a

r
X
i
v
:
h
e
p
-
t
h
/
9
9
0
1
0
1
1
v
1


4

J
a
n

1
9
9
9
OKHEP9901
The Casimir Eect:
Physical Manifestations of Zero-Point Energy

KIMBALL A. MILTON
Department of Physics and Astronomy,
The University of Oklahoma
Norman, OK 73019-0225 USA
December 16, 2010
Abstract
Zero-point uctuations in quantum elds give rise to observable forces be-
tween material bodies, the so-called Casimir forces. In these lectures I present
the theory of the Casimir eect, primarily formulated in terms of Greens func-
tions. There is an intimate relation between the Casimir eect and van der
Waals forces. Applications to conductors and dielectric bodies of various shapes
will be given for the cases of scalar, electromagnetic, and fermionic elds. The
dimensional dependence of the eect will be described. Finally, we ask the ques-
tion: Is there a connection between the Casimir eect and the phenomenon of
sonoluminescence?
1 Introduction
The Casimir eect is a manifestation of macroscopic quantum eld theory. Its origin
is intimately tied to the van der Waals forces between neutral molecules. In 1873 van
der Waals [1] introduced these weak intermolecular forces to explain derivations from
the ideal gas laws, but their physical basis did not begin to emerge until 1930 when
London [2] showed how the leading behavior of this force could be understood from
quantum mechanics. That is, the Hamiltonian describing the interaction between two
dipoles separated by a distance R is
H
int
=
d
1
d
2
3d
1
nd
2
n
R
3
, (1.1)

Invited Lectures at the 17th Symposium on Theoretical Physics, Seoul National University,
Korea, June 29July 1, 1998.
1
where n is the relative direction of the two dipoles. Because d
1
) = d
2
) = 0, this
Hamiltonian has vanishing expectation value, H
int
) = 0; however, in second order of
perturbation theory this operator appears squared, so there is an eective interaction
potential between the dipoles,
V
e

1
R
6
. (1.2)
This is a short distance electrostatic eect. The next step in the saga occurred in 1948
when Casimir and Polder [3] introduced retardation, resulting in a potential which
for large distances falls o faster,
V
e

1
R
7
. (1.3)
This result can be understood by dimensional considerations. For suciently weak
elds, the polarization of a molecule should be proportional to the electric eld acting
on the molecule,
d = E, (1.4)
where is the polarizability of the molecule. Each atom produces a dipole eld, so the
two atoms polarize each other. In terms of a unit of length L, we have dimensionally
[d] = eL, [E] = eL
2
, so [] = L
3
. (1.5)
Thus we have, at zero temperature,
V
e


1

2
R
6
hc
R
. (1.6)
[It may be useful to remember the conversion factor hc 2 10
5
eV cm.] The high
temperature limit is classical,
V
e


1

2
R
6
kT, T . (1.7)
Here we have thought of the interactions from an action-at-a-distance point of
view. From this viewpoint, the Casimir eect is simply a macroscopic manifestation
of the sum of a multitude of molecular van der Waals interactions. However, shortly
after the original Casimir and Polder paper, Bohr suggested to Casimir that zero-point
energy was responsible for the intermolecular force, and indeed that is an equivalent
viewpoint [4]. One can shift the emphasis from action at a distance between molecules
to local action of elds.
The connection between the sum of the zero-point energies of the modes and the
vacuum expectation value of the eld energy may be easily given. Let us regulate the
former with an oscillating exponential:
1
2

a
h
a
e
ia
=
h
2
_

d
2i
e
i

a
2

2
a

2
i
, 0, (1.8)
2
where a labels the modes, and the contour of integration in the second form may be
closed in the lower half plane. For simplicity of notation let us suppose we are dealing
with massless scalar modes, for which the eigenfunctions and eigenvalues satisfy

a
=
2
a

a
. (1.9)
Because these are presumed normalized, we may write the second form in Eq. (1.8)
as
h
_
d
2i

2
e
i
_
(dx)

a
(x)

a
(x)

2
a

2
i
=
h
i
_
(dx)
_
d
2

2
G(x, x; )e
i(tt

tt

=
_
(dx)
0

0
(x)(x

))

x
, (1.10)
where the Greens function G(x, t; x

, t

) satises
_

2
+

2
t
2
_
G(x, t; x

, t

) = (x x

)(t t

), (1.11)
and is related to the vacuum expectation value of the time-ordered product of elds
according to
G(x, t; x

, t

) =
i
h
T(x, t)(x

, t

)). (1.12)
For a massless scalar eld, the canonical energy-momentum tensor is
T

1
2
g

. (1.13)
The second term involving the Lagrangian in Eq. (1.13) may be easily shown not
to contribute when integrated over all space, by virtue of the equation of motion,

2
= 0 outside the sources, so we have the result, identifying the zero-point energy
with the vacuum expectation value of the eld energy,
1
2

a
h
a
=
_
(dx)T
00
(x)). (1.14)
In the vacuum this is divergent and meaningless. What is observable is the change
in the zero-point energy when matter is introduced. In this way we can calculate the
Casimir forces.
Alternatively, we can calculate the stress on the material bodies. Figure 1 shows
the original geometry considered by Casimir, where he calculated the quantum uc-
tuation force between parallel, perfectly conducting plates [5]. The force per unit
3
-
a
-
z
Figure 1: Geometry of parallel conducting plates.
area f on one of the plates is given in terms of the normal-normal component of the
stress tensor,
f =
_
dxdy T
zz
), (1.15)
where the integral is over the surface area of the plate in question. For electromagnetic
elds, the relevant stress tensor component is
T
zz
=
1
2
(H
2

H
2
z
+ E
2

E
2
z
). (1.16)
We impose classical boundary conditions on the surfaces,
H
z
= 0, E

= 0, (1.17)
and the calculation of the vacuum expectation value of the eld components reduces
to nding the classical TE and TM Greens functions. In general, one further has to
subtract o the stress that the formalism would give if the plates were not present,
the so-called volume stress, and then the result of a simple calculation, which is given
below, is
f =
_
dxdy [T
zz
T
zz
(vol)] =

2
240a
4
hc, (1.18)
an attractive force.
The dependence on a is, of course, completely determined by dimensional consid-
erations. Numerically, the result is quite small,
f = 8.11 MeVfma
4
= 1.30 10
27
Nm
2
a
4
, (1.19)
and will be overwhelmed by electrostatic repulsion between the plates if each plate
has an excess electron density n greater than 1/a
2
, from which it is clear that the
experiment must be performed at the m level. Nevertheless, it has been directly
measured to an accuracy of several percent [6, 7, 8, 9, 10, 11, 12, 13, 14, 15]. (The cited
4
measurements include insulators as well as conducting surfaces; the corresponding
theory will be given in Section 3.) Until recently, the most convincing experimental
evidence comes from the study of thin helium lms [16]; there the corresponding
Lifshitz theory [17, 18] has been conrmed over nearly 5 orders of magnitude in
the van der Waals potential (nearly two orders of magnitude in distance). Quite
recently, the Casimir eect between conductors has been conrmed to the 5% level
by Lamoreaux [19], and to 1% by Mohideen and Roy [20].
2 Casimir Force Between Parallel Plates
2.1 Dimensional Regularization
We begin by presenting a simple, modern, derivation of the Casimir eect in its
original context, the electromagnetic force between parallel, uncharged, perfectly con-
ducting plates. No attempt at rigor will be given, for the same formul will be derived
by a consistent Greens function technique in the following subsection. Nevertheless,
the procedure illustrated here correctly produces the nite, observable force starting
from a divergent formal expression, without any explicit subtractions, and is therefore
of great utility in practice.
For simplicity we consider a massless scalar eld conned between two parallel
plates separated by a distance a. (See Fig. 1.) Assume the eld satises Dirichlet
boundary conditions on the plates, that is
(0) = (a) = 0. (2.1)
The Casimir force between the plates results from the zero-point energy per unit
transverse area
u =
1
2

h =
1
2

n=1
_
d
2
k
(2)
2
_
k
2
+
n
2

2
a
2
, (2.2)
where we have set h = c = 1, and introduced normal modes labeled by the positive
integer n and the transverse momentum k.
To evaluate Eq. (2.2) we employ dimensional regularization. That is, we let the
transverse dimension be d, which we will subsequently treat as a continuous, complex
variable. It is also convenient to employ the Schwinger proper-time representation
for the square root:
u =
1
2

n
_
d
d
k
(2)
d
_

0
dt
t
t
1/2
e
t(k
2
+n
2

2
/a
2
)
1
(
1
2
)
, (2.3)
where we have used the Euler representation for the gamma function. We next carry
out the Gaussian integration over k:
u =
1
4

1
(4)
d/2

n
_

0
dt
t
t
1/2d/2
e
tn
2

2
/a
2
. (2.4)
5
Finally, we again use the Euler representation, and carry out the sum over n by use
of the denition of the Riemann zeta function:
u =
1
4

1
(4)
d/2
_

a
_
d+1

d + 1
2
_
(d 1). (2.5)
When d is an odd integer, this expression is indeterminate, but we can use the reec-
tion property

_
z
2
_
(z)
z/2
=
_
1 z
2
_
(1 z)
(z1)/2
(2.6)
to rewrite (2.5) as
u =
1
2
d+2

d/2+1
1
a
d+1

_
1 +
d
2
_
(2 + d). (2.7)
We note that analytic continuation in d is involved here: Eq. (2.4) is only valid if
Re d < 1 and the subsequent denition of the zeta function is only valid if Re d < 2.
In the physical applications, d is a positive integer.
We evaluate this general result (2.7) at d = 2. This gives for the energy per unit
area in the transverse direction
u =

2
1440
1
a
3
, (2.8)
where we have recalled that (4) =
4
/90. The force per unit area between the plates
is obtained by taking the negative derivative of u with respect to a:
f
s
=

a
u =

2
480
1
a
4
. (2.9)
The above result (2.9) represents the Casimir force due to a scalar eld. It is
tempting (and, in this case, is correct) to suppose that to obtain the force due to
electromagnetic eld uctuations between parallel conducting plates, we simply mul-
tiply by a factor of 2 to account for the two polarization states of the photon. Doing
so reproduces the classic result of Casimir [5]:
f
em
=

2
240
1
a
4
. (2.10)
A rigorous derivation of this result will be given in Sec. 2.5.
2.2 Scalar Greens Function
We now rederive the result of Sec. 2.1 by a physical and rigorous Greens function
approach. The equation of motion of a massless scalar eld produced by a source
K is

2
= K, (2.11)
6
from which we deduce the equation satised by the corresponding Greens function

2
G(x, x

) = (x x

). (2.12)
For the geometry shown in Fig. 1, we introduce a reduced Greens function g(z, z

)
according to the Fourier transformation
G(x, x

) =
_
d
d
k
(2)
d
e
ik(xx

)
_
d
2
e
i(tt

)
g(z, z

), (2.13)
where we have suppressed the dependence of g on k and , and have allowed z on the
right hand side to represent the coordinate perpendicular to the plates. The reduced
Greens function satises
_

2
z
2

2
_
g(z, z

) = (z z

), (2.14)
where
2
=
2
k
2
. Equation (2.14) is to be solved subject to the boundary conditions
(2.1), or
g(0, z

) = g(a, z

) = 0. (2.15)
We solve (2.14) by the standard discontinuity method. The form of the solution
is
g(z, z

) =
_
Asin z, 0 < z < z

< a,
Bsin (z a), a > z > z

> 0,
(2.16)
which makes use of the boundary condition (2.15) on the plates. According to
Eq. (2.14), g is continuous at z = z

, but its derivative has a discontinuity:

Asin z

Bsin (z

a) = 0, (2.17a)
A cos z

B cos (z

a) = 1. (2.17b)
The solution to this system of equations is
A =
1

sin (z

a)
sin a
, (2.18a)
B =
1

sin z

sin a
, (2.18b)
which implies that the reduced Greens function is
g(z, z

) =
1
sin a
sin z
<
sin (z
>
a), (2.19)
where z
>
(z
<
) is the greater (lesser) of z and z

.
7
From knowledge of the Greens function we can calculate the force on the bounding
surfaces from the energy-momentum or stress tensor. For a scalar eld, the stress
tensor
1
is given by Eq. (1.13). What we require is the vacuum expectation value of
T

which can be obtained from the Greens function according to Eq. (1.12), or
(x)(x

)) =
1
i
G(x, x

), (2.20)
a time-ordered product being understood in the vacuum expectation value. By virtue
of the boundary condition (2.1) we compute the normal-normal component of the
Fourier transform of the stress tensor on the boundaries to be
T
zz
) =
1
2i

z
g(z, z

zz

=0,a
=
i
2
cot a. (2.21)
We now must integrate on the transverse momentum and the frequency to get the
force per unit area. The latter integral is best done by performing a complex frequency
rotation,
i, i
_
k
2
+
2
i. (2.22)
Thus, the force per unit area is given by
f =
1
2
_
d
d
k
(2)
d
_
d
2
coth a. (2.23)
This integral does not exist.
What we do now is regard the right boundary at z = a, for example, to be a
perfect conductor of innitesimal thickness, and consider the ux of momentum to
the right of that surface. To do this, we nd the Greens function which vanishes at
z = a, and has outgoing boundary conditions as z , e
ikz
. A calculation just
like that which led to Eq. (2.19) yields for z, z

> a,
g(z, z

) =
1

_
sin (z
<
a)e
i(z>a)

. (2.24)
(This obviously has the correct discontinuity

z
g[
z

+0
z

0
= 1 and satises the appro-
priate boundary conditions.) The corresponding normal-normal component of the
stress tensor at z = a is
T
zz
)

z=z

=a
=
1
2i

z
g(z, z

z=z

=a
=

2
. (2.25)
1
The ambiguity in dening the stress tensor has no eect. We can add to T

an arbitrary
multiple of (

2
)
2
[21, 22]. But the zz component of this tensor on the surface vanishes
by virtue of Eq. (2.1).
8
So, from the discontinuity in T
zz
, that is, the dierence between (2.21) and (2.25), we
nd the force per unit area on the conducting surface:
f =
1
2
_
d
d
k
(2)
d
_
d
2
(coth a 1). (2.26)
We evaluate this integral using polar coordinates:
f =
A
d+1
(2)
d+1
_

0

d
d

e
2a
1
. (2.27)
Here A
n
is the surface area of a unit sphere in n dimensions, which is most easily
found by integrating the multiple Gaussian integral
_

d
n
xe
x
2
=
n/2
(2.28)
in polar coordinates. The result is
A
n
=
2
n/2
(n/2)
. (2.29)
When we substitute this into (2.27), employ the integral
_

0
dy
y
s1
e
y
1
= (s)(s), (2.30)
and use the identity
(2z) = (2)
1/2
2
2z1/2
(z)
_
z +
1
2
_
, (2.31)
we nd for the force per unit transverse area
f = (d + 1)2
d2

d/21

_
1 +
d
2
_
(d + 2)
a
d+2
. (2.32)
Evidently, Eq. (2.32) is the negative derivative of the Casimir energy (2.7) with respect
to the separation between the plates:
f =
u
a
; (2.33)
this result has now been obtained by a completely well-dened approach. The force
per unit area, Eq. (2.32), is plotted in Fig. 2, where a 2a and d = D 1. This
result was rst derived by Ambjrn and Wolfram [23].
9
-5.0 -3.0 -1.0 1.0 3.0 5.0
D
-15.0
-10.0
-5.0
0.0
5.0
10.0
15.0
20.0
25.0
f
a
D
+
1
Figure 2: A plot of the Casimir force per unit area f in Eq. (2.32) for 5 < D < 5
for the case of a slab geometry (two parallel plates). Here D = d + 1.
We can also derive the same result by computing the energy from the energy-
momentum tensor.
2
The relevant component is
3
T
00
=
1
2
(
0

0
+
1

1
+
2

2
+
3

3
), (2.34)
so when the vacuum expectation value is taken, we nd from Eq. (2.19)
T
00
) =
1
2i
1
sin a
[(
2
+ k
2
) sin z sin (z a)
+
2
cos z cos (z a)]
=
1
2i sin a
[
2
cos a k
2
cos (2z a)]. (2.35)
We now must integrate this over z to nd the energy per area between the plates.
Integration of the second term in Eq. (2.35) gives a constant, independent of a, which
will not contribute to the force. The rst term gives
_
a
0
dxT
00
) =

2
a
2i
cot a. (2.36)
2
Again, the ambiguity in the stress tensor is without eect, because the extra term here is
2

2
,
which upon integration over space becomes a vanishing surface integral.
3
As noted after Eq. (1.13), we would get the same integrated energy if we dropped the second,
Lagrangian, term in T
00
there, that is, used T
00
=
0

0
.
10
As above, we now integrate over and k, after we perform the complex frequency
rotation. We obtain
u =
a
2
_
d
d
k
(2)
d
_
d
2

coth a. (2.37)
If we introduce polar coordinates so that = cos , we see that this diers from
Eq. (2.23) by the factor of acos
2
). Here
cos
2
) =
_

0
cos
2
sin
d1
d
_

0
sin
d1
d
=
1
d + 1
, (2.38)
which uses the integral
_

0
sin
d1
d =
_
1
1
(1 x
2
)
(d2)/2
dx = 2
d1

_
d
2
_
2
(d)
. (2.39)
Thus, we again recover Eq. (2.7).
For the sake of completeness, we note that it is also possible to use the eigenfunc-
tion expansion for the reduced Greens function. That expansion is
g(z, z

) =
2
a

n=1
sin(nz/a) sin(nz

/a)
n
2

2
/a
2

2
. (2.40)
When we insert this into the stress tensor we encounter

z
g(z, z

)[
z=z

=0,a
=
2
a

n=1
n
2

2
/a
2
n
2

2
/a
2

2
. (2.41)
We subtract and add
2
to the numerator of this divergent sum, and omit the diver-
gent part, which is simply a constant in . Such terms correspond to functions in
space and time (contact terms), and should be omitted, since we are considering the
limit as the space-time points coincide. We evaluate the resulting nite sum by use
of the following expression for the cotangent:
cot x =
1
x
+
2x

k=1
1
x
2
k
2
. (2.42)
So in place of Eq. (2.21) we obtain
T
zz
) =
i
2

_
cot a
1
a
_
, (2.43)
which agrees with Eq. (2.21) apart from an additional contact term.
11
2.3 Massive Scalar
It is easy to modify the discussion of Sec. 2.2 to include a mass for the scalar eld.
The reduced Greens function now satises the equation
_

2
z
2

2
+
2
_
g(z, z

) = (z z

), (2.44)
instead of Eq. (2.14), so the reduced Greens function between the plates is
g(z, z

) =
1

sin z
<
sin (z
>
a)
sin a
(2.45)
where

2
=
2

2
. (2.46)
The calculation proceeds just as in Sec. 2.2, and we nd, in place of Eq. (2.27)
f =
A
d+1
(2)
d+1
_

0

d
d
_

2
+
2
e
2a

2
+
2
1
. (2.47)
When we substitute the value of A
d+1
given by Eq. (2.29), and introduce a dimen-
sionless integration variable, we nd for the force per unit area
f =
1
2
2(d+1)

(d+1)/2

_
d+1
2
_
a
d+2
_

2a
dxx
2
(x
2
4
2
a
2
)
(d1)/2
e
x
1
. (2.48)
For d = 2 this function is plotted in Fig. 3. The massive scalar was rst treated
by Hays in two dimensions [24], and again done in general number of dimensions by
Ambjrn and Wolfram [23].
2.4 Finite Temperature
We next turn to a consideration of the Casimir eect at nonzero temperature. In
this case, uctuations arise not only from quantum mechanics but from thermal
eects. In fact, as we will shortly see, the high-temperature limit is a purely classical
phenomenon.
Formally, we can easily obtain the expression for the Casimir force between parallel
plates at nonzero temperature. In Eq. (2.26) we replace the imaginary frequency by

n
=
2n

, (2.49)
where = 1/kT, T being the absolute temperature. Correspondingly, the frequency
integral is replaced by a sum on the integer n:
_
d
2
=
1

n=
. (2.50)
12
0.0 1.0 2.0 3.0 4.0 5.0
a
0.000
0.010
0.020
-
f
a
4
Figure 3: Scalar Casimir force between parallel plates as a function of mass for d = 2.
Thus, Eq. (2.26) is replaced by
f
T
=
1
2
_
d
d
k
(2)
d

n=
2
n
e
2na
1
, (2.51)
where
n
=
_
k
2
+ (2n/)
2
.
We rst consider the high-temperature limit. When T ( 0), apart
from exponentially small corrections, the contribution comes from the n = 0 term in
the sum in Eq. (2.51). That integral is easily worked out in polar coordinates using
Eqs. (2.29) and (2.30). The result is
f
T
kT
d
(2

a)
d+1

_
d + 1
2
_
(d + 1). (2.52)
In particular, for two and three dimensions, d = 1 and d = 2, respectively,
d = 1 : f
T
kT

24a
2
, (2.53a)
d = 2 : f
T
kT
(3)
8a
3
. (2.53b)
This high-temperature limit should be classical. Indeed, we can derive this same
result from the classical limit of statistical mechanics. The Helmholtz free energy for
massless bosons is
F = kT ln Z, ln Z =

i
ln(1 e
p
i
), (2.54)
13
from which the pressure on the plates can be obtained by dierentiation:
p =
F
V
. (2.55)
We make the momentum-space sum explicit for our d + 1 spatial geometry:
F = kTV
_
d
d
k
(2)
d+1

n=
ln(1 exp(
_
k
2
+ n
2

2
/a
2
)). (2.56)
Now, for high temperature, 0, we expand the exponential, and keep the rst
order term in . We can write the result as
F = V kT
1
2a
d
ds
_
d
d
k
(2)
d

n=
1
2

2s
_
n
2

2
a
2
+ k
2
_
s

s=0
, (2.57)
where we have used the identity
ln =
d
ds

s=0
. (2.58)
This trick allows us to proceed as in Sec. 2.1. After the k integration is done, the s
derivative acts only on 1/(s):
d
ds
1
(s)

s=0
= 1, (2.59)
so we easily nd the result from Eq. (2.6)
F = V
kT
(2a

)
d+1
(d + 1)
_
1 + d
2
_
. (2.60)
The pressure, the force per unit area on the plates, is obtained by applying the
following dierential operator to the free energy:

V
=
1
A

a
, (2.61)
where V = Aa, A being the d-dimensional area of the plates. The result of this
operation coincides with Eq. (2.52).
The low-temperature limit (T 0 or ) is more complicated because f
T
is
not analytic at T = 0. The most convenient way to proceed is to resum the series in
Eq. (2.51) by means of the Poisson sum formula, which says that if c() is the Fourier
transform of b(x),
c() =
1
2
_

b(x)e
ix
dx, (2.62)
14
then the following identity holds:

n=
b(n) = 2

n=
c(2n). (2.63)
Here, we take
b(n) =
_
d
d
k
(2)
d
2
n
e
2na
1
. (2.64)
Introducing polar coordinates for k, changing from k to the dimensionless integration
variable z = 2a
x
, and interchanging the order of x and z integration, we nd for the
Fourier transform
c() =
A
d
2
2d

d+1
a
d+1
_

0
dz z
2
e
z
1
_
z/4a
0
dx cos x
_
z
2

_
4ax

_
2
_
(d2)/2
. (2.65)
The x integral in Eq. (2.65) is easily expressed in terms of a Bessel function: It is [25]
z
d1
4a
_
1
0
du cos
_
z
4a
u
_
(1 u
2
)
(d2)/2
=
(1d)/2
z
(d1)/2
_
8a

_
(d3)/2

_
d
2
_
J
(d1)/2
_
z
4a
_
. (2.66)
We thus encounter the z integral
I(s) =
_

0
dz z
(d+3)/2
e
z
1
J
(d1)/2
(sz), (2.67)
where s = /4a.
The zero-temperature limit comes entirely from = 0:
f
T=0
=

c(0). (2.68)
So we require the small-x behavior of the Bessel function,
J
(d1)/2
(x)
_
x
2
_
(d1)/2
1

_
d+1
2
_, x 0, (2.69)
whence using Eq. (2.30)
I(s)
_
s
2
_
(d1)/2
2
d+1

d + 1
2

_
d
2
+ 1
_
(d + 2), s 0. (2.70)
15
Inserting this into Eq. (2.65) we immediately recover the zero-temperature result
(2.32).
We now seek the leading correction to this. We rewrite I(s) as
I(s) =
1
s
(d+5)/2
_

0
dy y
(d+3)/2
1 e
y/s
e
y/s
J
(d1)/2
(y)
=
1
s
(d+5)/2
_

0
dy y
(d+3)/2
e
y/s

n=0
e
ny/s
J
(d1)/2
(y), (2.71)
where we have employed the geometric series. The Bessel-function integral has an
elementary form [25]:
_

0
dy J
q
(y)e
by
=
(

b
2
+ 1 b)
q

b
2
+ 1
(2.72)
is the fundamental integral, and the form we want can be written as
_

0
dy J
q
(y)y
p
e
by
= (1)
p
_
d
db
_
p
(

b
2
+ 1 b)
q

b
2
+ 1
, (2.73)
provided p is a nonnegative integer. (For the application here, this means d is odd,
but we will be able to analytically continue the nal result to arbitrary d.) Then we
can write I(s) in terms of the series
I(s) =
(1)
(d+3)/2
s
(d+5)/2

l=1
F(l), l = n + 1, (2.74)
where
F(l) =
_
d
dl/s
_
(d+3)/2
_
_
_
l
s
_
2
+ 1
l
s
_
(d1)/2
_
_
l
s
_
2
+ 1
. (2.75)
We evaluate the sum in Eq. (2.74) by means of the Euler-Maclaurin summation
formula, which has the following formal expression:

l=1
F(l) =
_

1
dl F(l) +
1
2
[F() + F(1)]
+

k=1
1
(2k)!
B
2k
[F
(2k1)
() F
(2k1)
(1)]. (2.76)
Here, B
n
represents the nth Bernoulli number. Since we are considering ,= 0
(the = 0 term was dealt with in the previous paragraph), the low-temperature
16
limit corresponds to the limit s . It is easy then to see that F() and all its
derivatives there vanish. The function F at 1 has the general form
F(1) =
_
d
d
_
(d+3)/2
_

2
+ 1

(d1)/2

2
+ 1
, = 1/s 0. (2.77)
By examining the various possibilities for odd d, d = 1, d = 3, d = 5, and so on, we
nd the result
F(1) = (1)
(d1)/2
d!! = (1)
(d1)/2
2
(d+1)/2

1/2

_
d
2
+ 1
_
. (2.78)
Because it is easily seen that before the limit 0 is taken, F(1) is an even function
of , it follows that the odd derivatives of F evaluated at 1 that appear in Eq. (2.76)
vanish. Finally, the integral in Eq. (2.76) is that of a total derivative:
_

1
dl F(l) =
1

_
d
d
_
(d+1)/2
_

2
+ 1

(d1)/2

2
+ 1
= F(1). (2.79)
Thus, the nal expression for I(s) is
I(s)
1
s
(d+5)/2
2
(d1)/2

1/2

_
d
2
+ 1
_
. (2.80)
Note that a choice of analytic continuation to d other than an odd integer has been
made so as to avoid oscillatory behavior in d.
We return to Eq. (2.65). It may be written as
c() =
2
d/2
(d/2)
1
2
2d

d+1
a
d+1

(1d)/2
_
8a

_
(d3)/2

(d/2)I
_

4a
_
= 2
d1

d/21
_
4

_
d+1

_
d
2
+ 1
_
1

d+2
. (2.81)
The correction to the zero-temperature result Eq. (2.32) is obtained from
f
T0
corr
=
2

n=1
c(2n)
=
d/21
(d/2 + 1)(d + 2)
d2
. (2.82)
Thus, the force per unit area in the low-temperature limit has the form
f
T0

(d + 1)2
d2

d/21
a
d+2
(d/2 + 1)(d + 2)
_
1 +
1
d + 1
_
2a

_
d+2
_
, (2.83)
17
of which the d = 1 and d = 2 cases are familiar:
d = 1 : f
T0

1
8a
3
(3)
_
1 + 4
a
3

3
_
, (2.84a)
d = 2 : f
T0


2
480a
4
_
1 +
16
3
a
4

4
_
. (2.84b)
These equations are incomplete in that they omit exponentially small terms; for
example, in the last square bracket, we should add the term

240

e
/a
. (2.85)
Finite temperatures were rst discussed by Lifshitz [17], but then considered more
fully by Fierz, Sauer, and Mehra [26, 27, 28]. Hargreaves [29] analyzed the discrepancy
between the results of Refs. [17] and [27].
2.5 Electromagnetic Casimir Force
We now turn to the situation originally treated by Casimir: the force between parallel
conducting plates due to quantum uctuations in the electromagnetic eld. An ele-
gant procedure, which can be applied to much more complicated geometries, involves
the introduction of the Greens dyadic, dened as the response function between the
(classical) electromagnetic eld and the polarization source (this formalism is intro-
duced in Refs. [30, 31]):
E(x) =
_
(dx

) (x, x

) P(x

). (2.86)
In the following we will use the Fourier transform of in frequency:
(x, x

) =
_
d
2
e
i(tt

)
(r, r

; ), (2.87)
which satises Maxwells equations
= i, (2.88a)
i = i1(r r

). (2.88b)
The second Greens dyadic appearing here is solenoidal,
= 0, (2.89)
as is if a multiple of a function is subtracted:

= 0,

= +1(r r

). (2.90)
18
The system of rst-order equations Eqs. (2.88a), (2.88b) can be easily converted to
second-order form:
(
2
+
2
)

= (1)(r r

), (2.91a)
(
2
+
2
) = i1(r r

). (2.91b)
The system of equations Eqs. (2.91a), (2.91b) is quite general. We specialize to
the case of parallel plates by introducing the transverse Fourier transform:

(r, r

; ) =
_
(dk)
(2)
2
e
ik(r

)
g(z, z

; k, ). (2.92)
The equations satised by the various Cartesian components of g may be easily
worked out once it is recognized that
[(1)]
ij
=
i

ij

2
. (2.93)
In terms of the Fourier transforms, these equations are
_

2
z
2
k
2
+
2
_
g
zz
= k
2
(z z

), (2.94a)
_

2
z
2
k
2
+
2
_
g
zx
= ik
x

z
(z z

), (2.94b)
_

2
z
2
k
2
+
2
_
g
zy
= ik
y

z
(z z

), (2.94c)
_

2
z
2
k
2
+
2
_
g
xx
=
_
k
2
y
+

2
z
2
_
(z z

), (2.94d)
_

2
z
2
k
2
+
2
_
g
yy
=
_
k
2
x
+

2
z
2
_
(z z

), (2.94e)
_

2
z
2
k
2
+
2
_
g
xy
= k
x
k
y
(z z

). (2.94f)
We solve these equations subject to the boundary condition that the transverse
components of the electric eld vanish on the conducting surfaces, that is,
n

[
z=0,a
= 0, (2.95)
where n is the normal to the surface. That means any x or y components vanish at
z = 0 or at z = a. Therefore, g
xy
is particularly simple. By the standard discontinuity
method, we immediately nd [cf. Eq. (2.19)]
g
xy
= g
yx
=
k
x
k
y
sin a
(ss), (2.96)
19
where
(ss) = sin z
<
sin (z
>
a). (2.97)
To nd g
xx
we simply subtract a function:
g

xx
= g
xx
(z z

). (2.98)
Then, we again nd at once
g

xx
=
k
2
x

2
sin a
(ss), (2.99)
and similarly
g

yy
=
k
2
y

2
sina
(ss). (2.100)
To determine the boundary condition on g
zz
, we recall the solenoidal condition on

,
Eq. (2.90), which implies that

z
g
zz

z=0,a
= 0. (2.101)
This then leads straightforwardly to the conclusion
g
zz
=
k
2
sin a
(cc), (2.102)
where
(cc) = cos z
<
cos (z
>
a). (2.103)
The remaining components have the property that the functions are discontinuous,
while, apart from a function, their derivatives are continuous:
g
zx
=
ik
x
sin a
(cs), (2.104)
g
zy
=
ik
y
sin a
(cs), (2.105)
where, because of the analogue of Eq. (2.101), or because the components must vanish
when z

= 0, a,
(cs) =
_
cos z sin (z

a), z < z

,
sin z

cos (z a), z > z

.
(2.106)
20
Similarly,
g
xz
=
ik
x
sin a
(sc), (2.107)
g
yz
=
ik
y
sin a
(sc), (2.108)
where
(sc) =
_
sin z cos (z

a), z < z

,
cos z

sin (z a), z > z

.
(2.109)
The relation between Eqs. (2.104), (2.105) and Eqs. (2.107), (2.108) just reects the
symmetry

(r, r

) =
T
(r

, r), (2.110)
where T denotes transposition, which means that
g
ij
(z, z

; k) = g
ji
(z

, z; k). (2.111)
The normal-normal component of the electromagnetic stress tensor is
T
zz
=
1
2
(E
2
+ H
2
) (E
2
z
+ H
2
z
). (2.112)
The vacuum expectation value is obtained by the replacements
E(x)E(x

)) =
1
i
(x, x

), (2.113a)
H(x)H(x

)) =
1
i
1

2
(x, x

. (2.113b)
In terms of the Fourier transforms, we have
T
zz
) =
1
2i
2
_
(
2
k
2
)g
zz
+ (
2
k
2
y
)g
xx
+ (
2
k
2
x
)g
yy
+ ik
y
_

z
g
yz

g
zy
_
+ ik
x
_

z
g
xz

g
zx
_
+ k
x
k
y
(g
xy
+ g
yx
) +

z

(g
xx
+ g
yy
)
_
. (2.114)
When the appropriate Greens functions are inserted into the above, enormous sim-
plication occurs on the surface, and we are left with
T
zz
)

z=z

=0,a
= i cot a, (2.115)
which indeed is twice the scalar result (2.21), as claimed at the end of Sec. 2.1.
21
2.6 Fermionic Casimir Force
We conclude this section with a discussion of the force on parallel surfaces due to
uctuations in a massless Dirac eld. For this simple geometry, the primary distinc-
tion between this case and what has gone before lies in the boundary conditions. The
boundary conditions appropriate to the Dirac equation are the so-called bag-model
boundary conditions. That is, if n

represents an outward normal at a boundary

surface, the condition on the Dirac eld there is
(1 + in ) = 0. (2.116)
For the situation of parallel plates at z = 0 and z = a, this means
(1 i
3
) = 0 (2.117)
at z = 0 and z = a, respectively. In the following, we will choose a representation of
the Dirac matrices in which i
5
is diagonal, in 2 2 block form,
i
5
=
_
1 0
0 1
_
(2.118)
while

0
=
_
0 i
i 0
_
(2.119)
from which the explicit form of all the other Dirac matrices follow from = i
0

5
.
2.6.1 Summing modes
It is easiest, but not rigorous, to sum modes as in Sec. 2.1. We introduce a Fourier
transform in time and the transverse spatial directions,
(x) =
_
d
2
e
it
_
(dk)
(2)
2
e
ikx
(z; k, ), (2.120)
so that the Dirac equation for a massless fermion i = 0 becomes (if k lies along
the negative y axis)
_
i

z
_
u

kv

= 0, (2.121a)
ku

+
_
i

z
_
v

= 0. (2.121b)
where the subscripts indicate the eigenvalues of i
5
and u and v are eigenvectors of

3
with eigenvalue +1 or 1, respectively. This system of equations is to be solved
22
to the boundary conditions (2.117), or
u
+
+ u

[
z=0
= 0, (2.122a)
v
+
v

[
z=0
= 0, (2.122b)
u
+
u

[
z=a
= 0, (2.122c)
v
+
+ v

[
z=a
= 0. (2.122d)
The solution is straightforward. Each component satises
_

2
z
2
+
2
_
= 0, (2.123)
where
2
=
2
k
2
, so that the components are expressed as follows:
u
+
+ u

= Asin z, (2.124a)
v
+
v

= Bsin z, (2.124b)
u
+
u

= C sin (z a), (2.124c)

v
+
+ v

= Dsin (z a). (2.124d)

Inserting these into the Dirac equation (2.121a) and (2.121b), we nd, rst, an eigen-
value condition on :
cos a = 0, (2.125)
or
a = (n +
1
2
), (2.126)
where n is an integer. We then nd two independent solutions for the coecients:
A ,= 0, (2.127a)
B = 0, (2.127b)
C =
i

(1)
n
A, (2.127c)
D =
ik

(1)
n
A, (2.127d)
and
A = 0, (2.128a)
B ,= 0, (2.128b)
C =
k
i
(1)
n
B, (2.128c)
D =

i
(1)
n
B. (2.128d)
23
Thus, when we compute the zero-point energy, we must sum over odd integers, noting
that there are two modes, and remembering the characteristic minus sign for fermions:
Instead of Eq. (2.2), the Casimir energy is
u = 2
1
2

n=0
(dk)
(2)
2
_
k
2
+
(n + 1/2)
2

2
a
2
=
1
2

1
4

n=0
_

0
dt
t
t
3/2
e
t(n+1/2)
2

2
/a
2
=
1
8
3/2

3
2
_

n=0
(n + 1/2)
3

3
a
3
=

2
6a
3
7
8
(3), (2.129)
which is
7
8
2 times the scalar result (2.8) because (3) = B
4
/4 = 1/120. (The
factor of 2 refers to the two spin modes of the fermion.)
2.6.2 Greens Function Method
Again, a more controlled calculation starts from the equation satised by the Dirac
Greens function,

1
i
G(x, x

) = (x x

), (2.130)
subject to the boundary condition
(1 + in )G

z=0,a
= 0. (2.131)
We introduce a reduced, Fourier-transformed, Greens function,
G(x, x

) =
_
d
2
e
i(tt

)
_
(dk)
(2)
2
e
ik(xx

)
g(z, z

; k, ), (2.132)
which satises
_

0
+ k +
3
1
i

z
_
g(z, z

) = (z z

). (2.133)
Introducing the representation for the gamma matrices given above, we nd that the
components of g corresponding to the +1 or 1 eigenvalues of i
5
,
g =
_
g
++
g
+
g
+
g

_
, (2.134)
24
satisfy the coupled set of equations
_
k i
3

z
_
g

= 0, (2.135a)
_
k i
3

z
_
g

= i(z z

). (2.135b)
We then resolve each of these components into eigenvectors of
3
:
g

=
_
u
(+)

u
()

v
(+)

v
()

_
(2.136)
and similarly for g

. These components satisfy the coupled equations

_
i

z
_
u
()

kv
()

= 0, (2.137a)
ku
()

+
_
i

z
_
v
()

= 0, (2.137b)
_
i

z
_
u
(+)

kv
(+)

= i(z z

), (2.137c)
_
i

z
_
u
()

kv
()

= 0, (2.137d)
ku
(+)

+
_
i

z
_
v
(+)

= 0, (2.137e)
ku
()

+
_
i

z
_
v
()

= i(z z

), (2.137f)
which aside from the inhomogeneous terms are replicas of (2.121a) and (2.121b).
These equations are to be solved subject to the boundary conditions
u
()
++
u
()
+
[
z=a
= 0, (2.138a)
u
()
++
+ u
()
+
[
z=0
= 0, (2.138b)
u
()
+
u
()

[
z=a
= 0, (2.138c)
u
()
+
+ u
()

[
z=0
= 0, (2.138d)
v
()
++
+ v
()
+
[
z=a
= 0, (2.138e)
v
()
++
v
()
+
[
z=0
= 0, (2.138f)
v
()
+
+ v
()

[
z=a
= 0, (2.138g)
v
()
+
v
()

[
z=0
= 0. (2.138h)
25
Again, the solution is straightforward. We nd
u
(+)
++
= u
(+)

= v
()
++
= v
()

=
1
2 cos a
_
cos (z + z

a) +
i

sin (z + z

a)
_
, (2.139a)
v
(+)
++
= v
(+)

= u
()
++
= u
()

=
ik
2
1
cos a
sin (z + z

a), (2.139b)
u
(+)
+
= u
(+)
+
= v
()
+
= v
()
+
=
1
2 cos a
[(z z

) cos (z
>
z
<
a)

sin (z
>
z
<
a)
_
, (2.139c)
v
(+)
+
= v
(+)
+
= u
()
+
= u
()
+
=
ik
2
1
cos a
sin (z
>
z
<
a), (2.139d)
where
(z z

) =
_
1 if z > z

,
1 if z < z

.
(2.140)
We now insert these Greens functions into the vacuum expectation value of the
energy-momentum tensor. The latter is
T

=
1
2

0
1
2
_

1
i

1
i

_
+ g

L. (2.141)
We take the vacuum expectation value by the replacement

1
i
G, (2.142)
where G is the fermionic Greens function computed above. Because we are interested
in the limit as x

x we can ignore the Lagrangian term in the energy-momentum

tensor, leaving us with in the transform space
T
33
) =
1
2
Tr
3

3
g(z, z

z
=
i
2

z
tr
3
(g
+
+ g
+
)

z
=
i
2

z
[u
(+)
+
+ u
(+)
+
(v
()
+
+ v
()
+
)]

z
. (2.143)
26
When we insert the solution found above, Eqs. (2.139a)(2.139d), we obtain
T
33
) = 2i

z
Re u
(+)
+
. (2.144)
Carrying out the dierentiation and setting z = z

we nd
T
33
) = i tana, (2.145)
where again we ignore the -function term. [Compare the scalar result (2.21).]
We now follow the same procedure given in Sec. 2.2: The force per unit area is
f =
_
d
2
k
(2)
2
_
d
2
i tan a
=
_
d
2
k
(2)
2
_
d
2
tanh a
=
1
2
2
_

0
d
3
_
1
2
e
2a
+ 1
_
(2.146)
As in Eq. (2.23) we omit the 1 in the last square bracket: The same term is present
in the vacuum energy outside the plates, so cancels out when we compute the discon-
tinuity across the plates. We are left with, then,
f =
1
16
2
a
4
_

0
dxx
3
e
x
+ 1
. (2.147)
But
_

0
x
s1
dx
e
x
+ 1
= (1 2
1s
)(s)(s), (2.148)
so here we nd
f
F
=
7
2
1920a
4
, (2.149)
which is, indeed,
7
4
times the scalar force given in Eq. (2.9).
The eect of fermion uctuations was rst investigated by Ken Johnson [32], in
connection with the bag model [33].
3 Casimir Eect in Dielectrics
The formalism given in Sec. 2.5 can be readily extended to dielectric bodies [30]. The
starting point is the eective action in the presence of an external polarization source
P:
W =
_
(dx)[P (

A) + E (

A) H (A) +
1
2
(H
2
E
2
)],
(3.1)
27
which, upon variation with respect to H, E, A, and , yields the appropriate
Maxwells equations. Thus, because W is stationary with respect to these eld vari-
ations, its response to a change in dielectric constant is explicit:

W =
_
(dx)
1
2
E
2
. (3.2)
Comparison of i

W with the second iteration of the source term in the vacuum

persistence amplitude,
e
iW
= +
1
2
_
i
_
(dx)E P
_
2
+ , (3.3)
allows us to identify the eective product of polarization sources,
iP(x)P(x

e
= 1 (x x

). (3.4)
Thus, the numerical value of the action according to Eq. (2.86),
W =
1
2
_
(dx)P(x) E(x) =
1
2
_
(dx)(dx

)P(x) (x, x

) P(x

), (3.5)
implies the following change in the action when the dielectric constant is varied
slightly,
W =
i
2
_
(dx)(x)
kk
(x, x), (3.6)
which, in view of Eq. (3.2), is equivalent to the vacuum-expectation-value replacement
(2.113a).
For the geometry of parallel dielectric slabs, like that considered in Sec. 2, but
with dielectric materials in the three regions,
(z) =
_
_
_

1
, z < 0,

3
, 0 < z < a,

2
, a < z,
(3.7)
the components of the Greens dyadics may be expressed in terms of the TE (trans-
verse electric or H) modes and the TM (transverse magnetic or E) modes, given by
the scalar Greens functions satisfying
_

2
z
2
+ k
2

_
g
E
(z, z

) = (z z

), (3.8a)
_

z
1

+
k
2

2
_
g
H
(z, z

) = (z z

), (3.8b)
28
where, quite generally, = (z),

= (z

). The nonzero components of the Fourier

transform g given by Eq. (2.92) are easily found to be (in the coordinate system where
k lies along the +x axis)
g
xx
=
1

(z z

) +
1

z
1

g
H
, (3.9a)
g
yy
=
2
g
E
, (3.9b)
g
zz
=
1

(z z

) +
k
2

g
H
, (3.9c)
g
xz
= i
k

z
g
H
, (3.9d)
g
zx
= i
k

g
H
. (3.9e)
The trace required in the change of the action (3.6) is obtained by taking the limit
z

z, and consequently omitting delta functions:

g
kk
=
_

2
g
E
+
k
2

g
H
+
1

z
1

g
H
_

z=z

. (3.10)
This appears in the change of the energy when the second interface is displaced by
an amount a,
(z) = a(
2

3
)(z a), (3.11)
namely (A is the transverse area)
E
A
=
i
2
_
d
2
(dk)
(2)
2
dz (z)g
kk
(z, z

; k, ). (3.12)
Because g
E
, g
H
and
1

z
1

g
H
are all continuous, while

is not, we interpret the

trace of g in Eq. (3.12) symmetrically; we let z and z

approach the interface from

opposite sides, so the term
k
2

g
H

k
2

2
g
H
. Subsequently, we may evaluate the
Greens function on a single side of the interface. In terms of the notation

2
= k
2

2
, (3.13)
which is positive after a complex frequency rotation is performed (it is automatically
positive for nite temperature), the electric Greens function is in the region z, z

> a,
g
E
(z, z

) =
1
2
2
_
e

2
|zz

|
+ re

2
(z+z

2a)
_
, (3.14)
where the reection coecient is
r =

2

2
+
3
+
4
2

2
3

2
2
d
1
, (3.15)
29
with the denominator here being
d =

3
+
1

3
+
2

2
e
2
3
a
1. (3.16)
The magnetic Greens function g
H
has the same form but with the replacement
/

, (3.17)
except in the exponentials; the corresponding denominator is denoted by d

. [It is
easy to see that g
E
reduces to Eq. (2.24) when r = 1; the results in Sec. 2.5 follow
from Eqs. (3.9a)(3.9e) in the coordinate system adopted here.]
Evaluating these Greens functions just outside the interface, we nd for the force
on the surface per unit area
f =
i
2
_
d
2
(dk

)
(2)
2
__

2
+ 2
3
d
1

+
_

2
+ 2
3
d
1
_
, (3.18)
where the rst bracket comes from the E modes, and the second from the H modes.
The rst term in each bracket, which does not make reference to the separation a
between the surfaces, is seen to be a change in the volume energy of the system.
These terms correspond to the electromagnetic energy required to replace medium 2
by medium 3 in the displacement volume. They constitute the so-called bulk energy
contribution. The remaining terms are the Casimir force. We rewrite the latter by
making a complex rotation in the frequency,
i, real. (3.19)
This gives for the force per unit area at zero temperature
f
T=0
Casimir
=
1
8
2
_

0
d
_

0
dk
2
2
3
_
d
1
+ d
1
_
,
2
= k
2
+
2
. (3.20)
From this, we can obtain the nite temperature expression immediately by the sub-
stitution

2
n
= 4
2
n
2
/
2
, (3.21)
_

0
d
2

1

n=0

, (3.22)
the prime being a reminder to count the n = 0 term with half weight. These results
agree with those of Lifshitz et al. [17, 18].
Note that the same result (3.20) may be easily rederived by computing the normal-
normal component of the stress tensor on the surface, T
zz
, provided two constant
30
stresses are removed, terms which would be present if either medium lled all space.
The dierence between these two constant stresses,
T
vol
zz
= i
_
(dk

)
(2)
2
d
2
[
2

3
], (3.23)
precisely corresponds to the deleted volume energy in the previous calculation.
Various applications can be made from this general formula (3.20). In particular,
if we set the intermediate material to be vacuum,
3
= 1, and set
1
=
2
= ,

1
=

2
= 0, we recover the Casimir force between parallel, perfectly conducting
plates, Eq. (2.10), including the correct temperature dependence, given by Eq. (2.51)
with d = 2 and an additional factor of two representing the two polarizations. An-
other interesting result, important for the recent experiments [20], is the correction
for an imperfect conductor, where for frequencies above the infrared, an adequate
representation for the dielectric constant is [34]
() = 1

2
p

2
, (3.24)
where the plasma frequency is
4

2
p
=
4e
2
N
m
, (3.25)
where e and m are the charge and mass of the electron, and N is the number density
of free electrons in the conductor. A simple calculation shows, at zero temperature
[29, 30],
f

2
240a
4
_
1
8
3

1
ea
_
m
N
_
1/2
_
. (3.26)
Now suppose the central slab consists of a tenuous medium and the surrounding
medium is vacuum, so that the dielectric constant in the slab diers only slightly
from unity,
1 1. (3.27)
Then, with a simple change of variable,
= p, (3.28)
4
Although we have used rationalized Heaviside-Lorentz units in our electromagnetic action for-
malism, that is without eect, in that the one-loop Casimir eect is independent of electromagnetic
units. For considerations where the electric charge and polarizability appear, it seems more conve-
nient to use unrationalized Gaussian units.
31
we can recast the Lifshitz formula (3.20) into the form
f
1
32
2
_

0
d
3
_

1
dp
p
2
[() 1]
2
[(2p
2
1)
2
+ 1]e
2pa
. (3.29)
If the separation of the surfaces is large compared to the characteristic wavelength
characterizing , a
c
1, we can disregard the frequency dependence of the dielectric
constant, and we nd
f
23( 1)
2
640
2
a
4
. (3.30)
For short distances, a
c
1, the approximation is
f
1
32
2
1
a
3
_

0
d(() 1)
2
. (3.31)
These formulas are identical with the well-known forces found for the complementary
geometry in [30].
Now we wish to obtain these results from the sum of van der Waals forces, derivable
from a potential of the form
V =
B
r

. (3.32)
We do this by computing the energy (N = density of molecules)
E =
1
2
BN
2
_
a
0
dz
_
a
0
dz

_
(dr

)(dr

)
1
[(r

)
2
+ (z z

)
2
]
/2
. (3.33)
If we disregard the innite self-interaction terms (analogous to dropping the volume
energy terms in the Casimir calculation), we get
f =

a
E
A
=
2BN
2
(2 )(3 )
1
a
3
. (3.34)
So then, upon comparison with (3.30), we set = 7 and in terms of the polarizability,
=
1
4N
, (3.35)
we nd
B =
23
4

2
, (3.36)
or, equivalently, we recover the retarded dispersion potential [3],
V =
23
4

2
r
7
, (3.37)
whereas for short distances we recover from Eq. (3.31) the London potential [2],
V =
3

1
r
6
_

0
d ()
2
, (3.38)
which are the quantititive forms of Eqs. (1.3) and (1.2), respectively.
32
4 Casimir Eect on a Sphere
4.1 Perfectly Conducting Spherical Shell
The zero-point uctuations due to parallel plates, either conducting or insulating,
give rise to an attractive force, which seems intuitively understandable in view of
the close connection with the attractive van der Waals interactions. However, ones
intuition fails when more complicated geometries are considered.
In 1956 Casimir proposed that the zero-point force could be the Poincare stress
stabilizing a semiclassical model of an electron [35]. For deniteness, take a naive
model of an electron as a perfectly conducting shell of radius a carrying a total
charge e. The Coulomb repulsion must be balanced by some attractive force; Casimir
proposed that that could be the vacuum uctuation energy, so that the eective
energy of the conguration would be
E =
e
2
2a

Z
a
hc, (4.1)
where the Casimir energy is characterized by a pure number Z. The would open the
way for a semiclassical calculation of the ne-structure constant, for stability results
if E = 0 or
=
e
2
hc
= 2Z. (4.2)
Unfortunately as Tim Boyer was to discover a decade later [36], the Casimir force
in this case is repulsive, Z = 0.04618. The sign results from delicate cancellations
between interior and exterior modes, and between TE and TM modes, so it appears
impossible to predict the sign a priori.
Boyers calculation was rather complicated, involving nding the zeroes of Bessel
functions. In 1978 we found a simpler approach, based on Greens functions [31],
which I will describe here. In particular the Greens dyadic formalism of Sec. 2.5 may
be used, except now the modes must be described by vector spherical harmonics,
dened by [34, 37]
X
lm
= [l(l + 1)]
1/2
LY
lm
(, ), (4.3)
where L is the orbital angular momentum operator,
L =
1
i
r . (4.4)
The vector spherical harmonics satify the orthonormality condition
_
dX

m
X
lm
=
ll

mm
, (4.5)
33
as well as the sum rule
l

m=l
[X
lm
(, )[
2
=
2l + 1
4
. (4.6)
The divergenceless dyadics

and may be expanded in terms of vector spherical

harmonics as

lm
_
f
l
X
lm
+
i

g
l
X
lm
_
, (4.7a)
=

lm
_
g
l
X
lm

f
l
X
lm
_
, (4.7b)
where the second suppressed tensor index is carried by the coecient functions f
l
, g
l
,

f
l
, g
l
.
Inserting this expansion into the rst-order equations (2.88a) and (2.88b), and
using the properties of the vector spherical harmonics, we straightforwardly nd [31]
that the Greens dyadic may be expressed in terms of two scalar Greens functions,
the electric and the magnetic:
(r, r

; ) =

lm

2
F
l
(r, r

)X
lm
()X

lm
(

)
[G
l
(r, r

)X
lm
()X

lm
(

)]

+ -function terms, (4.8)

where the expression -function terms refers to terms proportional to spatial delta
functions. These terms may be omitted, as we are interested in the limit in which
the two spatial points approach coincidence. These scalar Greens functions satisfy
the dierential equation
_
1
r
d
2
dr
2
r
l(l + 1)
r
2
+
2
__
F
l
(r, r

)
G
l
(r, r

)
_
=
1
r
2
(r r

), (4.9)
subject to the boundary conditions that they be nite at the origin (the center of
the sphere), which picks out j
l
there, and correspond to outgoing spherical waves at
innity, which picks out h
(1)
l
. On the surface of the sphere, we must have
F
l
(a, r

) = 0,

r
rG
l
(r, r

r=a
= 0. (4.10)
The result is that
_
G
l
F
l
_
= G
0
l
+
_

G
l

F
l
_
, (4.11)
34
where G
0
l
is the vacuum Greens function,
G
0
l
(r, r

) = ikj
l
(kr
<
)h
(1)
l
(kr
>
), (4.12)
and in the interior and the exterior of the sphere respectively,
r, r

< a :
_

G
l

F
l
_
= A
G,F
ikj
l
(kr)j
l
(kr

), (4.13a)
r, r

> a :
_

G
l

F
l
_
= B
G,F
ikh
(1)
l
(kr)h
(1)
l
(kr

), (4.13b)
where the coecients are
A
F
= B
1
F
=
h
(1)
l
(ka)
j
l
(ka)
, (4.14a)
A
G
= B
1
G
=
[kah
(1)
l
(ka)]

[kaj
l
(ka)]

, (4.14b)
From the electromagnetic energy density we may derive the following formula for the
energy of the system
E =
_
(dr)
1
2i
_

d
2
e
i(tt

lm
k
2
[

F
l
(r, r

) +

G
l
(r, r

)]X
lm
() X

lm
(

)
X
lm
() [

F
l
(r, r

) +

G
l
(r, r

)] X

lm
(

r=r

. (4.15)
Note here that the vacuum term in the Greens functions has been removed, since
that corresponds to the zero-point energy that would be present in this formalism
if no bounding surface were present. Here we are putting the two spatial points
coincident, while we leave a temporal separation, = t t

, which is only to be set

equal to zero at the end of the calculation, and therefore serves as a kind of regulator.
The integration over the solid angle and the sum on m may be easily carried out,
and the radial integral over Bessel functions is simply done using recurrence relations.
After performing a complex frequency rotation, we obtain the following compact form
for the Casimir energy:
E =
1
2a

l=1
(2l + 1)
1
2
_

dy e
iy
x
d
dx
ln(1
2
l
), (4.16)
where

l
= [s
l
(x)e
l
(x)]

(4.17)
35
is written in terms of Ricatti-Bessel functions of imaginary argument,
s
l
(x) =
_
x
2
I
l+1/2
(x),
e
l
(x) =
_
2x

K
l+1/2
(x). (4.18)
Here, as a result of the Euclidean rotation,
x = [y[, y =
1
i
ka is real, as is =
1
i

a
0. (4.19)
The same formula may be derived by computing the stress on the surface through
use of the stress tensor [31].
A very rapidly convergent evaluation of this formula can be achieved by using the
uniform asymptotic expansions for the Bessel functions:
s
l
(x)
1
2
z
1/2
(1 + z
2
)
1/2
e

_
1 +

k=1
u
k
(t)

k
_
, (4.20a)
e
l
(x)
z
1/2
(1 + z
2
)
1/2
e

_
1 +

k=1
(1)
k
u
k
(t)

k
_
, l , (4.20b)
where
x = z, = l + 1/2, t = (1 +z
2
)
1/2
, = (1 +z
2
)
1/2
+ ln
z
1 + (1 +z
2
)
1/2
, (4.21)
and the u
k
(t) are polynomials in t of denite parity and of order 3k [38]. If we now
write
E =
1
2a

l=1
J(l, ), (4.22)
we easily nd
J(l, 0)
3
32
, l . (4.23)
In order to obtain a nite sum, therefore, we must keep ,= 0 until the end of the
calculation. By adding and subtracting the leading approximation to the logarithm,
we can write
J(l, ) = R
l
+ S
l
(), (4.24)
where
R
l
=
1
2
_

0
dz
_
(2l + 1)
2
ln(1
2
l
) +
1
(1 + z
2
)
3
_
= J(l, 0)
3
32
, (4.25)
36
and
S
l
() =
1
4
_

z dz e
iz
d
dz
1
(1 + z
2
)
3
. (4.26)
By use of the Euler-Maclaurin sum formula (2.76), we can work out the sum

l=1
S
l
() =
3
32
, (4.27)
precisely the negative of the value of a single term at = 0! The sum of the remainder,

l
R
l
, is easily evaluated numerically, and changes this result by less than 2%. Thus
the result for the Casimir energy for a spherical conducting shell is found to be
E =
0.092353
2a
. (4.28)
This agrees with the result found in 1968 by Boyer [36], evaluated more precisely by
Davies [39], and conrmed by a completely dierent method by Balian and Duplantier
in 1978 [40]. Recently, this result has been reconrmed, using a zeta function method,
by Leseduarte and Romeo [41]. A reconsideration using direct mode summation has
appeared very recently [42].
4.2 Fermion Fluctuations
The corresponding calculation for a massless spin-1/2 particle subject to bag model
boundary conditions (2.116) on a spherical surface,
(1 + in )G

S
= 0, (4.29)
was carried out by Ken Johnson [43] and by Milton [44]. The result is also a repulsive
stress, of less than one-half the magnitude of the electromagnetic result. (Recall that
for parallel plates, the reduction factor was 7/8.)
In this case we wish to solve the Greens function equation
_

1
i

_
G(x, x

) = (x x

) (4.30)
subject to the boundary condition (4.29). In the same representation for the gamma
matrices used before, this may be easily achieved in terms of the total angular mo-
mentum eigenstates (J = L + (1/2)):
Z
l=J1/2
JM
() =
_
l + 1/2 M
2l + 1
_
1/2
Y
lM1/2
()[+)

_
l + 1/2 M
2l + 1
_
1/2
Y
lM+1/2
()[). (4.31)
37
These satisfy
rZ
l=J1/2
JM
= Z
l=J1/2
JM
. (4.32)
Once the Greens function is found, it can be used in the usual way to compute the
vacuum expectation value of the stress tensor, which in the Dirac case is given by
T

=
1
2

0
1
2
_

1
i

1
i

_
+ g

L, (4.33)
L being the fermionic Lagrange function, which leads directly to
T
rr
=
1
2

r
tr rG(x, x

x
. (4.34)
The discontinuity of the stress tensor across the surface of the sphere gives the energy
according to
4a
2
[T
rr
(a) T
rr
(a+)] =

a
E(a). (4.35)
The result of a quite straightforward calculation (the details are given in Ref. [45])
gives the result for the sum of exterior and exterior modes, again, in terms of modied
spherical Bessel functions:
E =
2
a

l=0
(l + 1)
_

0
dxx cos x
d
dx
ln[(e
2
l
+ e
2
l+1
)(s
2
l
+ s
2
l+1
)]. (4.36)
This may again be numerically evaluated through use of the uniform asymptotic
approximants, with the result
E =
0.0204
a
. (4.37)
Somewhat less precison was obtained because, in this case, the leading uniform asymp-
totic approximation vanished.
5 Innite Circular Cylinder
Since parallel plates yield an attractive Casimir force, and a sphere has a repulsive
stress, one might guess that for a cylinder a zero stress results. The situation is not so
simple. The rst calculation was carried out in 1981 by DeRaad and Milton [46]. The
electrodynamic result turns out to be attractive but with rather small magnitude.
Consider a right circular perfectly conducting cylinder of innite length and radius
a. We compute the Casimir energy using the above Greens dyadic formalism, adapted
38
to this cylindrical basis. The necessary information about vector spherical harmonics
in this case is given in Stratton [37]. The results for the Greens dyadics are
(r, r

) =

m=
_
dk
2
_

2
MM

(d
m
k
2
)F
m
(r, r

)
+
1

2
NN

G
m
(r, r

)
_

mk
(, z)

mk
(

, z

), (5.1a)
(r, r

) =

m=
_
dk
2
_
iMN

G
m
(r, r

NM

F
m
(r, r

)
_

mk
(, z)

mk
(

, z

), (5.1b)
where the nonradial eigenfunctions are

mk
(, z) =
1

2
e
im
e
ikz
, (5.2)
the vector dierential operators M and N are
M =r
im
r

r
, N =rik

r

mk
r
zd
m
, (5.3)
and the scalar dierential operator d
m
is
d
m
=
1
r

r
r

r

m
2
r
2
. (5.4)
The scalar Greens functions in the interior and exterior regions are
r, r

< a :
1

2
F
m
(r, r

) =
i
2
2
_
J
m
(r
<
)H
m
(r
>
)
H

m
(a)
J

m
(a)
J
m
(r)J
m
(r

)
_
+
1

2
(
F
m
(r, r

), (5.5a)
1

G
m
(r, r

) =
i
2
2
_
J
m
(r
<
)H
m
(r
>
)
H
m
(a)
J
m
(a)
J
m
(r)J
m
(r

)
_
+
1

2
(
G
m
(r, r

), (5.5b)
(
G,F
m
(r, r

) =
1
2[m[
_
r
<
r
>
_
|m|

1
2[m[
r
|m|
r
|m|
a
2|m|
, m ,= 0, (5.5c)
39
r, r

> a :
1

2
F
m
(r, r

) =
i
2
2
_
J
m
(r
<
)H
m
(r
>
)
J

m
(a)
H

m
(a)
H
m
(r)H
m
(r

)
_
+
1

2
(
F
m
(r, r

), (5.6a)
1

G
m
(r, r

) =
i
2
2
_
J
m
(r
<
)H
m
(r
>
)
J
m
(a)
H
m
(a)
H
m
(r)H
m
(r

)
_
+
1

2
(
G
m
(r, r

), (5.6b)
(
G,F
m
(r, r

) =
1
2[m[
_
r
<
r
>
_
|m|

1
2[m[
a
2|m|
r
|m|
r
|m|
, m ,= 0, (5.6c)
where H
m
= H
(1)
m
is the Hankel function of the rst kind, and
2
=
2
k
2
. Although
(
G,F
0
are not determined, they do not contribute to physical quantities.
The result for the force per unit area is (z = a)
f = T
rr
(a

) T
rr
(a
+
)
=
1
2ia
2
_

d
2
()
1
2

m=
_

dk
2
_
z

z
ln
_
1

4
z
2
[(H
m
(z)J
m
(z))

]
2
_
_
.(5.7)
Here, because convergence is more subtle than in the previous cases, we have inserted
a frequency cuto function () that has the properties
(0) = 1,
()
1
[[
4
, [[ ,
() real for real or imaginary. (5.8)
A way of satisfying these conditions is to take
() =

i
_
a
i
a
2

2
i
+
a

i
a
2

i
_
, (5.9)
with the conditions on the residues and poles:

i
Re a
i
= 0, (5.10a)
2

i
Re
a
i

2
i
= 1. (5.10b)
Of course, the poles are to recede to innity. When we rotate the contour to imaginary
frequencies, the contribution of these poles is necessary to achieve a real result.
40
Once again we use the uniform asymptotic expansion to extract the leading be-
havior. We evaluate the m sum for the leading term by using the identity (2.42),
or
2

m=1
1
m
2
z
2
=

_
z
2

_
1 coth

z
2

_
+
1
z
2
+

_
z
2

. (5.11)
The result for the energy per unit length is
c = a
2
f =
1
8a
2
(S + R + R
0
), (5.12)
where
S =
1
2
_

0
dr (coth r 1)
+
1
2
_
5
d
d
+
7
2
d
2
d
2
+
1
2
d
3
d
3
__

0
dr
_

_
coth
r

1
_

1
r
_

=1
=
1
2
ln 2 +
5
8
, (5.13a)
R = 4

m=1
_

0
r dr
_
ln
_
1 (r(I
m
(r)K
m
(r))

)
2

+
r
2
4(m
2
+ r
2
)
3
_
= 0.0437, (5.13b)
R
0
= 2
_

0
r dr
_
ln
_
1 (r(I
0
(r)K
0
(r))

)
2

+
1
4r
2
_

1
4
=
1
2
ln + 0.6785. (5.13c)
Adding these numbers, we see that the infrared singularity cancels, and we obtain
an attractive result,
c =
0.01356
a
2
. (5.14)
Very recently, this result has been conrmed by two independent calculations.
First, Gosdzinsky and Romeo obtained the same answer, to eight signicant gures,
using a zeta-function technique [47]. Shortly thereafter, an earlier calculation by
Nesterenko was corrected to yield the same answer [48]. This later method is based
not on the Greens function, but rather an analytic evaluation of the sum of zero-point
energies through the formal formula
E =
1
2
_

dk
2

m=
1
2i
1
2
_
C
d

ln f
n
(k, , a), (5.15)
41
where f
n
is a function, the zeroes of which are the mode frequencies of the system,
and where C is a contour initially chosen to encircle those zeroes. The divergences
there are regulated by a zeta function technique; the results, both analytically and
numerical, are identical to those reported here and derived in Ref. [46].
6 Casimir Eect on a D-dimensional Sphere
Because of the rather mysterious dependence of the sign and magnitude of the Casimir
stress on the topology and dimensionality of the bounding geometry, we have carried
out calculation of TE and TM modes bounded by a spherical shell in D spatial
dimensions. We rst consider massless scalar modes satisfying Dirichlet boundary
conditions on the surface, which are equivalent to electromagnetic TE modes. Again
we calculate the vacuum expectation value of the stress on the surface from the
Greens function.
The Greens function G(x, t; x

, t

) satises the inhomogeneous Klein-Gordon equa-

tion (2.12), or
_

2
t
2

2
_
G(x, t; x

, t

) =
(D)
(x x

)(t t

), (6.1)
where
2
is the Laplacian in D dimensions. We will solve the above Greens function
equation by dividing space into two regions, region I, the interior of a sphere of radius
a and region II, the exterior of the sphere. On the sphere we will impose Dirichlet
boundary conditions
G(x, t; x

, t

|x|=a
= 0. (6.2)
In addition, in region I we will require that G be nite at the origin x = 0 and in region
II we will require that G satisfy outgoing-wave boundary conditions at [x[ = .
The radial Casimir force per unit area f on the sphere is obtained from the radial-
radial component of the vacuum expectation value of the stress-energy tensor [31]:
f = 0[T
rr
in
T
rr
out
[0)

r=a
. (6.3)
To calculate f we exploit the connection between the vacuum expectation value of the
stress-energy tensor T

(x, t) and the Greens function at equal times G(x, t; x

, t):
f =
1
2i
_

r

G(x, t; x

, t)
in

G(x, t; x

, t)
out
_

x=x

, |x|=a
. (6.4)
To evaluate the expression in (6.4) it is necessary to solve the Greens function
equation (6.1). We begin by taking the time Fourier transform of G:
(

(x; x

) =
_

dt e
i(tt

)
G(x, t; x

, t

). (6.5)
42
The dierential equation satised by (

is

2
+
2
_
(

(x; x

) =
(D)
(x x

). (6.6)
To solve this equation we introduce polar coordinates and seek a solution that
has cylindrical symmetry; i.e., we seek a solution that is a function only of the two
variables r = [x[ and , the angle between x and x

so that x x

= rr

cos . In terms
of these polar variables (6.6) becomes

2
+

2
r
2
+
D 1
r

r
+
sin
2D

r
2

sin
D2

_
(

(r, r

, )
=

_
D1
2
_
2
(D1)/2
r
D1
sin
D2

(r r

)(). (6.7)
Note that the D-dimensional delta function on the right side of (6.6) has been replaced
by a cylindrically-symmetric delta function having the property that its volume in-
tegral in D dimensional space is unity. The D-dimensional volume integral of a
cylindrically-symmetric function f(r, ) is
2
(D1)/2

_
D1
2
_
_

0
dr r
D1
_

0
d sin
D2
f(r, ). (6.8)
We solve (6.7) using the method of separation of variables. Let
(

(r, r

, ) = A(r)B(z), (6.9)
where z = cos . The equation satised by B(z) is then
_
(1 z
2
)
d
2
dz
2
z(D 1)
d
dz
+ n(n + D 2)
_
B(z) = 0, (6.10)
where we have anticipated a convenient form for the separation constant. The equa-
tion satised by A(r) is
_
d
2
dr
2
+
D 1
r
d
dr

n(n + D2)
r
2
+
2
_
A(r) = 0 (r ,= r

). (6.11)
The solution to Eq. (6.10) that is regular at [z[ = 1 is the ultraspherical (Gegenbauer)
polynomial [38]
B(z) = C
(1+D/2)
n
(z) (n = 0, 1, 2, 3, . . . ). (6.12)
The solution in region I to Eq. (6.11) that is regular at r = 0 involves the Bessel
function [38]
A(r) = r
1D/2
J
n1+
D
2
([[r). (6.13)
43
In Eq. (6.13) we assume that D 2 in order to eliminate the linearly independent
solution A(r) = r
1D/2
Y
n1+
D
2
([[r), which is singular at r = 0 for all n. The solution
in region II to Eq. (6.11) that corresponds to an outgoing wave at r = involves a
Hankel function of the rst kind [38]
A(r) = r
1D/2
H
(1)
n1+
D
2
([[r). (6.14)
Using a few properties of the ultraspherical polynomials, namely, orthonormality
[38]
_
1
1
dz (1 z
2
)
1/2
C
()
n
(z)C
()
m
(z) =
2
12
(n + 2)
n! (n + )
2
()

nm
( ,= 0), (6.15)
and the value of the ultraspherical polynomials at z = 1,
C
()
n
(1) =
(n + 2)
n! (2)
( ,= 0), (6.16)
as well as the duplication formula (2.31), we solve for Greens function in the two
regions. Adding the interior and the exterior contributions, and performing the usual
imaginary frequency rotation, we obtain the nal expression for the stress [49]:
f =

n=0
(n 1 +
D
2
)(n + D 2)
2
D1

D+1
2
a
D+1
n!
_
D1
2
_

_

0
dx
_
x
d
dx
ln
_
I
n1+
D
2
(x)K
n1+
D
2
(x)x
2D
_
_
. (6.17)
It is easy to check that this reduces to the known case at D = 1, for there the series
truncatesonly n = 0 and 1 contribute, and we easily nd
f =
F
2
=

96a
2
, (6.18)
which agrees with Eq. (2.32) for d = 0 and a 2a.
In general, however, although each x integral can be made nite, the sum over n
still diverges. We can make the integrals nite by replacing the x
2D
factor in the
logarithm by simply x, which we are permitted to do if D < 1 because

n=0
(n + )
n!
0 ( < 0, ,= N). (6.19)
Then the total stress on the sphere is
F =
1
a
2

n=0
(n + D 2)
n!(D 1)
Q

_
= n +
D
2
1
_
, (6.20)
44
where the integral is
Q

=
_

0
dx ln (2xI

(x)K

(x)) . (6.21)
We proceed as follows:
Analytically continue to D < 0, where the sum converges, although the integrals
become complex.
Add and subtract the leading asymptotic behavior of the integrals.
Continue back to D > 0, where everything is now nite.
We used two methods to carry out the numerical evaluations, which gave the same
results. In the rst method we use the uniform asymptotic expansions to yield
Q

2
+

128

35
32768
3
+
565
1048576
5
+ . . . . (6.22)
Then using the identities

n=0
(n + D 2)
n!
= 0 for D < 2, (6.23a)

n=0
(n + D 2)
n!

2
= 0 for D < 0, (6.23b)
we obtain the following expression convergent for D < 4 (further subtractions can be
made for higher dimensions):
F
1
a
2

1
(D 1)
_ [ND/2+1]

n=0
(n + D 2)
n!


Q

n=[ND/2+2]
(n + D 2)
n!
_

35
32768
2
+
565
1048576
4
_ _
, (6.24)
where

Q

= Q

/2 /128, and the square brackets denote the largest integer

less or equal to its argument. The innite sums are easily evaluated in terms of
gamma functions.
In the second method we carry out an asymptotic expansion of the summand in
(6.20) in n, n
(n + D 2)
n!
Q

1
2
_
n
D1
+
(D 1)(D 2)
2
n
D2
+
24D
4
176D
3
+ 504D
2
688D + 387
192
n
D3
+ . . .
_
.(6.25)
45
0.0 2.0 4.0
D
-0.4
-0.2
0.0
F

a
2
Figure 4: Scalar Casimir stress F for 0 < D < 5 on a spherical shell.
The sums on n in the terms in the asymptotic expansion are carried out according to

n=1
n
Dk
= (k D), D k < 1. (6.26)
In this way we obtain the following formula suitable for numerical evaluation:
F =
1
a
2
_
1
2
Q
D/21
+
1
(D1)

n=1
_
(n + D 2)
n!

1
2
l

k=1
n
Dk
c
k
(D)
_
+
1
2(D1)
l

k=1
(k D)c
k
(D)
_
, (6.27)
where the c
k
(D) are polynomials in D.
Both methods give the same results [49], which are shown in Fig. 4. Note the
following salient features:
Poles occur at D = 2n, n = 1, 2, 3, . . . .
As we will see in the next plot for negative D, branch points occur at D = 2n,
n = 0, 1, 2, 3, . . . , and the stress is complex for D < 0.
The stress vanishes at negative even integers, F(2n) = 0, n = 1, 2, 3, . . . , but
is nonzero at D = 0: F(0) = 1/2a
2
.
46
-2.0 -1.0 0.0 1.0 2.0 3.0 4.0
D
-1.0
0.0
1.0
2.0
F
a
2
F
TM
= solid
F
TE
= dashed
Figure 5: A plot of the TM and TE Casimir stress for 2 < D < 4 on a spherical
shell. For D < 2 (D < 0) the stress F
TM
(F
TE
) is complex and we have plotted the
real part.
The case of greatest physical interest, D = 3, has a nite stress, but one
which is much smaller than the corresponding electrodynamic one: F(3) =
+0.0028168/a
2
. (This result was conrmed in Ref. [41].)
The same kind of calculation can be carried out for the TM modes [50]. The TM
modes are modes which satisfy mixed boundary conditions on the surface [36, 51],

r
r
D2
G(x, t; x

, t

|x|=r=a
= 0, (6.28)
The results are qualitatively similar, and are shown in Fig. 5. In particular, removing
the n = 0 contribution from the sum of the TE and TM contributions, we recover
the Boyer result (4.28).
7 Dielectric Ball: A Model for Sonoluminescence?
In all the above illustrations, niteness of the Casimir energy has been the result of
delicate cancellation between interior and exterior modes. This cancellation occurs
47
because the speed of light is identical in the two regions. This indicates that if one
considered, for example, a dielectric ball, the procedure given here would not yield a
nite result, and indeed, such is the case. The same thing occurs for cavity modes,
which are relevant for gluon (or quark) uctuations in the bag model of hadronsfor
a discussion of these issues see Ref. [52].
A new application of such cavity modes has recently appeared, in connection
with the suggestion by Schwinger [53] that the dynamical Casimir eect could be
relevant for the roughly 1 million photons emitted per cycle in sonoluminescence [54].
There has been considerable controversy concerning this suggestion [55, 56, 57]. The
diculty is that it has been impossible to extract nite Casimir energies, so that it has
been suspected that the most divergent term, the so-called bulk energy contribution,
could give rise to the necessary large, virtual photonic, energy needed. Dynamically,
the diculty has seemed to be the extremely short time scales required, which seem
to imply superluminal velocities.
Schwinger [53] and Carlson et al. [56] simply use the unregulated dierence be-
tween the zero-point energy of the medium and that of the vacuum within the bubble
for their static estimates. That dierence is evidently
E
bulk
=
4a
3
3
_
(dk)
(2)
3
1
2
k
_
1
1
n
_
, (7.1)
where n is the index of refraction of the medium. This is quartically divergent, and
with a plausible cuto indeed can yield the required 10
6
optical photons per pulse.
However, as we have seen, even in the simplest geometries, the naive unregulated
zero-point energy bears no relation to the observable, nite, Casimir energy, either in
magnitude or sign. Although Schwinger had earlier embarked on a calculation of the
Casimir energy for a dielectric bubble, he abandoned the eort before its completion
[58]. However, I had considered the complementary situation, that of a dielectric ball,
already in 1980 [59]. The extension to a dielectic-diamagnetic ball, with permittivity

and permeability

, surrounded by a medium with permittivity and permeability

, was given in 1995 [60]. Using the methods given here, the following formula was
derived for the Casimir energy:
E =
1
4a
_

dy e
iy

l=1
(2l + 1)x
d
dx
ln S
l
, (7.2)
where
S
l
= [s
l
(x

)e

l
(x) s

l
(x

)e
l
(x)]
2

2
[s
l
(x

)e

l
(x) + s

l
(x

)e
l
(x)]
2
, (7.3)
where again the s
l
, e
l
are the spherical Bessel functions of imaginary argument, the
quantity is
=
_

1
_

+ 1
, (7.4)
48
the time-splitting parameter is now denoted by , and
x = [y[

, x

= [y[
_

. (7.5)
It is easy to check that this result reduces to that for a perfectly conducting spherical
shell if we set the speed of light inside and out the same,

=

, as well as set
= 1. However, if the speed of light is dierent in the two regions, the result is no
longer nite, but quartically divergent, and indeed the Schwinger result (7.1) follows
for that leading divergent term.
But a divergence is not an answer. This bulk term in fact was present in the simpler
geometry of the parallel dielectric slabs, treated in Sec. 3, where it was removed
as a change in the volume energy of the system. In other words, the term to be
removed is a contribution to the self energy of the material, since it renormalizes the
phenomenologically described bulk energy of the substance. This issue is discussed
in detail in Ref. [61]. Mathematically, this term is removed by subtracting o the
uniform medium part of the Greens function in each region, just as we did in Sec. 4.1.
We are then left with a cubically divergent energy.
To simplify the subsequent mathematics, let us henceforward assume that the
media are dilute and nonmagnetic, 1 1,

1 1, and =

= 1. This
should be suciently good for the qualitative analysis of the situation. Then the
subtracted energy has the divergence structure
E
(

)
2
4a
1

3
. (7.6)
Because the dimensionless time-splitting cuto here has the structure = /a, this
divergence has the form of a contribution to the surface tension, which therefore
should also be removed by renormalization of a phenomenological parameter. Then,
we are left with a nite, observable remainder.
Previously [59, 60, 61] I had kept merely the leading term in the uniform asymp-
totic expansion in estimating this nite result, believing, as with the perfectly con-
ducting sphere, that that approximation should be quite accurate. But Brevik and
Marachevsky calculated the next terms in that expansion and proved me wrong [62].
So jointly we then computed the Casimir energy exactly for a dilute dielectric sphere
[63]. That calculation can be done either by the point-split cuto method sketched
here, or directly by the zeta function trick, which swallows all the divergences. In
the latter approach, we add and subtract the rst two leading uniform asymptotic
approximants
E =
( 1)
2
64a

l=1
_

65
128
_
+ E
R
( = l + 1/2 ), (7.7)
where the remainder is the dierence between the exact expression and the asymptotic
49
one. The l sums here are evaluated according to the zeta function recipe,

l=0

s
= (2
s
1)(s). (7.8)
Keeping only the rst term in the asymptotic series gives the result that I previously
quoted, E
1
( 1)
2
/(256a), while including the rst two terms reverses the sign
but hardly changes the magnitude [62]: E
2
( 1)
2
33/(8192a). Stopping here
would give a 15% overestimate; including the remainder gives the result
5
[63]
E = ( 1)
2
0.004767
a
, (7.9)
which is approximated to better than 2% by keeping the third term in the asymptotic
expansion [62].
What is most remarkable about this result is that it coincides with the van der
Waals energy calculated two years earlier for this nontrivial geometry [61]. That
is, starting from the Casimir-Polder potential [3] (3.37) we summed the pairwise
potentials between molecules making up the media. A sensible way to regulate this
calculation is dimensional continuation, similar to that described in Sec. 6. That is,
we evaluate the integral
E
vdW
=
23
8

2
N
2
_
d
D
r d
D
r

(r
2
+ r
2
2rr

cos )
/2
, (7.10)
where is the angle between r and r

, by rst regarding D > so the integral exists.

The integral may be done exactly in terms of gamma functions, which when continued
to D = 3, = 7 yields [61]
E
vdW
=
23
1536a
( 1)
2
. (7.11)
This attractive result is numerically identical with the Casimir result (7.9). (Inci-
dentally, the dilute Casimir eect for a cylinder, or equivalently, the van der Waals
energy, vanishes [48, 65].)
Thus, there is no longer any doubt that the Casimir eect is coincident with
van der Waals forces. The fact that the latter is physically obviously proportional to
(1)
2
(which had been already emphasized in Ref. [59]) seems convincing proof that
the expression (7.1) is incorrect. In turn, this seems to demonstrate the irrelevance
of the Casimir eect to the light production mechanism in sonoluminescence. This
is because nearly certainly the adiabatic approximation is valid (because the ash
time is about 10
11
s, far longer that the characteristic optical time scale of 10
15
s. (Thus the instantaneous approximation of Ref. [57], which on the face seems to
require superluminal velocities, is not credible.) For details, see Ref. [66].
5
Immediately after I had completed this calculation, I received a manuscript from Barton with
the same result [64] calculated by a more elementary method using ordinary perturbation theory.
50
8 Conclusions
In this survey of the Casimir eect, I have been very selective. Emphasis has been
upon my own work, in part because of its greater familiarity. However, that emphasis
also reects my bias toward the superiority of Greens function techniques over, say,
zeta function methods, which may reach the correct answer by dening away phys-
ically interesting divergent contributions. The rst six sections of this report treat
unambiguously nite Casimir forces, about which there can be no argument. That is
not to say there is not yet work to be done there, for it remains true that even the
sign of the Casimir force is unpredictable without detailed mathematical calculation.
Section 7 is far more speculative, in that when interior and exterior modes are
divorced, even so mildly as through a dierent speed of light, the Casimir eect is
no longer nite. Removal of that divergence requires some sort of renormalization,
which remains controversial. Thus, it is still possible to disagree about the possible
relevance of the Casimir eect to sonoluminescence, although I, of course, have a
denite (negative) opinion on the matter. Recent developments in this domain of the
Casimir eect suggest that earlier calculations of the Casimir eect in the bag model
of hadrons be re-examined [67, 45, 52].
Due to lack of time and space, many other interesting topics had to be omitted.
For example, the Casimir eect likely plays a role in the compactication of higher
dimensional unied models [68]. It may also be relevant in 2-dimensional condensed
matter systems [69]. Most important are the small, but conceptually vital, inclusion
of radiative corrections, for everything discussed here has been at the one-loop level
[70]. For these, and many other applications and technical developments, the reader
is referred to the original literature, or to my forthcoming book on the subject [71].
Acknowledgements
I am very grateful to Choonkyu Lee for inviting me to the 17th Symposium on
Theoretical Physics at Seoul National University, and for his gracious hospitality,
and to all the participants in the Symposium for their interest and encouragement.
This work was supported in part by the U.S. Department of Energy.
References
[1] J. D. van der Waals, Over de Continueteit van den Gas- en Vloeistoftoestand
(thesis, Leiden, 1873).
[2] F. London, Z. Physik 63, 245 (1930).
[3] H. B. G. Casimir and D. Polder, Phys. Rev. 73, 360 (1948).
51
[4] H. B. G. Casimir, talk at the Fourth Workshop on Quantum Field Theory Under
the Inuence of External Conditions, Leipzig, 1418 September, 1998.
[5] H. B. G. Casimir, Proc. K. Ned. Akad. Wet. 51, 793 (1948).
[6] B. V. Deriagin and I. I. Abrikosova, Zh. Eksp. Teor. Fiz. 30, 993 (1956); 31, 3
(1956). [English transl.: Soviet Phys. JETP 3, 819 (1957); 4, 2 (1957).]
[7] B. V. Deriagin, I. Abrikosova, and E. M. Lifshitz, Quart. Rev. 10, 295 (1968).
[8] A. Kitchener and A. P. Prosser, Proc. Roy. Soc. (London) A 242, 403 (1957).
[9] M. Y. Sparnaay, Physica 24, 751 (1958).
[10] W. Black, J. G. V. de Jongh, J. T. G. Overbeck, and M. J. Sparnaay, Trans.
Faraday Soc. 56, 1597 (1960).
[11] A. van Silfhout, Proc. Kon. Ned. Akad. Wetensch. B 69, 501 (1966).
[12] D. Tabor and R. H. S. Winterton, Nature 219, 1120 (1968).
[13] D. Tabor and R. H. S. Winterton, Proc. Roy. Soc. (London) A 312, 435 (1969).
[14] R. H. S. Winterton, Contemp. Phys. 11, 559 (1970).
[15] J. N. Israelachivili and D. Tabor, Proc. Roy. Soc. (London) A 331, 19 (1972).
[16] E. S. Sabisky and C. H. Anderson, Phys. Rev. A 7, 790 (1973).
[17] E. M. Lifshitz, Zh. Eksp. Teor. Fiz. 29, 94 (1956). [English transl.: Soviet Phys.
JETP 2, 73 (1956).]
[18] I. D. Dzyaloshinskii, E. M. Lifshitz, and L. P. Pitaevskii, Usp. Fiz. Nauk 73, 381
(1961). [English transl.: Soviet Phys. Usp. 4, 153 (1961).]
[19] S. K. Lamoreaux, Phys. Rev. Lett. 78, 5 (1997).
[20] U. Mohideen and A. Roy, Phys. Rev. Lett. 81, 4549 (1998).
[21] J. Schwinger, Particles, Sources, and Fields, Vol. 1 (Addison-Wesley, Reading,
Massachusetts, 1970).
[22] C. G. Callen, Jr., S. Coleman, and R. Jackiw, Ann. Phys. (N.Y.) 59, 42 (1970).
[23] J. Ambjrn and S. Wolfram, Ann. Phys. (N. Y.) 147, 1 (1983).
[24] P. Hays, Ann. Phys. (N. Y.) 121, 32 (1979).
52
[25] I. S. Gradshteyn and I. M. Rhyzik, Table of Integrals, Series, and Products
(Academic Press, New York, 1965).
[26] M. Fierz, Helv. Phys. Acta 33, 855 (1960).
[27] F. Sauer, Ph.D. thesis, Gottingen, 1962.
[28] J. Mehra, Acta Physica Austriaca (1964); Physica 37, 145 (1967).
[29] C. M. Hargreaves, Proc. Kon. Ned. Akad. Wetensch. B 68, 231 (1965).
[30] J. Schwinger, J. L. L. DeRaad, and K. A. Milton, Ann. Phys. (N.Y.) 115, 1
(1978).
[31] K. A. Milton, L. L. DeRaad, Jr., and J. Schwinger, Ann. Phys. (N.Y.) 115, 388
(1978).
[32] K. Johnson, Acta Phys. Pol. B6, 865 (1975).
[33] A. Chodos, R. L. Jae, K. Johnson, C. B. Thorn, and V. Weisskopf, Phys. Rev.
D 9, 3471 (1974); A. Chodos, R. L. Jae, K. Johnson, and C. B. Thorn, Phys.
Rev. D 10, 2599 (1974): T. DeGrand, R. L. Jae, K. Johnson, and J. Kiskis,
Phys. Rev. D 12, 2060 (1975).
[34] J. Schwinger, L. L. DeRaad, Jr., K. A. Milton, and W.-y. Tsai, Classical Elec-
trodynamics (Perseus, Reading, Massachusetts, 1998).
[35] H. B. G. Casimir, Physica 19, 846 (1956).
[36] T. H. Boyer, Phys. Rev. 174, 1764 (1968).
[37] J. A. Stratton, Electromagnetic Theory (McGraw-Hill, New York, 1941).
[38] M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions (National
Bureau of Standards, Washington, D.C., 1964).
[39] B. Davies, J. Math. Phys. 13, 1324 (1972).
[40] R. Balian and B. Duplantier, Ann. Phys. (N.Y.) 112, 165 (1978).
[41] S. Leseduarte and A. Romeo, Europhys. Lett. 34, 79 (1996); Ann. Phys. (N.Y.)
250, 448 (1996).
[42] V. V. Nesterenko and I. G. Pirozhenko, Phys. Rev. D 57, 1284 (1997); M. E.
Bowers and C. R. Hagen, Phys. Rev. D 59, 0250007 (1999).
[43] K. Johnson, unpublished.
[44] K. A. Milton, Ann. Phys. (N.Y.), 150, 432 (1983).
53
[45] K. A. Milton, Phys. Rev. D 22, 1444 (1980).
[46] L. L. DeRaad, Jr. and K. A. Milton, Ann. Phys. (N.Y.) 136, 229 (1981).
[47] P. Gosdzinsky and A. Romeo, hep-th/9809199.
[48] K. A. Milton, A. V. Nesterenko, and V. V. Nesterenko, hep-th/9711168 v3.
[49] C. M. Bender and K. A. Milton, Phys. Rev. D 50, 6547 (1994).
[50] K. A. Milton, Phys. Rev. D 55, 4940 (1997).
[51] J. C. Slater and N. H. Frank, Electromagnetism (McGraw-Hill, New York, 1947),
p. 153.
[52] K. A. Milton, Phys. Rev. D 27, 439 (1983).
[53] J. Schwinger, Proc. Natl. Acad. Sci. USA 90, 958, 2105, 4505, 7285 (1993); 91,
6473 (1994).
[54] B. P. Barber, R. A. Hiller, R. Lofstedt, S. J. Putterman, and K. Weniger, Phys.
Rep. 281, 65 (1997).
[55] C. Eberlein, Phys. Rev. A 53, 2772 (1996); Phys. Rev. Lett. 76, 3842 (1996).
[56] C. E. Carlson, C. Molina-Pars, J. Perez-Mercader, and M. Visser, Phys. Lett.
B395, 76 (1997); Phys. Rev. D 56, 1262 (1997); C. Molina-Pars and M. Visser,
Phys. Rev. D 56, 6629 (1997). Apparently, this work grew out of the suggestion
that Gamma Ray Bursters have their origin in the Casimir eect: C. Carlson,
T. Goldman, and J. Perez-Mercader, Europhys. Lett. 36, 637 (1996).
[57] S. Liberati, M. Visser, F. Belgiorno, and D. W. Sciama. Sonoluminescence:
Bogolubov Coecients for the QED Vacuum of a Collapsing Bubble, quant-
ph/9805023; S. Liberati, F. Belgiorno, M. Visser, and D. W. Sciama, Sonolu-
minescence as a Quantum Vacuum Eect, quant-ph/9805031.
[58] J. Schwinger, Lett. Math. Phys. 24, 59, 227 (1992); Proc. Natl. Acad. Sci. USA
89, 4091, 11118 (1992).
[59] K. A. Milton, Ann. Phys. (N.Y.) 127, 49 (1980).
[60] K. A. Milton, Third Workshop on Quantum Field Theory Under the Inuence
of External Conditions, ed. M. Bordag (Teubner, Stuttgart, 1996), p. 3; K. A.
Milton and Y. J. Ng, Phys. Rev. E 55, 4207 (1997).
[61] K. A. Milton and Y. J. Ng, Phys. Rev. E 57, 5504 (1998).
54
[62] I. Brevik and V. N. Marachevsky, Casimir Surface Force on a Dilute Dielectric
Ball.
[63] I. Brevik, V. N. Marachevsky, and K. A. Milton, OKHEP9808, hep-
th/9810062.
[64] G. Barton, Perturbative Check on the Casimir Energy on a Nondispersive Di-
electric Ball.
[65] A. Romeo, private communication.
[66] K. A. Milton, talk at Fourth Workshop on Quantum Field Theory Under the
Inuence of External Conditions, Leipzig, September 1418, 1998, OKHEP98
11, hep-th/9811174.
[67] K. A. Milton, Phys. Rev. D 22, 1441 (1980).
[68] T. Applequist and A. Chodos, Phys. Rev. Lett. 50, 141 (1983); Phys. Rev. D
28, 772 (1983); S. Weinberg, Phys. Lett. 125B, 265 (1983); P. Candelas and S.
Weinberg, Nucl. Phys. B237, 397 (1964); A. Chodos and E. Myers, Ann. Phys.
(N.Y.) 156, 412 (1984); Phys. Rev. D 31, 3064 (1985); R. Kantowski and K. A.
Milton, Phys. Rev. D 35, 549 (1987); Phys. Rev. D 36, 3712); D. Birmingham,
R. Kantowski, and K. A. Milton, Phys. Rev. D 38, 1809 (1988); and references
therein.
[69] K. A. Milton and Y. J. Ng, Phys. Rev. D 42, 2875 (1990); Phys. Rev. D 46, 842
(1992).
[70] M. Bordag, D. Robaschik, and E. Wieczorek, Ann. Phys. (N.Y.) 165, 192 (1985);
M. Bordag and J. Lindig, Phys. Rev. D 58, 045003 (1998), and references therein.
[71] K. A. Milton, The Casimir Eect: Physical Application of Zero-Point Energy,
(World Scientic, in preparation).
55