With the external eld
present, s 0 requires 33.12/mgL
z

3
t
. This requires the system to be
more dilute (at xed T) or at higher temperature (at xed ) than without
the external eld.
In the opposite limit mgL
z
<< 1,
f = k
B
T
_
ln(
3
t
) 1

+
1
2
mgL
z
_
1
1
12
mgL
z
+
_
(5.27)
= k
B
T ln(
3
t
) +
1
2
mgL
z
_
1
1
12
mgL
z
+
_
(5.28)
s = k
B
_
5
2
ln(
3
t
)
1
24
(mgL
z
)
2
+
_
(5.29)
=
3
2
k
B
T +
1
2
mgL
z
_
1
1
6
mgL
z
+
_
(5.30)
P = k
B
T (5.31)

xx
=
yy
= k
B
T (5.32)

zz
= k
B
T
1
2
mgL
z
_
1
1
6
mgL
z
+
_
(5.33)
27
P
th
= k
B
T
1
6
mgL
z
_
1
1
6
mgL
z
+
_
(5.34)
The choices for zero point of the external potential may be considered to
be related by a Legendre transformation. For instance, the choice z
0
=
z
bot
+
1
2
L
z
may he considered to be related to the calculations done here
by introducing an energy E
1
2
NmgL
z
which will lead to a renormalized
chemical potential
1
2
mgL
z
, a renormaliized pressure P +
1
6
mgL
z
, and
so forth.
From the microscopic viewpoint, it should be noted that the exact particle-
density prole [17, 18] for the particle-density coupling is given by

loc
(r
1
) =
1
Z
can
_
d
3
p
1
d
3
p
N
_
d
3
r
2
d
3
r
N
e
H
(5.35)

loc
(z) =
loc
(0)e
mgz
(noninteracting system)
Due to the form for the noninteracting system local density (5.35), the local
momentum current or pressure tensor is simply
P
ij
(z) =
loc
(z)k
B
T
ij
+k
B
T
_
mgL
z
e
mgLz
1
1
_

iz

jz
(5.36)
The nonlocal contribution from the external gravitational eld is required
so that the volume average of (5.36) will give the correct thermodynamic
pressure. This clearly shows by explicit calculation that the equation of
state for noninteracting particles in an external gravitational eld is not of
the local equilibrium form, but is given by the form
P
loc
(z) = P
0
(
loc
(z), T) +f(, T, L
z
, g) (5.37)
In (5.37), P
0
(
loc
(z), T) is the local equilibrium equation of state, is the
average density of the system, and all the nonlocal eects of the external
28
coupling are contained in the function f. It should be noted that the local
equilibrium approach to the pressure (1.4) can be saved, at least to lowest
order in g, if the constant term is interpreted as including some nonlocal ef-
fects of the external coupling. As indicated previously, the local equilibrium
equation of state is sucient to calculate pressure dierences. However, the
nonlocal contributions from the external eld must be included if the cor-
rect equation of state for the thermodynamic pressure is to be obtained. It
is also of interest to note that the equation of state for the grand potential
suggests that
P
loc
(z) =
loc
(z)k
B
T = P
0
(
loc
(z), T) (5.38)
In other words, the local equilibrium equation of state yields the correct
grand potential, but is not sucient to give the correct thermodynamic
pressure.
VI Mean-Field Phase Transition in an External Field: Classical
Case
Since the main interest in this section will be in treating the external
eld as a perturbation, it is convenient to begin with Eq. (2.14),
F(T, N, L
i
, g) = F(T, N, V, g = 0) +
_
g
0
dg

H
ext

g
(6.1)
The gravitational eld was chosen as the external eld and the integration
over the gravitational eld strength is done at constant N, T, and L
i
. The
free energy in zero eld will be modeled to correspond to a mean-eld equa-
tion of state,
F(T, N, V, g = 0) = F
HS
(T, N, V ) a
N
2
V
(6.2)
where F
HS
(T, N, V ) is the free energy of a system of hard spheres. To
evaluate (6.1) treal g as a small parameter or, more exactly, treat

H
ext

g
as small for all g

between zero and g. To second order in g

,
29

H
ext

g
=
1
2
mg

NL
z
cNk
B
T(mg

L
z
)
2
+ (6.3)
where
z
i

0
=
1
2
L
z
(6.4)
c
1
4
_
4
NL
2
z
N

i=1
N

j=1
z
i
z
j

0
N
_
(6.5)
For noninteracting particles, c
FP
=
1
12
(the subscript FP means free par-
ticles) and computer simulations of Lennard-Jones particles indicate c is
about 0.07 for the liquid system near the triple point. It might be expected,
then, that c will be of the order
1
12
in general. It might also be expected,
however, that the uctuations in (6.5) would diverge in the critical region. If
so, then the analysis based on the expansion (6.3) would no longer be valid.
This would necessitate a nonperturbative treatment and suggests that the
gravitational contribution to Eq. (6.1) may contain signicant non-mean-
eld statements. Consequently, in the spirit of mean-eld theory, c will be
treated as a nite constant so that the full free energy of the system of
particles is given by
F(T, N, L
i
, g) = F
HS
(T, N, V ) a
N
2
V
+
1
2
mgNL
z

1
2
Nk
B
Tc(mgL
z
)
2
+ (6.6)
From thermodynamics, the pressure matrix is given by

ii
=
0
ii

1
2
mgL
z

i,z
+k
B
Tc(mgL
z
)
2

i,z
(6.7)

0
ii
is the thermodynamic pressure matrix in zero external eld and
i,k
is the Kronecker delta function. To examine the consequences of (6.7) in
30
greater detail, a mean-eld equation of state will be assumed for the zero-
eld pressure tensor. Introducing dimensionless variables by the denitions
= b
= bk
B
T/a

ii
= b
2

ii
/a (6.8)
= bmgL
z
/a
the zero-eld equation of state is given by

0
ii
= g
HS
()
2
(6.9)
where g
HS
() is the hard-sphere function and a and b are the usual van der
Waals parameters. Explicitly, the thermodynamic pressure matrix compo-
nents and the thermodynamic pressure are

xx
=
yy
= g
HS
()
2
(6.10)

zz
= g
HS
()
2

1
2
+
c
2

(6.11)
= g
HS
()
2

1
6
+
c
2

3
(6.12)
In Eq. (6.12), is the dimensionless thermodynamic pressure which is
1
3
of
the trace of
ii
. Following the elementary treatment of the critical point [11],
the inection points of
ii
and will now be investigated. To simplify the
analyses, a simple van der Waals gas will be assumed, that is, the hard-
sphere function will be taken to be g
HS
() = 1/(1 ). Thus the x and y
analyses give the usual van der Waals results [11]:

x
c
=
y
c
=
8
27
(6.13)
31

x
c
=
y
c
=
1
3
(6.14)

c
xx
=
c
yy
=
1
27
(6.15)
Equations (6.13) and (6.14) are the x and y direction critical point values.
The z component and pressure analyses are altered due to the terms linear
in ,
_

zz

_
,
=

(1 )
2
2

2
+
c
2

(6.16)
_

_
,
=

(1 )
2
2

6
+
c
2
3
(6.17)
while the second derivatives are the same as the zero external eld case.
The solution for the critical point parameters yields

z
c
=
8
27
+
2
9

_
27c 1
18
_

2
+O(
3
) (6.18)

z
c
=
1
3


6

9c
8

2
+O(
3
) (6.19)

c
=
8
27
+
2
27

_
51c 1
162
_

2
+O(
3
) (6.20)

c
=
1
3


18

3c
8

2
+O(
3
) (6.21)
Equations (6.18) and (6.19) give the z direction critical point; the pa-
rameters
c
and
c
are the critical point parameters for the thermodynamic
pressure .
To understand the signicance of these results, imagine a dilute gas with
<<
1
3
and at high temperature >>
8
27
. As particles are added to the
system at constant temperature, the density
z
c
will be reached before the
other critical densities since
z
c
<
c
<
x,y
c
. Now if the temperature is
lowered keeping the density =
z
c
xed, then
z
c
will be reached rst since
it is the largest of the three critical temperatures,
z
c
>
c
>
x,y
c
. For
32
=
z
c
and =
c
, both
xx
(=
yy
) and are in their single phase regions
while
zz
is at its inection or critical point. A further small decrease of
temperature will then bring
zz
into its two phase region while
xx
(=
yy
)
and remain single phase. Since these are volume-average properties, this
may be characterized as condensation in layers in a loose sense only. At the
local level, the x and y components of the pressure tensor will be of the local
equilibrium form while the z component will have the additional nonlocal
contribution from the external eld. This contribution is responsible for
the distinction in critical points described by the above global analysis. At
the height z = z
c
for which the local density is equal to the z direction
critical density
z
c
, and for the temperature xed at the z direction critical
temperature
z
c
, the local x and y pressure tensor components are in their
single phase domain while the z component is at its critical point. For
heights above that critical height z = z
c
the z component is in its single
phase domain, while for heights below z
c
, it is in its two phase domain. The
x and y components, however, will remain in their single phase domains for
any value of the height at the z direction critical temperature
z
c
. For each
value of the height z the system is in a distinct thermodynamic state, single
phase states above z
c
and two phase states below z
c
[19].
To close this section, it may be of interest to dene a generalized sus-
ceptibility for the pressure equation of state obtained from (6.7). Since
the derivation of (6.7) requires only that the zero-eld equation of state be
known, it will now be assumed that the zero-eld equation of state is the
exact equation of state and not of the van der Waals type. In analogy with
magnetic systems, the generalized susceptibility is dened as

g
=
_

g
_
T,P
(6.22)
The equation of state obtained from (6.7), ignoring terms of second order in
g, yields

g
=
1
6
mL
z
_
P
0
(T,)

_
T

1
6
mgL
z
(6.23)
Consequently, the zero-eld susceptibility, that is, the limit g 0 of
g
di-
verges as the critical point is approached in the same manner as the isother-
mal compressibility,
33

g=0

T
|T T
c
|

(6.24)
Equation (6.24) is quite analogous to the zero-eld susceptibility in the
magnetic system case. On the other hand, at the zero-eld critical point,
the rst term in the denominator of (6.23) vanishes and consequently

T=T
0
,=
0
=

0
g
(6.25)
where T
0
is the zero-eld critical temperature and
0
is the zero-eld critical
density. The fact that (6.25) shows a negative divergence in zero g has,
apparently, no simple interpretation. But since
g
measures how easy it is
to change the average particle density at constant temperature and pressure
by means of an external gravitational eld, (6.25) may be the indication that
such changes are virtually impossible in a microgravity environment [19].
This is expected since, in zero external eld, the density cannot be changed
if both the temperature and pressure are held xed.
VII Noninteracting Particles in an External Temperature
Gradient: Classical Case
The canonical partition function for noninteracting classical particles
with the energy-density-external-eld coupling (4.1) can be calculated in a
way similar to the particle-density-external-eld case of Sec. V. For such a
system of Maxwellons, the partition function is given by
Z
can
(T, N, L
i
) = Z
FP
(T, N, V )
_
1
V
_
d
3
r
_
1 +U
ext
(r)

3/2
_
N
(7.1)
where L
x
, L
y
, and L
z
appear in the limits of integration and Z
FP
is the
Maxwellon free-particle partition function. Consequently, the thermody-
namics for this system is given by the following set of equations:
F = F
FP
Nk
B
T ln
_
1
V
_
d
3
r
_
1 +U
ext
(r)

3/2
_
(7.2)
34
=
FP
k
B
T ln
_
1
V
_
d
3
r
_
1 +U
ext
(r)

3/2
_
(7.3)
= PV = F N =
FP
= Nk
B
T (7.4)
S = S
FP
+Nk
B
ln
_
1
V
_
d
3
r
_
1 +U
ext
(r)

3/2
_
(7.5)
E = F +TS = E
FP
=
3
2
Nk
B
T (7.6)

ii
= k
B
T
3
2
k
B
T
_
d
3
r r
i

i
U
ext
(r)
_
1 +U
ext
(r)

5/2
_
d
3
r
_
1 +U
ext
(r)

3/2
(7.7)
P
th
=
1
3
3

i=1

ii
= k
B
T
1
2
k
B
T
_
d
3
r r U
ext
(r)
_
1 +U
ext
(r)

5/2
_
d
3
r
_
1 +U
ext
(r)

3/2
(7.8)
Again, it should he noted that P
th
= P. Specializing to the case of a
constant external-eld gradient in the z direction, U
ext
(r) = z, and taking
the bottom of the system container to be at z = z
bot
= 0 and the zero point
of the potential z
0
to be z
0
= z
bot
, Eqs. (7.1) (7.8) become
Z
can
(T, N, L
i
, ) = Z
FP
(T, N, V )
_
2
L
z
_
1
_
1 +L
z
_
1/2
_
_
N
(7.9)
F = Nk
B
T
_
ln(
3
t
) 1

Nk
B
T ln
_
2
L
z
_
1
_
1 +L
z
_
1/2
_
_
(7.10)
= k
B
T
_
ln(
3
t
) 1

k
B
T ln
_
2
L
z
_
1
_
1 +L
z
_
1/2
_
_
(7.11)
S = Nk
B
_
5
2
ln(
3
t
) + ln
_
2
L
z
_
1
_
1 +L
z
_
1/2
_
__
(7.12)
35
E =
3
2
Nk
B
T (7.13)
= Nk
B
T = PV (7.14)

xx
= k
B
T (7.15)

yy
= k
B
T (7.16)

zz
= k
B
T k
B
T
_
1
1
2
L
z
(1 +L
z
)
3/2
(1 +L
z
)
_
(7.17)
P
th
= k
B
T
1
3
k
B
T
_
1
1
2
L
z
(1 +L
z
)
3/2
(1 +L
z
)
_
(7.18)
where
t
is the thermal de Broglie wavelength h/(2mk
B
T)
1/2
. As for the
external gravitational eld case, it is instructive to examine these equations
in two distinct limits: for L
z
>> 1 and L
z
<< 1. The usual bulk limit
requires L
x
, L
y
, L
z
, and N , but in such a way that the thermodynamic
parameters per unit volume (, , etc.) are nite. With xed, the L
z

limit corresponds to L
z
>> 1,
f = k
B
T
_
ln(
3
t
) 1

+k
B
T ln
_
1
2
L
z
_
(7.19)
= k
B
T ln(
3
t
) +k
B
T ln
_
1
2
L
z
_
(7.20)
s = k
B
_
5
2
ln(
3
t
)
_
k
B
ln
_
1
2
L
z
_
(7.21)
=
3
2
k
B
T (7.22)
P = k
B
T (7.23)
36

xx
=
yy
= k
B
T (7.24)

zz
= 0 (7.25)
P
th
=
2
3
k
B
T (7.26)
Note the logarithmic divergence in the free energy density, chemical poten-
tial, and entropy density as for the external gravitational coupling case. Note
also that the external-eld contribution to the energy density is zero unlike
the gravitational case, while the z component of the volume-averaged pres-
sure tensor is zero and the free Maxwellon pressure has been lowered from
k
B
T to
2
3
k
B
T as for the gravitational case. This change in pressure indi-
cates that the system of noninteracling Maxwellons has been compressed
by the external eld into a quasi-two-dimensional system but, unlike the
gravitational eld case, without changing the energy. Since s 0, the usual
free-particle result requires 12.18
3
t
. With the external eld present,
s 0 requires 24.36/L
z

3
t
. This requires the system to be more dilute
(at xed T) or at higher temperature (at xed ) than without the external
eld as was found for the external gravitational eld case.
In the opposite limit L
z
<< 1,
f = k
B
T
_
ln(
3
t
) 1

+
3
4
k
B
TL
z
_
1
11
24
L
z
+
_
(7.27)
= k
B
T ln(
3
t
) +
3
4
k
B
TL
z
_
1
11
24
L
z
+
_
(7.28)
s = k
B
_
5
2
ln(
3
t
)
_

3
4
k
B
L
z
_
1
11
24
L
z
+
_
(7.29)
=
3
2
k
B
T (7.30)
P = k
B
T (7.31)

xx
=
yy
= k
B
T (7.32)
37

zz
= k
B
T
3
4
k
B
L
z
_
1
11
12
L
z
+
_
(7.33)
P
th
= k
B
T
1
4
k
B
L
z
_
1
11
12
L
z
+
_
(7.34)
As for the gravitational coupling case, the choices for zero point of the exter-
nal potential may be considered to be related by Legendre transformations.
From the microscopic viewpoint, it should be noted that the exact local
particle density, obtained from the denition of the single-particle distribu-
tion function (5.35), and the exact local energy-density prole [17, 20] for
the noninteracting system are given by

loc
(r) =
loc
(0)
_
1 +U
ext
(r)

3/2

loc
(r
1
) =
1
Z
can
_
d
3
p
1
d
3
p
N
_
d
3
r
2
d
3
r
N
H e
H
(7.35)

loc
(r) =
loc
(0)
_
1 +U
ext
(r)

3/2
Due to the form for the local particle density and energy density (7.35), the
local momentum current or pressure tensor is simply
P
ij
(r) = 2
loc
(r)
_
1 +U
ext
(r)

ij
+C
ij
(7.36)
The C
ij
are the nonlocal contributions from the external eld and are re-
quired to obtain the correct thermodynamic pressure. The diagonal elements
are determined by comparing the volume average of (7.36) with the ther-
modynamically dened
ij
in (7.7). This clearly shows that the equation
of state for noninteracting particles in an external temperature eld is
neither of the local equilibrium form nor the modied form for the particle-
density coupling discussed in Secs. III and V. For the present noninteracting
case, the form for (7.36) is
P
loc
(r) = P
0
(
loc
(r), T)
_
1 +U
ext
(r)

+f
ext
(, T, L
x
, L
y
, L
z
) (7.37)
38
where P
0
(
loc
(r), T) is the local equilibrium equation of state, T is the con-
stant thermodynamic temperature, is the average density of the system,
and all the nonlocal eects of the external coupling are coniained in the
function f
ext
. It is also of interest to note that the equation of state for the
grand potential suggests that
P
loc
(r) =
loc
(r)k
B
T = P
0
(
loc
(r), T) (7.38)
As for the gravitational coupling case, the local equilibrium equation of state
yields the correct grand potential, but is not sucient to give the correct
thermodynamic pressure.
Since the usual free-particle energy equation of state (7.13) remains valid
in this case, the conditions
=
1
V
_
d
3
r
loc
(r) =
3
2
k
B
T
(7.39)
=
1
V
_
d
3
r
loc
(r)
together with the forms for the local particle density and local energy density
in (7.35) imply that

loc
(0) =
3
2

loc
(0)k
B
T (7.40)
Consequently, the local energy-density equation of state is simply

loc
(r) =
3
2

loc
(r)k
B
T (7.41)
A local temperature eld may he dened in terms of the local kinetic energy
per particle,
T
loc
(r) =
2
3

loc
(r)
k
B
=

loc
(r)

T (7.42)
39
This local mechanically dened temperature has the property that its
volume average is given by the thermodynamically dened temperature T.
The gradient of this local temperature is given by
T
loc
(r) =
T

loc
(r)
(7.43)
T
loc
(r) =
3
2
T

loc
(r)

U
ext
(r)
1 +U
ext
(r)
The rst equation in (7.43) reects the conventional wisdom, mentioned in
Sec. IV, that density gradients are not independent of temperature gradi-
ents [15]. That is, a density gradient will be produced by establishing a
temperature gradient. The second equation in (7.43) makes a connection
with the linear response theory interpretation of U
ext
(r) as a tempera-
ture gradient [3, 4]. It should he pointed out that the basis for this local
temperature is similar to the basis for the local pressure discussed above in
Secs. III and IV. That is, the local quantity is given a microscopic mechani-
cal denition such that the volume average of that local quantity is equal to
the corresponding thermodynamically dened variable. Strictly speaking, a
local thermodynamic variable was not dened.
VIII Conclusions
It was shown that an external scalar eld has two main eects on the
equilibrium properties of a many-particle assembly. First, the external eld
breaks the usual translational invariance of the system and the usual exten-
sivity arguments do not hold, in general, in the presence of an external eld.
Two intensive parameters may be identied by a choice of independent vari-
ables (for instance, T and ), but other quantities such as the pressure, free
energy density, entropy density, etc., may no longer be intensive, that is,
they may show a size dependence that does not vanish in the limit of a geo-
metrically large system. Second, the external eld will select a direction in
space as special in some way which breaks the usual rotational invariance of
the bulk system and this is reected in an anisotropy in the thermodynamic
relations. External strains show up in the Gibbs-Duhem relation as well as
40
the thermodynamics for unit volume quantities (E/V, F/V , etc.) which are
shape-dependent work terms. Specic examples of extensivity breaking and
anisotropy were seen in the case of noninteracting particies in an external
gravitational eld and an externally imposed temperature eld. Another
aspect of anisotropy in the thermodynamics was illustrated in the case of
the phase transition of a van der Waals gas in an external gravitational
eld. In addition, the form of the local equilibrium assumption, Eq. (1.3),
for the particle-density coupling was shown to be incorrect in principle due
to an inappropriate identication of the thermodynamic pressure with the
negative of the grand canonical potential per unit volume. This was also
illustrated for the case of noninteracting particles in an external gravita-
tional eld. The condition of local equilibrium is valid, but a purely local
equilibrium equation of state for the pressure is not correct. Nonlocal con-
tributions from the external eld are required to obtain the correct equation
of state for a system of particles in an external eld. A local equilibrium
equation of state or a nonlocal modication of a local equilibrium equation
of state does not appear to be sucient for the energy-density coupling. For
this system, the local equilibrium equation of state requires local and non-
local modication. The externally imposed temperature eld seems to have
a more profound eect on the thermodynamic structure of a system than
an externally imposed mechanical constraint like an external gravitational
eld.
Acknowledgments
I would like to thank Dr. Raymond D. Mountain of the Thermophysics
Division of the National Institute of Standards and Technology, Gaithers-
burg and Professor Robert D. Pu of the Department of Physics at The
University of Washington for reading a preliminary version of this paper and
for their many helpful and clarifying comments. This work was supported
by the NASA Lewis Research Center under Contract No. NAS3-25266 with
R. Allen Wilkinson as monitor.
Erratum
In Secs. III and IV, some equations were inappropriately identied as
equilibrium conditions when, in fact, they should be identied as steady-
41
state equations. The equilibrium label may not be incorrect but it can be
misleading. While the time-independent momentum equations, (3.11) and
(4.12), may be considered to be both steady-state and equilibrium equa-
tions, the time-independent continuity equation [conservation of particles,
Eq. (4.11)], and the time-independent conservation of energy equations,
(3.18) and (4.13), should be considered as steady-state equations. The equi-
librium solutions of these equations may then be dened as those with zero
particle current and zero energy current.
References
[1] J. E. Mayer and M. G. Mayer, Statistical Mechanics (Wiley, New York.
1940).
[2] S. K. Ma, Modern Theory of Critical Phenomena (Ben-
jamin/Cummings, Reading MA, 1976).
[3] L. P. Kadano and P. C. Martin, Ann Phys. (N.Y.), 24, 419 (1963).
[4] R. D. Pu and N. S. Giillis, Ann. Phys. (N.Y.), 46, 364 (1968).
[5] D. Forster, Hydrodynamic Fluctuations, Broken Symmetry, and Corre-
lation Functions (Benjamin/Cummings, Reading MA, 1983).
[6] F. F. Abraham, Phys. Rep., 53, 93 (1979).
[7] R. Evans, Adv. Phys., 28, 143 (1979).
[8] B. Widom, in Phase Transitions and Critical Phenomena, edited by C.
Domb and M. S. Green (Academic, New York, 1972), Vol. 2.
[9] A. L. Fetter and J. D. Walceka, Quantum Theory of Many-Particle
Systems (McGraw-Hill, New York, 1971).
[10] L. P. Kadano and G. Baym, Quantum Statistical Mechanics: Greens
Function Methods in Equilibrium and Nonequilibrium Problems (Ben-
jamin/Cummings, Reading MA, 1978).
[11] L. D. Landau and E. M. Lifshitz, Statistical Physics (Pergamon, Ox-
ford, 1980), Part I.
[12] D. E. Wolf and L. Tang, Phys. Rev. A, 36, 5337 (1987).
[13] R E. Salvino and R. D. Pu, Phys Rev. B, 34, 6351 (1986).
42
[14] The linear response version of Eq. (4.4) and a model-dependent state-
ment corresponding to (4.5) may be found in Ref. [4].
[15] L. D. Landau and E. M. Lifshitz, Fluid Mechanics (Pergamon, Oxford.
1987).
[16] B. G. Levich, Theoretical Physics, Vol. 2: Statistical Physics and Elec-
tromagnetic Processes in Matter (North-Holland, Amsterdam, 1970).
[17] D. Chandler, Introduction to Modern Statistical Mechanics (Oxford,
New York, 1987).
[18] J. P. Hansen and I. R. McDonald, Theory oj Simple Liquids (Academic,
London, 1986). For a density functional approach to nonuniform sys-
tems, sec Refs. 6 and 7.
[19] M. R. Moldover, J. V. Sengers, R. W. Gammon, and R. J. Hocken,
Rev. Mod. Phys., 51, 79 (1979).
[20] G. E. Uhlenbeck and G. W. Ford, Lectures in Statistical Mechanics
(American Mathematical Society, Providence, 1963).
43

Das könnte Ihnen auch gefallen