Fluctuation-induced forces in confined ideal and imperfect Bose gases

H. W. Diehl and Sergei B. Rutkevich

Fakultät für Physik, Universität Duisburg-Essen, D-47058 Duisburg, Germany
(Dated: July 2, 2018)
Fluctuation-induced (“Casimir”) forces caused by thermal and quantum fluctuations are investi-
gated for ideal and imperfect Bose gases confined to d-dimensional films of size ∞d−1 × D under
periodic (P), antiperiodic (A), Dirichlet-Dirichlet (DD), Neumann-Neumann (NN), and Robin (R)
boundary conditions (BCs). The full scaling functions ΥBC d (xλ = D/λth , xξ = D/ξ) of the residual
−(d−1) BC
reduced grand potential per area, ϕBC res,d (T, µ, D) = D Υd (xλ , xξ ), are determined for the
arXiv:1704.00507v2 [cond-mat.quant-gas] 11 Jun 2017

ideal gas case with these BCs, where λth and ξ are the thermal de-Broglie wavelength and the bulk
correlation length, respectively. The associated limiting scaling functions ΘBC BC
d (xξ ) ≡ Υd (∞, xξ )
describing the critical behavior at the bulk condensation transition are shown to agree with those
previously determined from a massive free O(2) theory for BC = P, A, DD, DN, NN. For d = 3, they
are expressed in closed analytical form in terms of polylogarithms. The analogous scaling functions
ΥBC R
d (xλ , xξ , c1 D, c2 D) and Θd (xξ , c1 D, c2 D) under the RBCs (∂z − c1 )φ|z=0 = (∂z + c2 )φ|z=D = 0
with c1 ≥ 0 and c2 ≥ 0 are also determined. The corresponding scaling functions ΥP ∞,d (xλ , xξ ) and
ΘP∞,d (xξ ) for the imperfect Bose gas are shown to agree with those of the interacting Bose gas with
n internal degrees of freedom in the limit n → ∞. Hence, for d = 3, ΘP ∞,d (xξ ) is known exactly
in closed analytic form. To account for the breakdown of translation invariance in the direction
perpendicular to the boundary planes implied by free BCs such as DDBCs, a modified imperfect
Bose gas model is introduced that corresponds to the limit n → ∞ of this interacting Bose gas.
Numerically and analytically exact results for the scaling function ΘDD ∞,3 (xξ ) therefore follow from
those of the O(2n) φ4 model for n → ∞.

I. INTRODUCTION nite direction and chemical potentials µ < µc,id , where

µc,id = 0 is the bulk (D = ∞) critical value of µ at the
When a macroscopic system consisting of a medium in Bose-Einstein transition [8, 9]. Let
which long-wavelength low-energy excitations can occur ϕBC 1−d
ln ΞBC
d (T, µ, D) = − lim L d (T, µ, D, L) (1.1)
is confined along a given direction, fluctuation-induced L→∞
effective forces can emerge. These fluctuations can be of be the reduced grand potential per cross-sectional area
quantum mechanical or classical (i.e., thermal) nature. A, where ΞBC denotes the grand partition function
d
An example of the first kind of fluctuation-induced forces and the superscript BC indicates the type of bound-
are the Casimir forces [1] between two grounded parallel ary conditions chosen along the finite direction, e.g.,
metallic plates caused by the modification of the vac- BC = P, DD, NN, DN. We will also consider antiperi-
uum fluctuations of the electromagnetic field due to the odic (BC = A) and Robin BC (BC = R) [10–12]. The
presence of the plates. Familiar examples of fluctuation- latter will be specified below.
induced forces of the second kind are the critical Casimir Writing
forces that appear near continuous phase transitions with
a bulk critical temperature Tc > 0 [2–4]. ϕBC BC
d (T, µ, D) = Dϕb,d (T, µ) + ϕs,d (T, µ)
In confined quantum systems, generally both quantum + ϕBC
res,d (T, µ, D), (1.2)
and thermal fluctuations occur. At a conventional criti-
cal point with Tc > 0, quantum fluctuations are expected we decompose this reduced grand potential into a con-
to be irrelevant, i.e., they give only corrections to the tribution involving the reduced bulk potential ϕb,d , a D-
leading asymptotic behavior on large length scales [5, 6]. independent surface term ϕBC s,d , and a D-dependent re-
However, at sufficiently low temperature or near quan- mainder ϕBCres,d , which we call residual reduced potential
tum critical points, quantum fluctuations are crucial and [13, 14]. For the ideal Bose gas, one has the well-known
must not be neglected. An important prototype class of result (see, e.g., [15])
systems exhibiting both quantum and thermal fluctua-
ϕb,d (T, µ < 0) = −λ−d βµ


tions are Bose gases. In this paper we are concerned with th Lid/2+1 e . (1.3)
fluctuation-induced forces of ideal and interacting Bose
gases confined to a hypercuboid of size A × D of cross- Aside from the width D, there are two lengths in the
sectional hyperarea A = Ld−1 → ∞ and finite width D problem [16, 17]. One is the thermal de-Broglie wave-
in dimensions 2 < d < 4. length
Consider first the ideal Bose gas case. For d = 3 this
p
λth = ~ 2πβ/m, (1.4)
has been investigated in some detail in [7] for the cases
of periodic (P), Dirichlet-Dirichlet (DD), and Neumann- where m is the mass of the Bose particles and β =
Neumann (NN) boundary conditions (BCs) along the fi- (kB T )−1 . The other is the bulk correlation length ξ,
 2

which in the ideal Bose gas case is given by an n-component real-valued order parameter φ [19] for
the choice n = 2.
~ Following, we generalize these ideal Bose gas results in
ξid = p . (1.5)
2m(−µ) four ways:
(i) In addition to PBCs, DDBCs, and NNBCs, we also
The latter is finite or infinite depending on whether µ < 0 consider DNBCs, ABCs, and RBCs [10–12, 20]. For the
or µ = 0. On dimensional grounds, the residual potential latter the eigenfunctions h(z) of the operator −∂z2 are
can therefore be written in the form required to satisfy along the finite z direction the BCs
−(d−1) BC
ϕBC
res,d (T, µ, D) = D Υd (D/λth , D/ξ). (1.6) (∂z − c1 )h|z=0 = 0 = (∂z + c2 )h|z=D (1.15)
Let us also introduce the reduced Casimir pressure at the boundary planes z = 0 and D, respectively, where
c1 and c2 are general non-negative parameters. In this
∂ BC case, the scaling functions ΥR R
βFCBC = − ϕ (T, µ, D). (1.7) d and Θd also depend on
∂D res,d the two additional scaling variables c1 D and c2 D.
This can be written in the analogous scaling form (ii) We show the equality of ΥBCd (∞, xξ ) with the clas-
sical n = 2 free field scaling functions ΘBC d (xξ ) for all
βFCBC (T, µ, D) = D−d YdBC (D/λth , D/ξ), (1.8) d > 2 and BC = P, A, DD, DN, NN.
(iii) Summing the series of the functions ΘBC 3 (x) for
whose scaling function can be expressed in terms of YdBC BC = P, A, DD, DN, NN, we derive closed analytical ex-
and its derivatives as pressions for them in terms of polylogarithms.
  (iv) Finally, we determine ΥR d (D/λth , D/ξ, c1 D, c2 D)
∂ ∂ for all d > 2, generalizing previous ξ = ∞ results to
YdBC (xλ , xξ ) = d − 1 − xλ − xξ ΥBC
d (xλ , xξ ).
∂xλ ∂xξ ξ < ∞ [11, 12, 21].
(1.9) Note that unless we state the contrary, we will restrict
In [7], the ideal Bose gas functions ΥBC d=3 were com- ourselves in our analysis of the ideal Bose gas to the bulk
puted at d = 3 for the bulk disordered phase µ < 0 in disordered phase. Since for d ∈ (2, 3] a phase with long-
the form of double series. The asymptotic scaling forms range order is not possible for T > 0 and D < ∞, this is
of the functions ϕBCres,d for ξ → ∞ are expected to be in- not a severe restriction.
dependent of quantum effects. The dependence on the In extending our analysis to the interacting Bose gas
thermal de Broglie wavelength λth should drop out in case, we consider a model which is called imperfect Bose
the limit D/λth → ∞, so that Eqs. (1.6) and (1.8) must gas according to common, though debatable, terminol-
asymptotically reduce to ogy [22, 23]. This is a model for a gas of interacting
bosons in a region V whose interaction energy is approx-
ϕBC
res,d (T, µ, D) ' D−(d−1) ΘBC
d (D/ξ) (1.10) imated by aN̂ 2 /(2V ), where N̂ is the number operator.
λth D
In a recent paper [23] a d-dimensional model of such
and an imperfect Bose gas confined to a hypercuboid of size
Ld−1 × D and subject to periodic boundary conditions
βFCBC (T, µ, D) ' D−d ϑBC
d (D/ξ) (1.11) was investigated. Considering the appropriate thermo-
λth D
dynamic limits L, D → ∞ and L → ∞ at fixed D < ∞,
with the authors derived expressions for the bulk and residual
grand potentials. They found that the critical exponents
ΘBC BC
d (xξ ) = Υd (∞, xξ ) (1.12) that characterize the critical behavior at the transition
at fixed chemical potential for all d > 2 dimensions agree
and with those of the spherical model. This prompted them
to conclude that the universality class of the critical be-
ϑBC BC
d (xξ ) = Yd (∞, xξ ), (1.13) havior at the condensation transition of the imperfect
Bose gas is represented by the spherical model. On the
where
other hand, they found that at d = 3 the Casimir ampli-
d BC tude ∆C associated with the residual free energy at the
ϑBC BC
d (xξ ) = (d − 1) Θd (xξ ) − xξ Θ (xξ ) (1.14)
dxξ d critical point takes twice the value of its analog for the
spherical model [24, 25].
as a consequence of Eq. (1.9). We clarify this issue by showing that the bulk, sur-
As pointed out in [18], the results of [7], when appro- face, and residual potentials of the imperfect Bose gas
priately normalized, confirm this expectation and yield agree with the corresponding quantities for a system of
the functions ΘBC
d (xξ ) for d = 3 and PBCs, DDBCs, and interacting bosons with n internal degrees of freedom α
NNBCs in the form of series. The authors of [18] fur- and a pair potential of the form (ů/n) δαβ δ(x − x0 ) in
thermore showed that these functions agree with those the limit n → ∞. This latter model, henceforth called n-
previously determined for a free Gaussian theory with component interacting Bose gas, is defined in detail in the
 3

next section. Using a coherent-state functional-integral denote the operators of the kinetic and potential en-
formulation, we show that the leading asymptotic behav- ergy, respectively. The integration region V is the d-
ior near its bulk condensation transition is described by dimensional hypercuboid [0, L]d−1 × [0, D]. We write
the n = ∞ analogs ΘP ∞,d of the above scaling functions x = (y, z) and choose PBCs along the first d − 1 Carte-
ΘPd of the classical O(2n) real-valued φ4 theory. This sian directions y. Along the remaining z direction we
means, in particular, that for d = 3 the critical behavior consider PBCs, ABCS, DDBCs, NNBCs, DNBCs, and
of the imperfect Bose gas with PBCs is described by twice RBCs. From a physical point of view, DDBCs and RBCs
the exactly known scaling function of the mean spherical are the most relevant BCs; they correspond to the cases
model [24, 25]. of bosons confined along the z direction by infinitely or
In view of the equivalence of the imperfect Bose gas finitely high potential barriers, respectively. PBs are
with the interacting n-component Bose gas in the limit of interest because they are a preferred choice in nu-
n → ∞, it is natural to ask whether generalizations of the merical analyses. The remaining BCs (ABCS, NNBCs,
former model to other BCs such as free ones can be de- and DNBCs) are mainly of theoretical interest, though
fined so that this equivalence prevails. We introduce such some of them have been considered in the literature [7];
an imperfect Bose gas model with free BCs in Sec. IV D. whether and how they can be realized in experiments is
Its scaling function ΘDD
∞,3 (tD) may be obtained for all val-
not clear to us.
ues of the scaling field t ∝ µc − µ R 0 from the numerical The operators ψ̂(x) and ψ̂ † (x) are Bose field annihi-
lation and creation operators which for L = ∞ can be
solution of the O(∞) φ4 model determined in [26–29].
expressed as
Furthermore, a variety of exact analytical results may be
inferred from those known for the O(∞) φ4 model subject
dd−1 p
X Z
to DDBCs [30–35]. ψ̂ BC (y, z) = hBC
k (z) eip·y bp,k (2.4)
The remainder of this paper is divided into four addi- k
(2π)(d−1)/2
tional sections and three appendixes. Section II serves
to define the ideal, imperfect, and interacting Bose gas in terms of Bose annihilation and creation operators sat-
models with which we are concerned, including their BCs, isfying the commutation relations
and to recall their coherent-space functional-integral rep-
resentations. In Sec. III our results for the scaling func- [bp,k , b†p0 ,k0 ] = δ(p − p0 ) δk,k0 ,
tions of the ideal Bose gas in the disordered bulk phase [bp,k , bp0 ,k0 ] = [b†p,k , b†p0 ,k0 ] = 0. (2.5)
and their quantum corrections are presented. Section IV
deals with the n → ∞ limit of the n-component Bose gas, The hBC
k (z) are the orthonormalized eigenfunctions with
the equivalent imperfect Bose gas, and their scaling func- eigenvalues (k BC )2 of the operator −∂z2 for the specified
tions for PBCs and DDBCs. The exact scaling functions BCs. Explicitly, one has [3, 11, 12, 19]
ΘP P
∞,3 and ϑ∞,3 are determined in closed analytic forms
and shown to coincide with those of the O(∞) φ4 model 1 ikz 2π
hP
k (z) = √ e , k= ν, ν ∈ Z, (2.6a)
up to a factor of 2. For the case of DDBC, a number of D D
exact properties of the scaling functions ΘDD DD
∞,3 and ϑ∞,3
are deduced from known exact results for the latter clas-
sical model. Finally, Sec. V contains a brief summary of 1 ikz 2π
hA
k (z) = √ e , k= (ν + 21 ) ν ∈ Z, (2.6b)
our results and concluding remarks. D D

r
II. IDEAL, IMPERFECT, AND INTERACTING 2 π
BOSE GASES hDD
k (z) = sin(kz), k = ν, ν = 1, 2, . . . , ∞,
D D
(2.6c)
We consider an interacting Bose gas described by the (
Hamiltonian D−1/2 , k = ν = 0,
hNN
k (z) = p π
†
Ĥ = T [ψ̂ , ψ̂] + U [ψ̂ , ψ̂], †
(2.1) 2/D cos(kz), k=D ν, ν = 1, 2, . . . , ∞,
(2.6d)
where
r
~2 2 π
Z
†
T [ψ̂ , ψ̂] = d †
d x [∇ψ̂ (x)] · ∇ψ̂(x) (2.2) hDN
k (z) = sin(kz), k = (ν+ 12 ), ν = 0, 1, 2, . . . , ∞,
2m V D D
(2.6e)
and and
1
Z
† d d 0
d x d x ψ̂ † (x)ψ̂ † (x0 )

U [ψ̂ , ψ̂] =
r
2 sin(kz) + (k/c1 ) cos(kz)
2 hR
k (z) = p , k = kν ,
V×V DγkD 1 + k 2 /c21
×u(x − x0 )ψ̂(x0 )ψ̂(x)

(2.3) ν = 1, 2, . . . , ∞. (2.6f)
 4

In the latter case of RBCs, the discrete values kν ≡ kν /D [23, 38, 39]. This is analogous to the well-known rigorous
follow from the BC (1.15) at z = D. Here, the di- derivation of the van der Waals theory for a classical gas
mensionless kν are given by the zeros of the function of particles interacting through an attractive Kac pair po-
[11, 12, 21, 36] tential and a repulsive hard core [40], and hence reveals
the mean-field nature of the approximation to which the
Rc1 ,c2 (k) = (c1 c2 /k − k) sin(k) + (c1 + c2 ) cos(k), (2.7) model corresponds. The mentioned equivalence of the
imperfect Bose gas with an interacting Bose gas with n
where cj = cj D, j = 1, 2. Further, γk=kD denotes the internal degrees of freedom in the limit n → ∞ we are go-
normalization factor ing to prove below provides an even nicer justification of
c1 c1 the former model because the latter involves a physically
γk = 1 + 2 2
+ 2 , (2.8)
c1 + k c 1 + k2 reasonable short-ranged pair potential.
To define the interacting Bose gas with n internal de-
chosen such that the eigenfunctions are orthonormalized. grees of freedom (called interacting n-component Bose
The BCs of the above eigenfunctions satisfy, e.g., gas model), we replace ψ̂ by an n-component operator
hP P
hA A ψ̂ = (ψ̂α ) with α = 1, . . . , n and consider the Hamilto-
k (z) = hk (z + D), k (z) = −hk (z + D),
nian
hDD DD
k (0) = hk (D) = 0, hDN DN
k (0) = ∂z hk (D) = 0,
ů
Z
(2.9) †
Ĥn = T [ψ̂ , ψ̂] + dd x ψ̂α† (x)ψ̂β† (x) ψ̂β (x)ψ̂α (x),
2n
(2.14)
and Eq. (1.15), carry over to the field operators ψ̂(x) and
where pairs of internal indices α, β are to be summed
ψ̂ † (x). The commutation relations of the latter are from 1 to n. It will become clear below that the imperfect
[ψ̂(x), ψ̂ † (x0 )] = δ(x − x0 ), Bose gas is equivalent to this model in the limit n → ∞.
Let Ξ(T, µ, D, L) denote the grand partition function
[ψ̂(x), ψ̂(x0 )] = [ψ̂ † (x), ψ̂ † (x0 )] = 0. (2.10)
Following common practice [17, 37], we take the pair Ξ( T, µ, D, L) = Tr e−β(Ĥ−µN̂ ) (2.15)
potential u(x) to be short ranged and of the form
of any of these models with Hamiltonian Ĥ. With a view
4π d/2 ~2 ad−2 to our subsequent analysis it will be helpful to recall its
s
u(x) = ů δ(x), ů = , (2.11) coherent-state path-integral representation. To this end,
Γ(d/2 − 1) m
we introduce complex-valued fields ψ(x, τ ) = (ψα (x, τ ))
where as is the s-wave scattering length [17]. and ψ ∗ (x, τ ) = (ψα∗ (x, τ )) satisfying PBCs ψ(x, τ ) =
We will be concerned with three different Bose gas ψ(x, τ + β~) and ψ ∗ (x, τ ) = ψ ∗ (x, τ + β~) in imaginary
models on a film [0, L]d−1 × [0, D]: the ideal Bose gas, time τ . Owing to these BCs, the Bose fields ψ(x, τ ) and
the imperfect Bose gas model investigated in [23], [38], ψ ∗ (x, τ ) can be decomposed into Fourier series (see, e.g.,
and [39], and the interacting Bose gas with n internal [41])
degrees of freedom in the limit n → ∞. In the case of 1 X 2π
the ideal Bose gas, the interaction u(x) is zero. The im- ψα (x, τ ) = ψα,ρ (x) e−iωρ τ , ωρ = ρ, (2.16)
β~ β~
perfect Bose gas model results from the above specified ρ∈Z
interacting one if the potential energy term U [ψ̂] is ap-
involving the bosonic Matsubara frequencies ωρ and the
proximated by UimpBG = aN̂ 2 /2V , where
Fourier coefficients
Z Z β~
N̂ = dd x ψ̂ † (x)ψ̂(x) (2.12) ψα,ρ (x) = dτ ψα (x, τ ) eiωρ τ . (2.17)
V
0
is the number operator and V = AD means the hyper- For later use, we split ψα (x, τ ) into its τ -independent
volume of the system. Thus the Hamiltonian of the im- (ρ = 0) part Ψα (x) ≡ (β~)−1 ψα,0 (x) and a remainder
perfect Bose gas reads as Ψ6=
α (x, τ ), writing [42–44]

N̂ 2 ψα (x, τ ) = Ψα (x) + Ψ6=

ĤImpBG = T [ψ̂ † , ψ̂] + a . (2.13) α (x, τ ). (2.18)
2V
All BCs along the z direction considered above for the
This model looks unphysical in that each of the N (N − operators ψ̂α (y, z) and ψ̂α∗ (y, z) carry over to the Bose
1)/2 = N 2 /2 + O(N ) pairs of bosons gives the same con- fields ψα (y, z, τ ) and ψα∗ (y, z, τ ).
tribution to UimpBG independent of the separation of the The coherent-state path-integral representation of the
two bosons of each pair. It can be obtained by taking the grand partition function reads as
limit γ → 0 of a repulsive integrable Kac-type pair poten-
tial such as uγ (x) = γ d e−γx whose strength and inverse
Z
∗

range are both controlled by the same parameter γ > 0 Ξ(T, µ, D, L) = D[ψ ∗ , ψ] e−S[ψ ,ψ]/~ , (2.19)
 5

where the action S[ψ ∗ , ψ], in the case of the interacting gleaned from [7]. Since their n > 1 analogs follow upon
Bose gas with Hamiltonian Ĥn , is given by multiplication by n, we set n = 1 unless stated otherwise
(see Sec. IV). In Appendix A we present a slightly dif-
S[ψ ∗ , ψ] = S0 [ψ ∗ , ψ] + Sint [ψ ∗ , ψ] (2.20a) ferent calculation of ΥP A
d , determine Υd , and recapitulate
with how ΥDDd and Υ NN
d can be computed. The results read
Z β~  Z  as
S0 [ψ ∗ , ψ] = dd x ψα∗ (x, τ ) ~∂τ
dτ
0 V ΥP
d (xλ , xξ )
~2 2
 
∞ ∞
− ∇ − µ ψα (x, τ ) (2.20b) X X 2 2 2 2
2m = −2xdλ s−d/2−1 e−πj xλ /s−sxξ /4πxλ
and s=1 j=1
β~ ∞
ů
Z Z X 2 2 2
= −xdλ s−d/2−1 e−sxξ /4πxλ ϑ3 0, e−πxλ /s − 1 , (3.1)


Sint [ψ ∗ , ψ] = dd x ψα∗ (x, τ )ψβ∗ (x, τ )

dτ
2n 0 V s=1

×ψβ (x, τ )ψα (x, τ ) . (2.20c)
Near the bulk critical temperature, the part of the ac-
tion involving Matsubara frequencies ωρ 6= 0 is expected ΥA
d (xλ , xξ )
∞ ∞
to give only exponentially small corrections. In fact, we X X 2 2 2 2

will verify√ below that these corrections are down by fac- = −2xdλ s−d/2−1 (−1)j e−πj xλ /s−sxξ /4πxλ
s=1 j=1
tors e−2π 2 D/λth at the bulk critical point√
for PBCs and
∞
−4π 2 D/λth
ABCs, but smaller by its square e for free X 2 2 2
= −xdλ s−d/2−1 e−sxξ /4πxλ ϑ4 0, e−πxλ /s − 1 , (3.2)


BCs such as DDBCs, DNBCs, NNBCs, and RBCs. Ig-
s=1
noring these ωρ 6= 0 parts of the action by making the
replacement ψ → Ψ, we arrive at the “classical” action and
S[Ψ∗ , Ψ], which can be conveniently written in terms of
a rescaled real-valued (2n)-component order-parameter ΥDD NN
d (xλ , xξ ) = Υd (xλ , xξ )
field φ = (φα2 )2nα2 =1 formed from the real and imaginary ∞ ∞
parts of Ψα such that X X 2 2 2 2

√ = −2xdλ s−d/2−1 e−4πj xλ /s−sxξ /4πxλ

2π s=1 j=1
Ψα (x) = [φ2α−1 (x) + iφ2α (x)], α = 1, . . . , n. ∞
λth X 2 2 2
= −xdλ s−d/2−1 e−sxξ /4πxλ ϑ3 0, e−4πxλ /s − 1 ,(3.3)


(2.21)
One finds s=1

cl ∗
S [φ] ≡ S[Ψ , Ψ] P∞ j 2 2πiz
  where ϑ3 (z, q) = j=−∞ q e and ϑ4 (z, q) =
1 1 −2 2 g/n 4
Z P∞
d 2 j j 2 2πiz
= d x φ · (−∇ φ) + ξid φ + φ j=−∞ (−1) q e are Jacobi theta functions [45, 46].
V 2 2 4! P∞
(2.22) In the limit xλ → ∞, the R∞sums s=1 . . . become Rie-
mann sums for integrals 1 ds . . .. Furthermore, the
with lower integrationR 1 limit can be changed from 1 to 0 since
12m2 ů 96 ad−2 π d/2+1 the differences 0 ds . . . vanish in the limit xλ → 0. Per-
g= 4
= 2 s . (2.23) forming the integrals yields
β~ λth Γ(d/2 − 1)
2 1/2
P2n
, (∇φ)2 stands for ΘP


Here φ = α2 =1 φα2 d (xξ )
P2n 2 ∞
α2 =1 (∇φα2 ) , and we have added the subscript id to d/2
X
emphasize that ξid is the ideal Bose gas quantity defined = −4(2π)−d/2 xξ j −d/2 Kd/2 (jxξ ) (3.4a)
in Eq. (1.5). j=1
∞
2 Kd−1 d (t2 − 1)(d−1)/2
Z
=− x dt , (3.4b)
d−1 ξ 1 etxξ − 1
III. SCALING FUNCTIONS OF THE IDEAL
BOSE GAS

A. The cases of periodic, antiperiodic, ΘA

d (xξ )
∞
Dirichlet-Dirichlet, Dirichlet-Neumann, and d/2
X (−1)j+1
Neumann-Neumann boundary conditions = 4(2π)−d/2 xξ Kd/2 (jxξ ) (3.5a)
j=1
j d/2
∞
2 Kd−1 d (t2 − 1)(d−1)/2
Z
For the case of the ideal Bose gas (with n = 1 compo-
= x dt ., (3.5b)
nents) the scaling functions ΥP DD
d , Υd , and Υd
NN
can be d−1 ξ 1 etxξ + 1
 6

ΘBC
and 3

A
ΘDD NN
d (xξ ) = Θd (xξ )
∞
d/2
X
= − 4(2π)−d/2 xξ j −d/2 Kd/2 (2jxξ ), (3.6a)
j=1 DN
xξ
∞ 2 (d−1)/2
2 Kd−1 d (t − 1)
Z
DD, NN
= − x dt , (3.6b)
d−1 ξ 1 e2txξ − 1
where
2
Kd ≡ , (3.7) P
(4π)d/2 Γ(d/2)
while Kd/2 (x) is a modified Bessel function of the second
kind. FIG. 1. Classical scaling functions ΘBC
3 (xξ ) for n = 1 and
For DNBCs, the corresponding result reads as BC = P, A, DD, NN, and DN. Their values at xξ are the
corresponding Casimir amplitudes ∆BCd .
ΘDN
d (xξ )
∞
d/2
X (−1)j+1
= 4(2π)−d/2 xξ Kd/2 (2jxξ ) (3.8a)
j=1
j d/2
They read as
∞
2 Kd−1 d (t2 − 1)(d−1)/2
Z
= x dt . (3.8b)
d−1 ξ 1 e2txξ + 1 1 2
ϑP xξ ln 1 − e−xξ − 2 Li3 − e−xξ



3 (xξ ) =
This result most easily follows by setting (c1 , c2 ) = (∞, 0) π
− 2xξ Li2 e−xξ ,


in the result for RBCs derived in the next subsection [see (3.12a)
Eqs. (3.56) and (3.58)] and integrating by parts. 1
ϑA x2 ln 1 + e−xξ − 2 Li3 e−xξ



The integral forms given in Eqs. (3.4b)–(3.8b) follow 3 (xξ ) =
π ξ
from Eq. (6.8) of [19] upon setting n = 2 there. To − 2xξ Li2 − e−xξ ,


(3.12b)
verify that the series (3.4a)–(3.8a) can be summed in this
1
manner, one can substitute the expansion ϑDD x2 ln 1 − e−2xξ − Li3 e−xξ



3 (xξ ) =
∞
4π ξ
− 2xξ Li2 e−2xξ ,


(3.12c)
X
[etx ∓ 1]−1 = (±1)j+1 e−jtx (3.9)
j=1 1
ϑDN 2x2ξ ln 1 + e−2xξ − Li3 − e−xξ



3 (xξ ) =
with x = xξ or 2xξ into these integrals and integrate 4π
− 2xξ Li2 − e−2xξ ,


termwise. (3.12d)
It should be obvious that the above results for ΘBC
d (xξ )
with BC = P, A, DD, NN, and DN are identical to those and are plotted in Fig. 2.
that follow from the g = 0 analog of S cl [φ].
When d = 3, the results for ΘBCd given in Eqs. (3.4a)– ϑBC
3
(3.8b) can be expressed in closed form in terms of poly- A
logarithms [45, 46]; one has
1
ΘP
3 (xξ ) = − [Li3 (e−xξ ) + xξ Li2 (e−xξ )], (3.10a)
π DN
1 −xξ xξ
ΘA
3 (xξ ) = − [Li3 (−e ) + xξ Li2 (−e−xξ )], (3.10b)
π DD, NN

ΘDD NN
3 (xξ ) = Θ3 (xξ )
1
= − [Li3 (e−2xξ ) + 2xξ Li2 (e−2xξ )], (3.10c)
8π
1 P
ΘDN
3 (xξ ) = − [Li3 (−e−2xξ ) + 2xξ Li2 (−e−2xξ )].
8π
(3.10d)
FIG. 2. Classical scaling functions ϑBC
3 (xξ ) for n = 1 and
These functions are plotted in Fig. 1. BC = P, A, DD, NN, and DN.
The corresponding results for the scaling functions of
the Casimir force follow in a straightforward fashion via
d BC The Casimir amplitudes ∆BC = ΘBC
d (0) that follow
ϑBC BC
d (xξ ) = (d − 1) Θd (xξ ) − xξ Θ (xξ ). (3.11) d
dxξ d from the above Eqs. (3.4a)–(3.8b) and (3.10a)–(3.10d)
 7

for the scaling functions, which can be expressed as

−d/2
∆P
d = −2π Γ(d/2) ζ(d) 1 1
ρ(E; D, c1 , c2 ) = − Im T (E + i0) ≡ − Im Tr G(E + i0)
= −ζ(3)/π, (3.13a) π π
d=3 (3.15)
∆A
d = 2(1 − 21−d )π −d/2 Γ(d/2) ζ(d) in terms of the trace T (z) ≡ T (z; D, c1 , c2 ) of the Green’s
= 3 ζ(3)/4π, (3.13b) function
d=3
−1
G(z) ≡ G(z; D, c1 , c2 ) = z + ∂z2 , (3.16)
∆DD
d = ∆NN
d = −21−d π −d/2 Γ[d/2] ζ(d)
where the operator ∂z2 is subject to the RBCs specified
= −ζ(3)/8π, (3.13c) in Eq. (1.15).
d=3
With the aid of the generalized function ρ(E; D, c1 , c2 )
and the reduced grand potential ϕR d can be written as
∆DN
d = (1 − 21−d )21−d π −d/2 Γ(d/2)ζ(d)
ϕR
d (T, µ, D, c1 , c2 )
= 3 ζ(3)/32π, (3.13d) Z ∞
d=3  2

= −λ1−d
th dE Li d+1 eβµ−λth E/4π ρ(E; D, c1 , c2 ).
agree with the familiar one-loop results for the O(2) 0
2

massless φ4 theory (cf. Eq. (5.7) of [19] and Eq. (4.4) (3.17)
of [47]).
The function T (z; D, c1 , c2 ) is given by
∞
B. Case of Robin boundary conditions X
−1
T (z; D, c1 , c2 ) = z − kν2 , z ∈ C \ [0, ∞), (3.18)
ν=1
We now turn to the case of Robin boundary conditions and has the scaling property
with c1 ≥ 0 and c2 ≥ 0. We assume that c1 + c2 > 0,
so that not both variables vanish simultaneously. In this T (z; D, c1 , c2 ) = D2 T (D2 z; 1, Dc1 , Dc2 ). (3.19)
case, the discrete k values kν , ν = 1, 2, . . . , ∞, introduced
Recalling that the scaled momenta kν = Dkν are the
in Eq. (2.6f) are all positive. It is convenient to introduce
zeros of the function Rc1 ,c2 (k) and that this
√ function is
the density of states
even in k, one concludes that ∂ζ ln Rc1 ,c2 ( ζ) has simple
∞
X poles at ζ = k2ν with residues 1 and therefore agrees with
ρ(E; D, c1 , c2 ) = δ(E − kν2 ), E > 0, (3.14) T (ζ; 1, c1 , c2 ). Hence,
ν=1

√ √ √
1 (ζ − c1 c2 ) ζ cos ζ + [c1 c2 + (1 + c1 + c2 )ζ] sin ζ
T (ζ; 1, c1 , c2 ) = − √ √ √ , (3.20)
2ζ (c1 + c2 ) ζ cos ζ + (c1 c2 − ζ) sin ζ
√
where the branch cut of ζ is taken along the positive real axis.

Choosing Im ζ > 0, we can expand the result in powers The reader may want to check that the result (3.22) is
of D about D = ∞.We thus obtain the decomposition consistent with what one gets from Eqs (3.15) and (3.16)
2 using the fact that Tb (z) is independent
R ∞ dk of 2the BC along
with limD→∞ D−1 Tr f (−∂z2 ) = −∞ 2π f (k ). The asso-
X
T (z; D, c1 , c2 ) = D Tb (z)+ Ts (z; cj )+Tres (z; D, c1 , c2 )
j=1
ciated bulk density of states becomes
(3.21) 1
into bulk, surface, and residual contributions with ρb (E) = √ . (3.24)
2π E
−1
Tb (z; D, c1 , c2 ) ≡ lim D T (z; D, c1 , c2 )
D→∞ One can easily check that Eq. (1.3) for ϕb,d is recovered
−i from Eq. (3.17) upon substituting ρ(E) by the foregoing
= √ (3.22)
2 z result for ρb (E).
and In order to determine the surface quantity
Ts (z; c) ≡ lim [T (z; D, c, c) − D Tb (z)]/2, 1
D→∞ ρs (E; c) = − Im Ts (E + i0; c) (3.25)
√ π
1 z − ic
= √ . (3.23)
4z z + ic from Eq. (3.23), the familiar Sokhatsky-Weierstraß iden-
 8

tity as Fisher-renormalized exponents [17, 48] of the Gaussian

model; i.e., the specific-heat, order-parameter, suscepti-
1 P bility, and correlation-length exponents αid , βid , γid , and
= − iπ δ(E) (3.26)
E + i0 E νid , respectively, follow from their Gaussian counterparts
is needed, where P denotes the principal value. One finds αG ,. . . via the relations
νG 1 1
θ(E) c 1 νid = = = γid /2, νG = ,
ρs (E; c) = √ − δ(E), (3.27) 1 − αG d−2 2
2
2π E c + E 4 βG 1 d−2 d−2
βid = = , β G = νG = ,
where we have included the Heaviside function θ(E) in 1 − αG 2 2 4
the first term, which originated from the principal-value −αG d−4 d
αid = = , αG = 2 − . (3.33)
term on the right-hand side of Eq. (3.26), to emphasize 1 − αG d−2 2
that this distribution is integrated only over the positive The asymptotic critical behavior is known to be purely
real axis. By contrast, the E integration involving δ(E) classical. The bulk universality class is that of a Gaus-
is to be extended over the full real axis, so that its action sian model for a two-component real-valued order param-
on a test function g(E) yields the usual result g(0). eter with a mass term ξid−2
∼ (−µ) → 0, but the above-
The result (3.27) can be substituted into Eq. (3.17) mentioned renormalization of the critical exponents due
to determine √ϕs,d . Upon transforming to the integration to the fixed-density constraint must be taken into ac-
variable k = E and exploiting the evenness of the inte- count. The length λth (T ) should drop out from the
grand in k, one arrives at asymptotic critical behavior in appropriately normalized
quantities. Let us therefore determine the limiting be-
ϕR
s,d (T, µ, c)
 havior of ϕRs,d for λth → 0. One possibility is to use the
1 small-x expansion [49]
= λ1−dth Li d+1 (eβµ )
4 2
Z ∞  Li(d+1)/2 (1 − x)
dk  2 2
 c
− Li d+1 eβµ−λth k /4π 2
(
. (3.28) ζ( d+1 ) + ζ( d−1 ) x + Γ( 1−d
d−1
)x 2 + ... for 3 6= d,
−∞ 2π
2 c + k2 = 1 2 2 2 2
6 π − x(1 − ln x) + . . . for d = 3.
In the limit c → ∞ (corresponding to DBCs), the contri-
(3.34)
bution from the integral vanishes. Thus
for the polylogarithms in Eq. (3.28). In order to bene-
1 1−d
ϕD R
s,d (T, µ) ≡ ϕs,d (T, µ, ∞) = λ Li d+1 (eβµ ). (3.29) fit from dimensional regularization we insert the above
4 th 2
expansion for general d ∈ (2, 4) into the integral in
To determine ϕR Eq. (3.28). This leads us to
s,d (T, µ, 0) (corresponding to NBCs), one
can use ϕR R R
s,d (T, µ, c) = ϕs,d (T, 0) + fs,d (ξ, c) + o(λth ) (3.35)
c
−−−→ πδ(k) (3.30) with
c2 + k 2 c→0
1 1−d  d + 1 
to find ϕR
s,d (T, 0; c) = − λ ζ (3.36)
4 th 2
1 1−d and
ϕN R
s,d (T, µ) ≡ ϕs,d (T, µ, 0) = − λth Li d+1 (eβµ ). (3.31)
4 2
 1 − d  1
R
These results (3.29) and (3.31) are consistent with the fs,d (ξ, c) =Γ (4πξ 2 )(1−d)/2
2 4
integral expressions given in Eq. (11) of [7] for the total Z ∞ (d−1)/2
dk k 2 + ξ −2
 
surface contributions of ϕBC c
d=3 for DDBCs and NNBCs. − ,
They also imply that the surface contribution of ϕd van- 0 π 4π k 2 + c2
ishes for DNBCs. (3.37)
At the Bose-Einstein transition in d > 2 dimensions,
where ξ again means ξid and the omitted o(λth ) terms
the correlation length ξ = ξid diverges ∼ (−µ)−1/2 as
contain quantum corrections.
µ → 0− and becomes much larger than the thermal de
The result requires two comments. First, the integral
Broglie wavelength λth (T ≈ Tc ). If the approach to crit-
in Eq. (3.28) converges in the ultraviolet (UV) because
icality occurs along a temperature path at fixed density
the k integration is smoothly cut off at kλ ≈ 4π/λth .
ρ = −∂ϕb,d /∂(βµ), then −µ ∼ δT 2/(d−2) , where
However, since λth drops out from the integral of the
R
δT = [T − Tc (ρ)]/Tc (ρ), (3.32) expansion term associated with fs,d , a UV cutoff is no
longer present in it. Convergence of this integral is en-
so that ξid ∼ δT −νid with νid = (d − 2)−1 . sured only for d < 2. For 2 < d < 4, it must be reg-
The correlation-length exponent νid and the other crit- ularized either by reintroducing a UV cutoff or else di-
ical exponents of the ideal Bose gas may be understood mensionally. We prefer to use dimensional regularization.
 9

Assuming that d < 2, the integral in Eq. (3.37) can be for the surface energy density, where we adopted the con-
computed by means of Mathematica [50]. One obtains venient notation
dd−1 p
R
Z (d−1) Z
fs,d (ξ, c) = cd−1 fs,d (ξc, 1)
= d−1
. (3.44)
1 − d  1 (c2 ξ 2 − 1)(d−1)/2 Rd−1 (2π)
   
1 1−d 1−d p
= ξ (4π) 2 Γ −
2 2 2 sin(πd/2) Integrating the result with respect to ξ −2 yields
√
π 2 F̃1 [1/2, 1; d/2 + 1; (cξ)−2 ]

Z (d−1) 
+ , (3.38) d−1 1 + κ 
cξ sin(πd/2) fs,d (ξ, c) = c ln √ 
p κ κ=κp (ξc)
where 2 F̃1 (a, b; c; z) is the regularized hypergeometric = cd−1 Kd−1 Jd (ξc) (3.45)
function 2 F1 (a, b; c; z)/Γ(c). The result provides the an-
alytic continuation to dimensions d ≥ 2. with
A second necessary comment is that this expression ∞  
1 + κp (v)
Z
does not reproduce the c → ∞ limit of fs,d R
given by the Jd (v) = dp pd−2 ln p
0 κp (v)
first term in Eq. (3.37) (since the regularized integral van- ∞
1 dp pd κp (v) − 1
Z
ishes for c → ∞). The reason is that the sequence of the = , (3.46)
limits in which the lengths 1/c and λth go to zero (or the 1−d 0 2κ2p (v) κp (v) + 1
cutoff kλ → ∞) do not commute. This noncommutabilty
where the second form follows upon integration by parts.
of the limits c → ∞ of the bare surface-enhancement vari-
The latter integral, which is UV divergent for dimen-
able c and the UV momentum cutoff is well known from
sions d > 1, can be analytically continued to d > 2 in
the classical theory [10].
a straightforward manner. In the first form, Jd (v) is
However, the result given in Eq. (3.37) yields the cor-
IR divergent for d < 1 and UV divergent for d > 1.
rect limits c → 0 with ξ > 0 and ξ → ∞ with 0 < c < ∞,
To avoid the IR divergence, one can subtract from the
namely,
integrand’s R ∞logarithm its value at p = 0, using the
π cd−1 fact that 0 dp pd−2 = 0 in dimensional regularization.
R
fs,d (∞, c) = −Kd−1 (3.39) The resulting integral thereby becomes well-defined for
d − 1 sin(dπ) d ∈ (−1, 1) and can be analytically continued.
and Consistency with the results for (ξ, c) = (∞, c) and
(ξ, 0) given in Eqs. (3.39) and (3.40) is easily checked
R Kd−1 π by performing the required integrals. A proof that
fs,d (ξ, 0) = − ξ 1−d . (3.40) Eqs. (3.45) and (3.46) are consistent with the integral
d − 1 4 cos(πd/2)
representation (3.37) is harder and relegated to Ap-
Equation (3.39) complies with the result obtained in [11], pendix B.
R
[12], and [21]. Further, Eq. (3.40) can be confirmed easily Both ways of calculating fs,d used above can be gener-
via Eq. (3.37) using Eq. (3.30). alized to determine the quantum corrections of ϕR s,d . To
The result (3.37) can be checked by means of a purely generalize the second, note that the Fourier transform
classical calculation. Consider the free classical theory with respect to y and τ of the propagator hψ(y, z, τ ) ·
associated with the action (2.22) with g = 0 for the semi- ψ ∗ (0, z 0 , 0)i/2n associated with the action (2.20b) can
infinite system z ≥ 0 and subject to the RBC (1.15) be written as
with c ≡ c1 at z = 0. The Fourier transform of the
2m
corresponding free propagator hφ(y, z)·φ(0, z 0 )i/2n with gρ (p; z, z 0 ) = Ĝ[κp,ρ (ξ, λth ); z, z 0 ; c] (3.47)
respect to the y coordinate is given by ~2
with
1 −κ|z−z0 | c − κ −κ(z+z0 )
 
Ĝ[κp (ξ); z, z 0 ; c] = e − e
q
2κ c+κ κ=κp (ξ) κp,ρ (ξ, λth ) = p2 + ξ −2 − i8π 2 ρ/λ2th . (3.48)
(3.41)
where We thus arrive at the expansion
p ϕR
s,d (T, µ, c)
κp (ξ) = p2 + ξ −2 . (3.42)
R
= ϕs,d (T, µ, 0) + fs,d (ξ, c)
Using this leads us to X∞ h
−1/2 i
+ 2cd−1 Kd−1 ReJd c ξ −2 − i8π 2 ρ/λth .
Z (d−1) Z ∞ h i
ρ=1
es,d (ξ, c) = dz Ĝ[κp (ξ); z, z; c] − (z = ∞)
p 0 (3.49)
(d−1)
1 κ−c
Z  
= (3.43) Here, the first (c = 0) term can be expanded about
p 4κ2 κ + c κ=κp (ξ) µc = 0 to obtain contributions analytic in µc − µ. The
 10

R
√ R0
second term, fs,d , describes the asymptotic critical be- where Im E > 0. The integrals −∞ duu . . . are con-
P∞
havior. Finally, the sum ρ=1 . . . contains the quantum vergent for d < 2 and defined for d > 2 by analytic
corrections. They are exponentially small. continuation.
An alternative way of calculating ϕR
s,d (T, µ, c) is to go The residual potential ϕR
res,d can be computed in a sim-
back to the integral representation of ϕR ilar fashion using
s,d in terms of
Ts (E; c):
dE h
Z
2 E+ξ−2

ϕR ϕR 1−d
res,d (T, µ, c1 , c2 ) = −λth Li d+1 e−λth 4π
s,d (T, µ, c) C1 2πi
2

dE
Z
2 E+ξ−2
  i
= −λ1−d
th Li d+1 e−λth 4π Ts (E; c). (3.50) ×Tres (E; D, c1 , c2 ) . (3.54)
C1 2πi 2

Here C1 is the contour in the complex E plane de- Since Tres decays exponentially as E → ∞, the inte-
picted in Fig. 3. The branch point of the polylogarithm grand needs no subtraction to deform the contour C1 into
Li(d+1)/2 (z) at z = 1 yields infinitely many branch points ∪∞j=−∞ γj . Exploiting again Eq. (3.52), one finds that the
in the complex energy plane located at
scaling function ΥRd can be written as
Ej = −ξ −2 + i8π 2 λ−2
th j, j ∈ Z. (3.51)
The thick lines in the figure denote the associated branch ΥR
d (xλ , xξ , c1 , c2 )
∞ Z
cuts. −(4π)(1−d)/2 X 0
du (−u)(d−1)/2

=
Im E Γ[(d + 1)/2] j=∞ −∞
γ1
×Tres (u − x2ξ + i8π 2 jx2λ ); 1, c1 , c2 ) .

(3.55)
ε1

γ0
Re E As we show in Appendix C, this result can be trans-
ε0 k12 k22 formed into
γ−1 C1
−ic
ε−1 ΥR
d (xλ , xξ , c1 , c2 )
Z ∞ 
dp pd−2 gc1 ,c2 κp (1/xξ )
 
FIG. 3. Deformation of the contour C1 into the union of = Kd−1
0
contours γj , j ∈ Z plus a circle around the isolated pole at ∞ 
E = −ic. X  
+2 Re gc1 ,c2 κp,ρ (1/xξ , 1/xλ ) , (3.56)
ρ=1

We now wish to deform the contour C1 into the union

∞ where
of contours ∪ γj . In order to be able to do this, we
j=−∞
(κ − c1 )(κ − c2 ) −2κ
R  
must first ensure that the integrand of C1 dE decays suf- gc1 ,c2 (κ) = ln 1 − e (3.57)
ficiently fast at E = ∞. This can be achieved by adding (κ + c1 )(κ + c2 )
and subtracting from Ts (E; c) its asymptote for E → ∞
given by Ts (E; 0) = (4E)−1 . We can then deform C1 in and κp,ρ (ξ, λth ) was defined in Eq. (3.48). In Appendix C,
the intended fashion. The polylogarithms have jump dis- we also present a somewhat easier, alternative calcula-
continuities across the branch cuts separating the upper tion, which directly leads to Eq. (3.56).
and lower rims of the contours γj . These discontinuities The first contribution to ΥR d , namely, the ρ = 0 analog
are implied by the jump of the remaining ones, is the classical scaling function
z=x+i0 2πi ln(d−1)/2 (x) Z ∞
Li d+1 (z)z=x−i0 = for x > 1. (3.52) ΘR dp pd−2 gc1 ,c2 [κp (1/xξ )].
2 Γ[(d + 1)/2] d (xξ , c1 , c2 ) = Kd−1
0
The functions Ts (E; c) −√Ts (E, 0) are continuous across (3.58)
these branch cuts since E (whose only branch cut is In the special case xξ = 0, it reduces to its analog at the
along the positive real axis) has this property. We thus bulk critical point, the scale-dependent amplitude
obtain
∆R R
d (c1 , c2 ) ≡ Θd (0, c1 , c2 )
ϕR
s,d (T, µ, c) Z ∞
−1 1−d 1 X∞ Z 0  = Kd−1 dp pd−2 gc1 ,c2 (p) (3.59)
λth Li d+1 eβµ −


= du 0
Γ d+1


4 2
2 j=−∞ −∞
 −u  d−1    obtained in [11, 12, 20]. Likewise, Eqs. (3.6a)–(3.8b) can
2 1 c be recovered from the result (3.58) in the cases xξ > 0
√ , (3.53)
4π 2E i E − c E=εj +u and (c1 , c2 ) = (∞, ∞), (0, 0) and (∞, 0).
 11

C. Quantum corrections to the critical behavior IV. INTERACTING n-COMPONENT BOSE GAS

The results given in Eqs.. (3.56) and (3.57) enable us

to determine the form of the leading quantum correc- A. The limit n → ∞
tions. They are associated with the contributions from
the Matsubara frequencies ωρ with ρ = ±1. Let We now turn to the interacting n-component Bose
q gas. In order to investigate its limit n → ∞, we use the
γ(xξ , xλ ) ≡ κ0,−1 (1/xξ , 1/xλ ) = x2ξ + i8π 2 x2λ Hubbard-Stratonovich transformation
r Z ∞
= |x2ξ + i8π 2 x2λ |1/2 ei arctan(8π
2
x2λ /x2ξ )/2
. (3.60) −(ů/2n)|ψ|4 n 2 2
e = dχ e−i χ| ψ| −(n/2ů)χ (4.1)
2πů −∞
We expand the function gc1 ,c2 (κp,−1 ) to linear order
in e−2κp,−1 , write the corresponding expansion term to rewrite the interaction term of the action (2.20). It fol-
2 2
as A e−2γ(1+2p /γ ) , and then expand the coefficient to lows Rthat e−S[ψ∗,ψ]/~ can be written
as a functional inte-
O(γ −1 , p4 /γ 3 ). Upon performing the p integral, we ar- gral D[χ] exp − Seff [ψ ∗ , ψ, χ]/~ over the real-valued
rive at the expansion field χ(x, τ ). The effective action is given by

~2 2
Z   
ΥR R
d (xλ , xξ , c1 , c2 ) − Θd (xλ , xξ , c1 , c2 ) Seff [ψ ∗ , ψ, χ] = ψα∗ ~∂τ − µ + iχ − ∇ ψα
   d−1 τ,V 2m
γ 2
e−2γ 1 + O e−2γ


= −2Re

n 2
γ→∞ 4π + χ , (4.2)
2ů
d2 − 1
 
c1 + c2 −2


× 1+ −2 +O γ . (3.61) R
16γ γ up to an unimportant constant. Here, τ,V stands for
R β~ R d
At the bulk critical point, where xξ = 0, we have 0
dτ V d x. Upon integrating out the fields ψ ∗ and ψ
and introducing the potential
|γ| = 23/2 πxλ . Thus, the quantum corrections to
the Casimir amplitude ∆R d (c1 , c2 ) are down by a factor V(x, τ ) = iχ(x, τ ) − µ, (4.3)
exp(−25/3 πD/λth ).
An analogous analysis of quantum corrections can be one arrives at
made in the cases of the PBCs and ABCs. To see this, Z
note that the classical scaling functions ΘP,A
∗
d can be writ- ln D[ψ ∗ , ψ, χ] e−Seff [ψ ,ψ]/~ + const
ten in a form analogous to Eqs. (3.58) using integration
~2 2 i
Z
by parts. Adding the quantum corrections resulting from
nD  h E
= −n x, τ  ln ~∂τ + V − ∇ x, τ

the Matsubara frequencies ωρ 6= 0 then gives τ,V 2m
1 o
Υd
P/A
(xλ , xξ )/n − (µ + V)2 /~, (4.4)
Z ∞  2ů
dp pd−2 ln 1 ∓ e−κp (1/xξ )
 
= ±2Kd−1 where a self-explanatory Dirac notation is used.
0 The explicit factor of n in the result tells us that the
∞
functional integral over χ for n → ∞ can be calculated
X 
−κp,ρ (1/xξ ,1/xλ )
 
+2 Re ln 1 ∓ e , (3.62) by evaluating the integrand at the stationary point. The
ρ=1
stationary potential V∗ (x) = V(x) (where the subscript
asterisk indicates values at the stationary point) is in-
where the upper (lower) signs refer to BC = P and A,
dependent of τ by translation invariance along the time
respectively. It follows that the leading quantum correc-
direction. Since we also have spatial translation invari-
tions are of the form
ance along the y directions, it is sufficient to consider
potentials that depend on z or are independent of z in
ΥP,A P,A
d (xλ , xξ )/n − Θd (xλ , xξ )/n the cases of free BCs and PBCs, respectively. For such
d2 − 1
   d−1  
γ 2 potentials
e−γ 1 + + O γ −2


= −4 Re
γ→∞ 2π 8γ
 ~2
× 1 + O e−γ


. (3.63) V(y, z) = V(z) ≡ v(z) (4.5)
2m

P/A
Thus, the quantum corrections to the amplitudes ∆d
are smaller by a factor exp(−23/2 πD/λth ), the decay
length λth /23/2 is twice as large as in the cases of DDBCs,
NNBCs, DNBCs, and RBCs.
the eigenvalues of the operator v(z) − ∇2 are of the form
p2 + εν where p denotes the wave vector conjugate to y
and εν ≡ εBC
ν , ν = 1, 2, . . . , ∞, are the eigenvalues of the
operator −∂z2 + v(z) on the interval [0, D] subject to the
BCs considered.
 12

Let us introduce the grand partition function set of noninteracting bosons with single-particle energies
~2 2
2m (p + εν + µ). Accordingly, we have
Z Z
∗
Ξd,n (T, µ, D, L, ů) = D[ψ , ψ] D[χ] e−Seff [ψ ,ψ,χ]/~
BC ∗
1 XX  β~2 2
ln 1 − e− 2m (p +εν )

Φd ([V]; T, µ, ů, D, L) =
(4.6) A p ν
by analogy with Eq. (2.19). In the large-n limit the as- Z D
sociated grand potential per hyperarea A and number n β
− dz [V(z) + µ]2 . (4.15)
of components 2ů 0
1 1 We now wish to take the limit A = Ld−1 → ∞. To
ϕBC
∞,d (T, µ, ů, D, L) ≡= − lim ln ΞBC
d (T, µ, ů, D, L) allow for macroscopic occupancy of p = 0 states, we
A n→∞ n P
(4.7) separate the p = 0 from p using the asymptotic equiv-
is given by the maximum of the functional alence
Z (d−1)
1 X D
Z 
β 1 X
Φ([V]; T µ, ů, D, L) = dz − [V(z) + µ]2 f (p) ' A−1 f (0) + f (p). (4.16)
A p 0 2ů A p A→∞ p
Z β~
dτ D  h ~2 2 i E
The self-consistency condition of Eqs. (4.9) and (4.14)
+ z, τ  ln ~∂τ + V(z) + (p − ∂z2 ) z, τ ,
 
0 ~ 2m thus becomes
(4.8) X
− β~
2
|hν (z)|2 λ1−d 2m εν


V∗ (z) + µ = ů th Li d−1 e
A→∞ 2
where we have chosen PBCs along the y directions and ν
L
p is a (d − 1)-dimensional wave vector with 2π p ∈ Zd−1 . ů X 1
+ |hν (z)|2 β~2 (4.17)
Functional differentiation with respect to V(z) yields the A ν ε
e 2m ν − 1
stationarity condition
and for the A → ∞ limit of the grand potential we find
1 X
ů gp (z, z, 0−) = V∗ (x) + µ, (4.9) ϕBC BC
A p ∞,d (T, µ, ů, D) ≡ ϕ∞,d (T, µ, ů, D, ∞)
Z D
X  1 2 
− 4π λth εν β 2
= −λ1−d

Li d+1 e − dz µ + V∗ (z)
where gp (z, z 0 , τ − τ 0 ) is a Matsubara Green’s function th 2 2ů 0
ν
satisfying
1 X  1 2

+ lim ln 1 − e− 4π λth εν . (4.18)
~2 2 A→∞ A
 
p − ∂z2 gp (z, z 0 , τ ) = δ(τ ) δ(z − z 0 ). ν


~∂τ + V∗ +
2m The contributions from the p = 0 terms to both equa-
(4.10)
tions [last terms in Eqs. (4.17) and (4.18)] vanish in the
Let εν and hν (z) be the eigenvalues and orthonormalized
limit A → ∞.
eigenfunctions of the Schrödinger equation
It is an easy matter to check that these equations re-
 2 
−∂z + v∗ (z) − εν hν (z) = 0, (4.11) duce to those of [23] if we choose PBCs. To see this, note
that in this case the self-consistent potential is indepen-
and np,ν denote the occupation number dent of z by translational invariance, we have
~2 P

β~2 2
  −1 V∗P ≡ V∗P (T, µ, ů, D) = v (T, µ, ů, D),
np,ν = exp (p + εν ) − 1 . (4.12) 2m ∗
2m εP P P 2
ν = v∗ + (kν ) , (4.19)
Then we have where kνPare the discrete k values given in Eq. (2.6a). We
Xn insert these results into Eqs. (4.17) and (4.18) along with
gp (z, z 0 , τ ) = hν (z)h∗ν (z 0 ) np,ν θ(−τ )

the eigenfunctions hP ν (z) given in Eq. (2.6a). Taking into
ν
o account that the variables s0 and a of [23] correspond to
2
+ (1 + np,ν )θ(τ ) e− 2m (εν +p )τ , (4.13) −βV P and ů, respectively, one recovers the corresponding
 ~

equations of [23].
and the self-consistency Eq. (4.9) becomes
X 1 X B. Bulk properties
V∗ (z) + µ = ů |hν (z)|2 np,ν . (4.14)
ν
A p
The foregoing statements carry over to the bulk quan-
Furthermore, the contribution to the functional R tities
Φd ([V]; T, µ, D, L, ů) associated with the integral dτ
is nothing else than the reduced grand potential of a ϕ∞,b,d (T, µ, ů) ≡ lim D−1 ϕBC
∞,d (T, µ, ů, D) (4.20)
D→∞
 13

and and

Vb,∗ ≡ Vb,∗ (T, µ, ů) = V∗P (T, µ, D = ∞, ů) ϕ∞,b,d [T, µc (1 + δµ), ů] − ϕ∞,b,d (T, µc , ů)
V∗BC (z
2−α∞
= = ∞; T, µ, D = ∞, ů),

Γ(−d/2) ζ(d/2)
' − d (−δµ) + O(δµ)
BC = DD, NN, DN, R. (4.21) δµ→0− λth (T ) −Γ(1 − d/2)
(4.28)
Note that Vb,∗ can be defined either as the D → ∞ limit
of V∗P or in terms of the z → ∞ limit of the poten- for δµ ≡ (µ/µc − 1) → 0− , where γ∞ = 2ν∞ =
tial V∗ |D=∞ for the semi-infinite case with free boundary 2/(d − 2) ≡ γid and α∞ = (4 − d)/(d − 2) = αid
conditions. are the familiar (n = ∞) critical exponents of the
The equations these quantities satisfy, spherical model.
  β
2
The linear scaling field δµ varies linearly in δT . How-
ϕ∞,b,d (T, µ, ů) = −λ−d
th Li d+2 e−βVb,∗ − µ + Vb,∗
2 2ů ever, unlike its Gaussian analog αG , the exponent α∞
1  −βVb,∗

is negative so that δµ ∼ δT as δT → 0. Therefore, the
+ lim ln 1 − e (4.22)
V →∞ V constraint of constant ρ does not lead to a Fisher renor-
and malization of the (n = ∞) critical exponents.
Comparison of the n → ∞ action with that of the ideal
  ů 1 Bose gas shows that the bulk correlation length ξ is given
Vb,∗ + µ ' ů λ−d th Lid/2 e−βVb,∗ + , by the analog of Eq. (1.5) obtained by the replacement
V →∞ V eβVb,∗ − 1
(4.23) −µ → Vb,∗ , i.e.,
are again in accordance with those of [23]. The easiest
way to obtain these bulk equations is to choose PBCs. ~ λth
−1/2
ξ=p = √ βVb,∗ . (4.29)
Alternatively, one can consider the semi-infinite case and 2mVb,∗ 2 π
investigate the limit z → ∞.
For later use, let us also mention that the following re- As an immediate consequence, one obtains for the
sults (obtained in [23]) follow for 2 < d < 4 in a straight- static pair correlation function in the disordered phase
forward fashion from the above bulk equations: 1
∗ 
gb,d (x; λth , ξ) = lim ψα (x, τ )ψα (0, τ −) b,d
n→∞ n
(i) The critical line µc (T, ů) across which the bulk Z (d)
transition occurs is given by eiq·x
= β(V +~ 2 q 2 /2m)
q e b,∗ −1
µc (T, ů) = λ−d
th ů ζ(d/2). (4.24) ∞
X 2 2 2 2
= λ−d
th s−d/2 e−sλth /4πξ −πx /sλth .
(ii) The bulk potential Vb,∗ vanishes on the critical line s=1
µc (T, ů) and in the bulk ordered phase in the ther- (4.30)
modynamic limit V → ∞.
To determine its asymptotic behavior in the regime
(iii) The bulk grand potentials ϕ< >
b,d (T, µ) and ϕb,d (T, µ) λth  x, ξ, one can rescale according to Eq. (2.21) and
in the bulk disordered and bulk ordered phases can take the limit
be written as
λ2th
lim gb,d (x; λth , ξ)
−d λth →0 2π
ϕ< Lid/2+1 e−βVb,∗


∞,b,d (T, µ, ů) = − λth (T ) d−2

β = 2(2π)−d/2 (xξ)− 2 K d−2 (x/ξ),

(Vb,∗ + µ)2
2
− (4.25) cl
2ů ≡ 2gb,d (x; ξ), (4.31)
and which eliminates quantum corrections of order
O e−const x/λth . The right-hand side is twice the propa-


β 2  −d P2n
ϕ<
∞,b,d (T, µ, ů) = − µ − λth (T ) ζ(d/2 + 1), cl
gator gb,d (x; ξ) ≡ limn→∞ α2 =1 hφα2 (x)φα2 (0)ib,d /2n
2ů
(4.26) of the classical O(2n) φ4 theory in the disordered phase,
respectively. as it should. This shows that ξ is the true correlation
length, namely, the scale on which this function decays
(iv) Analysis of the bulk equations (4.23) and (4.22) exponentially in the large-distance limit x → ∞.
yields the limiting behaviors All above-mentioned results for the bulk critical be-
 γ∞ havior of the imperfect Bose gas and our interacting n-
ζ(d/2) component Bose gas are in accordance with the fact that
βVb,∗ [T, µc (1 + δµ), ů] = (−δµ)
µ→µc − −Γ(1 − d/2) this behavior is representative of the universality class
(4.27) of the O(2n) φ4 model in the limit n → ∞. For PBCs,
 14

is over all kνP = 2πν/D with

P
this identification of the universality class for the scaling Here, the summation kP
behavior of both of these Bose gas models near bulk criti- ν ∈ Z, and
cality carries over to the case of finite thickness D. In the
next subsection we explicitly verify that the asymptotic ~ λth
−1/2
ξD = p = √ βV∗P (4.38)
scaling behaviors of the residual grand potential ϕP ∞,d 2mV∗P 2 π
P
and the associated Casimir force FC,∞ near the bulk
transition are indeed described by the scaling functions means the analog of the bulk correlation length (4.29).
of the classical O(2n) model in the n → ∞ limit. The last line of Eq. (4.37) can be obtained via the method
of images or Poisson’s summation formula
∞ Z ∞
C. Scaling functions for periodic boundary 1 X X dk
f (k) = f (k) eijDk (4.39)
conditions D P j=−∞ −∞ 2π
k=k

We begin by considering the case µ < µc where both for functions f (k).
the bulk system and the D < ∞ strip are disordered Expressed in terms of the pair correlation function
P
whenever T > 0. Let us generalize Eqs. (1.6), (1.8), gd,D , the grand potential ϕP
∞,d becomes
(1.10), and (1.13) to the present interacting case. The
interaction constant ů or, equivalently, the rescaled in- ϕP 2 P
∞,d (T, µ, ů, D)/D = −λth gD,d+2 (0; λth , ξD )
teraction constant g defined in Eq. (2.23), gives rise to β
2
a further dimensionless variable. As dimensionless vari- − µ + V∗P . (4.40)
2ů
ables, we choose
The corresponding formula for ϕ∞,b,d (T, µ, ů), which fol-
xλ = D/λth , xξ = D/ξ, xg = gD4−d , (4.32) lows upon taking the limit D → ∞, should be obvious.
The respective form of the self-consistency condition fol-
where ξ now denotes the bulk correlation length (4.29)
lows in a straightforward fashion by equating the deriva-
rather than its ideal Bose gas counterpart ξid . We can
tive ∂/∂V∗P to zero, using its stationarity at V = V∗P .
then write the n = ∞ analogs of Eqs. (1.6) and (1.8) as
This yields
−(d−1) BC
ϕBC
∞,res,d (T, µ, ů, D) = D Υ∞,d (xλ , xξ , xg ). (4.33) ∞
X
and V∗P + µ = ů gb,d (Dj; λth , ξD ) (4.41)
j=−∞
BC
βF∞,C (T, µ, ů, D) = D−d Y∞,d
BC
(xλ , xξ , xg ), (4.34)
and the bulk analog
respectively. In the scaling limit of the Bose-Einstein
transition, where D and ξ both become large compared Vb,∗ + µ = ů gb,d (0; λth , ξ). (4.42)
to all other lengths (namely, λth and g −1/(4−d) ), the be-
havior should simplify to Eliminating µ from these two equations and expressing ů
−(d−1) in terms of the coupling constant g defined in Eq. (2.23)
ϕBC
∞,res,d (T, µ, ů, D) ' D ΘBC
∞,d (D/ξ) (4.35)
D,ξ→∞ yields an equation whose solution gives us D/ξD as a
function of xξ , xλ , and xg :
and
BC
βF∞,C (T, µ, ů, D) ' D−d ϑBC D/ξD = Xd (xλ , xξ , xg ), (4.43)
∞,d (D/ξ), (4.36)
D,ξ→∞
where Xd (xλ , xξ , xg ) satisfies
where Θ∞,d and ϑ∞,d are scaling functions of the classical
O(2∞) φ4 model. The dependence on both λth and g (or

6 2 1
Xd − x2ξ = gb,d 0; x−1


ů) should drop out except from nonuniversal amplitudes 2 λ , 1/Xd
xg 2πxλ
such as that of ξ.
− gb,d 0; x−1 −1


To determine the residual grand potential ϕP ∞,res,d , we λ , xξ
∞
must compute the grand potentials ϕP and ϕP∞,d .

∞,b,d
X
The latter functions can be conveniently written in terms +2 gb,d (j; x−1
λ , 1/Xd ). (4.44)
j=1
of the ideal Bose gas bulk correlation function (4.30) and
its finite-D analog under PBCs, In the limit xλ → ∞, the solution Xd agrees with the
P
gD,d (y, z; λth , ξD ) classical value Xdcl (xξ , xg ) ≡ Xd (0, xξ , xg ) up to exponen-
tially small quantum corrections:
1 X (d−1) ei(p,k)·(y,z)
Z
=
Xd (xλ , xξ , xg ) = Xdcl (xξ , xg ) + O e−1/xλ , (4.45)
P 2 2 2


D P p eβ[V∗ +~ (p +k )/2m] − 1 xλ →∞
k=k
∞
where Xdcl (xξ , xg ) solves the classical analog of Eq. (4.44)
X p 
= gb,d y 2 + (z + jD)2 ; λth , ξD . (4.37)
j=−∞ one obtains upon taking the limit xλ → 0. The limit
 15

value of the right-hand side is easily determined

P∞ with is given by
the aid of Eq. (4.31) or by replacingRthe sums s=1 the  cl
d
ΘP − xdξ + ΘP cl

∞,d (xξ ) = 4πAd+2 Xd,as
∞
functions gb,d involve by integrals 0 ds. Introducing d (Xd,as )
the coefficient + Ad xξd−2 (Xd,as
cl
)2 − x2ξ .
 
(4.53)
Ad = −(4π)−d/2 Γ(1 − d/2) (4.46) One easily checks that this equation is consistent with
and using the result (3.4a) for the ideal Bose gas scal- published results [24, 25, 52] for the classical scaling func-
ing function ΘP tion. For example, setting the scaled magnetic field ȟ = 0
d , we see that the resulting equation for
Xdcl (xξ , xg ) can be written as in Eq. (5.14) of [52], one sees that the function 2Y0 (x2ξ , 0)
of this reference is identical to ΘP∞,d (xξ ), as it should.
6  cl
2
d−2
Xd − x2ξ = −2Ad Xdcl − xd−2 Until now we restricted ourselves to the bulk disor-
  
ξ
xg dered phase δµ ≥ 0. Specializing to the case of d = 3, we
1 P now consider negative and positive values of δµ. Since
Θd−2 Xdcl .


+ (4.47)
2π ν = 1 at d = 3, 1/ξ behaves linearly in |δµ| for δµ → 0−.
The xg depending terms on the left-hand side of Instead of xξ , we choose the variable
Eq. (4.47) yield corrections to scaling to ξD /D of the D
form g −1 D−(4−d) with the familiar n = ∞ exponent ωg = x = −sgn(δµ) for d = 3, δµ R 0. (4.54)
4 − d. As is discussed in some detail in [26] and [27], they ξ(−|δµ|)
can be eliminated by taking the limit g → ∞ [51]. In fact, Because the bulk correlation length is infinite in the or-
it follows from Eq. (4.47) that the function Xdcl (xξ , xg ) dered phase δµ > 0, the coefficient of the x3 term of
behaves asymptotically as ΘP∞,3 for x < 0 differs from its x > 0 analog. On the
Xdcl (xξ , xg ) = cl
Xd,as (xξ ) + O(1/xg ), (4.48) other hand, the result for the scaled inverse finite-size
cl
xg →∞ correlation length X3,as (x) given in Eq. (4.49) remains
cl valid for x < 0. It follows that
where Xd,as is the zero of the right-hand side of this equa-
tion.  cl 2  cl 3
−1/xξ P x3 x X3,as (x) X3,as (x)
At d = 3, where ΘP d−2 simplifies to −2 ln(1 − e ), Θ∞,3 (x) = − θ(x) + −
12π 4π 6π
this zero can be determined in closed analytic form. One
1  −X cl (x) 
finds [52, 53] − Li3 e 3,as
π
cl
X3,as (xξ ) = 2 arsinh(exξ /2 /2) 1 cl cl
(x) Li2 e−X3,as (x) , (x R 0),
 
√ − X3,as
= 2 ln exξ /2 + 4 + exξ /2 ,


(4.49) π
(4.55)
in agreement with [24], [25], and [52].
The residual grand potential ϕP which is again consistent with the results of [24] and [25].
∞,res,d can be computed
along similar lines. Upon eliminating µ in favor of λth The scaling function ΘP ∞,3 (x) is plotted in Fig. 4 along
1
and ξ by means of Eq. (4.42), we can express ϕP ∞,res,d
with 2 times the associated Casimir force scaling function
in terms of the variables xλ , xξ and xg to determine the ϑP
∞,3 (x) one obtains via Eq. (3.11).
function ΥP∞,d (xλ , xξ , xg ). We obtain
x
ΥP
∞,d (xλ , xξ , xg )
h  2 2
  2 2
i ΘP
3,∞ (x)
= xdλ Li1+d/2 e−xξ /4πxλ − Li1+d/2 e−Xd /4πxλ
3 ϑP
3,∞ (x)/2
+ ΥP
d (xλ , Xd ) − (X 2 − x2ξ )2
2xg d
1 d−2 2 2 2

− x (Xd − x2ξ ) Lid/2 e−xξ /4πxλ . (4.50)
4π λ
In the limit xλ → ∞, this becomes
ΥP
∞,d (∞, xξ , xg )

d
= 4πAd+2 Xdcl − xdξ + ΘP cl
 
d (Xd )
3  cl 2 2 FIG. 4. The scaling functions ΘP P
3,∞ (x) and ϑ3,∞ (x).
+ Ad xd−2
 cl 2
(Xd ) − x2ξ − (Xd ) − x2ξ . (4.51)

ξ
2xg
Ignoring the corrections to scaling due to the O(1/xg )
term, we set xg = 0 and find that the classical scaling The results described in this subsection provide ex-
function plicit proof of the fact that the finite-size critical behav-
ior which the interacting n-component Bose gas on a d-
ΘP P
∞,d (xξ ) = Υ∞,d (∞, xξ , 0) (4.52) dimensional strip of finite width D with 2 < d < 4 and
 16

PBCs exhibits in the limit n → ∞ in the vicinity of the This is in fact possible. Let N̂l (z) be the number of
bulk Bose-Einstein transition point is represented by the bosons in layer z:
universality class of the corresponding classical O(2∞)
φ4 model. Because of the equivalence of the interacting
Z
N̂l (z) = dd−1 y ψ̂ † (y, z)ψ̂(y, z). (4.57)
∞-component Bose gas and the imperfect Bose gas, the
same statement applies to the latter model, a fact which
answers the questions about its universality class raised We now modify the potential energy term in Eq. (2.13)
in [23]. and consider the Hamiltonian
It should be clear that the equivalence of these models Z D
also holds for ABCs. Consequently, the critical behavior a
ĤImpBG = T [ψ̂ † , ψ̂] + dz [N̂l (z)]2 . (4.58)
of both models must be described up to quantum correc- 2A 0
tions by the O(2∞) φ4 model with ABCs. The univer-
sality class of the latter classical model corresponds (up We can impose free BCs, such as RBCs or DDBCs, but
to a trivial factor of 2 for free energies) to that of the alternatively also PBCS.
mean spherical model with ABCs studied in [54]. The equivalence of this modified imperfect Bose gas
We refrain from computing the full scaling func- with the interacting ∞-component Bose gas can be seen
tions ΥA A
∞,d (∞, xξ , xg ) and Y∞,d (∞, xξ , xg ) here. How- as follows. The potential-energy term of ĤImpBG yields
ever, information about the associated classical scaling the contribution
functions ΘA A A
∞,d (xξ ) = Υ∞,d (∞, xξ , 0) and ϑ∞,d (xξ ) = Z β~ Z D Z 
A
Y∞,d (∞, xξ , 0) can be inferred from the results of [54]. ∗ a d−1 2
S1 [ψ , ψ] = dτ dz d y |ψ(y, z, τ )| ,
Note that in their work on the mean spherical model 2A 0 0
with ABCs the authors of the latter reference allowed for (4.59)
different values Jk and J⊥ for the ferromagnetic nearest- to the coherent-state action SImpBG = S0 + S1 of this
neighbor bonds parallel and perpendicular to the planes model. Using the Hubbard-Stratonovich transforma-
z = const. In the continuum limit this model maps on a tion (4.1), we can rewrite the exponential exp (S1 /~) as a
φ4 model whose derivative term gyy (∇y φ)2 + gzz (∂z φ)2 functional integral over a real-valued field χ(z, τ ) so that
involves a diagonal, yet anisotropic metric. This intro- the analog of the effective action (4.2) becomes
duces a source of nonuniversality that can be eliminated
~2 2
Z   
by an appropriate rescaling of z (see, e.g., p. 15–17 of eff ∗
SImpBg [ψ , ψ, χ] = ∗
ψ ~∂τ − µ + iχ − ∇ ψ
[55]). To obtain the universal scaling functions ΘA ∞,d and τ,V 2m
ϑA and the Casimir amplitude from [54], one can sim-

∞,d 1
ply set Jk = J⊥ . Specifically, one finds from Eq. (3.48) + χ2 . (4.60)
2a
of [54] the Casimir amplitude
We can now introduce a potential V(z, τ ) by analogy with
2 ζ(3) Eq. (4.3),
∆A
∞,3 = Im[Li2 (eiπ/3 )] − , (4.56)
3 3π
V(z, τ ) = iχ(z, τ ) − µ, (4.61)
whose numerical value 0.549086 . . . is also known from
[56]. We leave it to the reader to extract from [54] the cor- R
perform the functional integral D[ψ, ψ] and exploit the
responding predictions for the scaling functions ΘA ∞,3 (x) translation invariance along the y directions to obtain
A
and ϑ∞,3 (x). Z
∗
ln D[ψ ∗ , ψ] e−Seff [ψ ,ψ,χ]/~ + const
D. Generalizing the imperfect Bose gas model to Z  Z (d−1) D  h
allow for nontranslation-invariant boundary = −A z, τ  ln ~∂τ + V

conditions τ,z p
~2 2

i
2 
E 1 2
As we discussed, the interacting n-component Bose gas
defined by the Hamiltonian (2.14) can be considered for
different BCs along the z direction and its n → ∞ limit
formulated. For PBC and ABCs, its bulk critical be-
havior and finite-size critical behavior on a strip of finite
width D are the same as those of the imperfect Bose
gas. This raises the question as to whether appropriate
nontranslation-invariant generalizations of the imperfect
Bose gas model can be defined that are equivalent to our
interacting n-component Bose gas with boundary condi-
tions along the z direction, such as RBCs or DDBCs.
+ (p − ∂z ) z, τ − (µ + V) /~. (4.62)
2m 2a
Since the right-hand side is proportional
R to hyperarea
A, the remaining functional integral D[χ] can be calcu-
lated in the thermodynamic limit A → ∞ by evaluating
its integrand at the stationary point. The result shows
that our modified imperfect Bose gas model defined by
Eqs. (4.58) and (4.57) is in one-to-one correspondence
with the n → ∞ limit of the interacting Bose gas, if
we identify the interaction strength a with the coupling
constant ů of the latter.
 17

E. Results for d = 3 with Dirichlet-Dirichlet and the diagonal potential matrix v = diag(v1 , . . . , vN ).
boundary conditions Since the bulk limit Nz → ∞ of these equations is inde-
pendent of the BC, we can choose PBCs to study it. The
From our general considerations based on the coherent- spectrum {εν } becomes dense as Nz → ∞. Because of
state functional-integral approach in Sec. II and our anal- the modified (lattice) dispersion relation, the bulk eigen-
ysis in Sec. IV it should be clear that the asymptotic values εdct
b (k) of our discretized model are given by
large-scale finite-size behavior of our n = ∞ Bose film
2
model with free BCs in the scaling regime of the bulk εdct
b (k) = 4 sin (k/2) + vb,∗ , 0 ≤ k ≤ π, (4.67)
Bose-Einstein transition point is described by the clas-
sical O(2∞) φ4 model. Unfortunately, exact analytic rather than by k 2 + vb,∗ . The changes this implies for
solutions for free BCs are neither known for the self- our results for the bulk grand potential, the bulk self-
consistent potential V∗ (z) nor for the eigenvalues εν and consistency equation, and the critical value µc (T, ů) of
eigenfunctions hν even if quantum corrections are ne- the chemical potential are equivalent to the replacement
glected. Of particular interest is the (d = 3)-dimensional of the pair correlation function gb,d (x; λth , ξ) by its ana-
case on which we now focus. For it, a number of exact an- log for our discretized model, namely,
alytic results have been obtained for the classical theory dct
gb,d (x, λth , ξ)
with DDBCs [26, 27, 30–35], which are known to apply
Z π
to this theory with free BCs at d = 3 asymptotically in dk (d−1) eix·(p,k)
Z
the large length scale limit. Using a combination of tech- ≡ 2 2 −2 2 . (4.68)
0 π p eλth [p +ξ +4 sin (k/2)]/4π − 1
niques such as direct solutions of the self-consistent equa-
tions [32], short-distance and boundary-operator expan- Thus the analogs of Eqs. (4.25), (4.26), and (4.24) for our
sions [27], trace formulas [34], inverse scattering methods discrete model can be written as
for the semi-infinite case D = ∞ and matched semiclassi-
cal expansions for D < ∞ [33], exact analytic results for β
ϕdct,< 2 dct
∞,b,d (T, µ, ů) = −λth gb,d+2 (0; λth , ξ) − (µ + Vb,∗ )2 ,
several series expansion coefficients of the self-consistent 2ů
potential v∗ (z) and for the asymptotic x → ∞ behaviors (4.69)
of the eigenvalues εDD DD
ν , eigenfunctions hν , and the clas- β 2
sical scaling functions of the residual free energy and the ϕdct,> 2 dct
∞,b,d (T, µ, ů) = −λth gb,d+2 (0; λth , 0) − µ , (4.70)
Casimir force have been determined. However, the com- 2ů
putation of these scaling functions for all values x R 0 of
the scaling variable x introduced in Eq. (4.54) required dct
Vb,∗ + µ = ů gb,d (0; λth , ξ), (4.71)
the use of numerical methods [26, 27, 29].
In order to solve the self-consistent Schrödinger equa- and
tion numerically, it must be discretized. We do this in the
same manner as in the treatment of the classical model µdct dct
c (T, ů) = ůgb,d (0; λth , 0), (4.72)
called A in [26] and [27], i.e., we discretize only in the
z direction, keeping the y coordinates continuous. Let respectively. Note that these bulk quantities also depend
az be the corresponding lattice spacing. Then, the dis- on the discretization length az , which we have set to 1.
dct
cretized system consists of Nz = L/az layers located at The classical limit λth → 0 of gb,d (x; λth , ξ) is related
to the correlation function
zj = (j − 1/2)az , j = 1, . . . , Nz . (4.63) Z π
dk (d−1) eix(p,k)
Z
cl,dct
For convenience we set az = 1. The discrete analog of gb,d (x; ξ) ≡
0 π p p + 4 sin2 (k/2) + ξ −2
2
the Schrödinger Eq. (4.11) is the eigenvalue equation (4.73)
Nz
of the correspondingly discretized classical O(∞) φ4
X
Hj,j 0 hν,j 0 = εν hν,j (4.64) model. By analogy with Eq. (4.31), we have
j=1
λ2th dct cl,dct
lim gb,d (x; λth , ξ) = 2gb,d (x; ξ). (4.74)
for the matrix λth →0 2π

H = −D2 + v (4.65) The integrals in Eq. (4.73) can be computed using di-
mensional regularization for the p integral. One obtains
with the tridiagonal discrete Laplacian (for DDBCs) [27, 57]
−2 1
 
dct
gb,d (x, λth , ξ) = −Ad−1 ξ 3−d 2 F1 ( 12 , 3−d 2
2 ; 1; −4ξ ).
 1 ... ... (4.75)
 
2

D =   (4.66) This result can be exploited in a straightforward fash-
. .
. . . . 1
  ion to derive the analog of Eq. (4.30). Upon considering
1 −2 the λth → 0 limit of the self-consistency equation and
 18

dct
substituting the large-ξ expansion of gb,d (x, λth , ξ) into and
dct −1/(d−2)
it, one finds ξ ∼ |µ−µc | , where the proportion- dct (1)
ality constant differs from the one implied by Eq. (4.30) lim [gb,d+2 (0, λth , ξ) − Sd (ξ)]
d→3
because of the different dispersion relation of the dis- 1 h p i
crete model. Moreover, upon introducing a variable τ̊ = 2 2
(2 + 4ξ 2 ) arsinh(1/2ξ) − 1 + 4ξ 2 . (4.81)
16π ξ
such that
(ii) The first difference is encountered automatically
βµ = −λ2th τ̊ /4π (4.76) if one subtracts from the classical bulk self-consistency
equation its analog at the bulk transition point. To make
and expressing ů = ~2 λ2th g/24πm in terms of the coupling dct
the bulk free energy f∞,b,d UV finite, we can follow [27]
constant g introduced in Eq. (2.23) and Vb in terms of vb ,
and subtract from it its Taylor expansion to first order
we can take the limit λth → 0 of the bulk grand potential
in τ ,
at fixed τ̊ and g to obtain the associated classical bulk
free energy density. One obtains dct
S(τ, g) = f∞,b,d (0, g) − Ad−1 τ, (4.82)
lim ϕdct
∞,b,d (T, µ, ů) ≡ dct
f∞,b,d (τ, g) defining the renormalized bulk free energy
λ→0
τ̊ ,g
dct,ren dct
cl,dct −1/2
3 2 f∞,b,d (τ, g) ≡ f∞,b,d (τ, g) − Sd (τ, g). (4.83)
= −4πAd+1 gb,d+2 0; vb − (τ + τ̊c − vb ) ,
g
(4.77) Its limit for d → 3 is twice the expression given in
Eq. (4.16) of [27], namely,
where
ren 1 p 2 2 + 1/ξ 2
τ = τ̊ − τ̊c = 4πλ−2 (µc − µ) f∞,b (τ, g) = 4 ξ + 1 − arsinh(1/2ξ)
th β (4.78) 4πξ 2 2π
3
is the deviation of τ̊ from its bulk critical value τ̊c . − (τ − 1/ξ 2 )2 , (4.84)
Evaluated at the stationary point vb,∗ , the result is the g
bulk free energy density of the classical model with the where ξ, the bulk correlation length, satisfies
coherent-state action (2.22) in the limit n → ∞. Hence,
the correspondence of the classical limit of our n = ∞
(
g
τ − 12π arsinh(1/2ξ) for τ > 0,
Bose model with the O(2∞) φ4 model carries over to the ξ −2 = (4.85)
discretized versions of these models. We have explicitly 0 for τ ≤ 0.
verified this here only for the bulk grand potential, but
it should be obvious that the correspondence of ϕDD (iii) By analogy with Eqs. (4.77) and (4.83), we can
∞,d
in the classical limit with the reduced free energy per take the classical limit of the layer grand potential at
DD
layer, f∞,d (τ̊ , ů, D), of the O(2∞) φ4 also holds for the finite D to obtain the reduced layer free energy
discretized versions. However, to understand in detail dct
f∞,d (τ, g, D) ≡ lim ϕdct
∞,d (T, µ, ů, D) (4.86)
that, and how, the results for d = 3 of [26], [27], and [29] λ→0
τ̊ ,g
are related to the scaling behavior of our Bose model, a
few explanatory remarks will be helpful. and introduce the renormalized quantity
(i) One cannot simply set d = 3 in the classical
theory because the dimensionally regularized functions dct,ren
f∞,d dct
(τ, g, D) ≡ f∞,d (τ, g, D) − D Sd (τ, g). (4.87)
dct dct
gb,d (0, λth , ξ) and gb,d+2 (0, λth , ξ) have simple poles at
d = 3 corresponding to UV singularities. The Laurent ex- Its UV-finite limit at d = 3 is twice the result given in
pansions of these functions about d = 3 are known from Eq. (4.15) of [27]. We do not give it here since we are
[21], [26], and [27]. For our purposes it is sufficient to not going to use it in the following.
know that the differences of the first function and its (iv) Once the classical limit of our Bose model has
value at the bulk transition point, and that of the second been taken to eliminate the corrections to scaling due to
function and its Taylor series expansion to first order in quantum effects, the analyses of the corrections to scaling
ξ −2 , performed in [26] and [27] fully apply to the remaining
classical ones of our Bose model. In particular, one can
1
(1)
X eliminate corrections to scaling by taken the limit g → ∞.
ξ −2j ∂rj gb,d+2
dct
(0, λth , r−1/2 ) r=0 , (4.79)
 
Sd (ξ) =
To this end, one defines at d = 3 a linear scaling variable
j=0

t = 24πτ /g (4.88)
have finite d → 3 limits. We have

dct dct arsinh(1/2ξ) in which the amplitude of the correlation length ξ for
lim [gb,d (0, λth , ξ) − gb,d (0, λth , ∞)] = − τ > 0 (µ < µc ) has been absorbed. Upon adding to
d→3 2π dct,ren
(4.80) f∞,3 (τ, g, D) the term D3τ 2 /g, one can perform the
 19

limit g → ∞ to obtain the finite t-dependent layer free whose quoted numerical value is taken from [27]. More-
energy over, the scattering data that are equivalent to the poten-
tial vd=3 (z; t, D = ∞) for the semi-infinite case D = ∞
dct,ren 1 tD
f∞,3 (t, D) = Tr[H (1 + t − ln H)] − . (4.89) are known in closed analytical form; they can be found
4π 2π in Eqs. (4.54), (4.64), and (4.66)–(4.68) of [34].
Furthermore, the self-consistency equation, the bulk free Third, also known from [33] are the leading singular
energy, and the bulk correlation length at d = 3 simplify behaviors of ΘDD ∞,3 (x) and
in this limit g → ∞ to
d DD
t = hz| ln H|zi, (4.90) ϑDD DD
∞,3 (x) = 2 Θ∞,3 (x) + x Θ (x) (4.96)
dx ∞,3
as x → 0±. One has
(
dct,ren 1 sinh(t) − t for t > 0,
f∞,b,3 (t) = (4.91)
2π 0 for t ≤ 0,
h
(s) x i 2
ΘDD DD
∞,3 (x) = ∆∞,3 − ∆A0 + 2x θ(x)
x→0± 24π
and 3
1 X
+ 3 x2 ln |x| + Θj xj + o(x3 ), (4.97)
(
[2 sinh(t/2)]−1 for t > 0, π
ξ= (4.92) j=1
∞ for t ≤ 0.
where
These equations were used in [26] and [27] to determine
Z ∞
the classical scaling functions of the residual free en- (s) 1 coth(u) − 1/u
ergy and the Casimir force quite accurately by numer- ∆A0 = du
8π 0 u2 + π 2 /4
ical means. The corresponding finite-g equations were
= 0.01888264398 . . . . (4.98)
also studied there and their consistency with the g = ∞
results verified. is a universal amplitude difference while the Θj are co-
The upshot of these considerations is that the numeri- efficients of regular contributions. The result yields the
cal results of [26, 27, 29] for the self-consistent potential exact value
give us directly the potential v(z) up to the ignored ex-
ponentially small quantum corrections, whereas those for  DD 00
ϑ∞,3 (0) = −2/π 3 . (4.99)
the bulk, layer, and residual free energies and the Casimir
force must be multiplied by a factor of 2 to give us their Finally, we mention that three terms of the asymptotic
analogs for the Bose gas in the scaling regime near the expansions of the functions ΘDD DD
∞,3 (x) and ϑ∞,3 (x) for x →
bulk critical point up to exponentially small quantum −∞ are known. For the first function, it reads as [33]
corrections. For example, the plot of the critical poten-
−ζ(3) 1 ζ 0 (3)
  
tial shown in Fig. 3 of [27] directly applies to our Bose 2 4|x|
ΘDD
∞,3 (x) = 1 + γ E − − + ln
model, and the scaling functions Θ(x) and ϑ(x) displayed x→−∞ 8π |x| 2 ζ(3) π
in Fig. 4 of this reference correspond to the functions  1  
ΘDD DD
∞,3 (x)/2 and ϑ∞,3 (x)/2. +o . (4.100)
|x|
The exact analytic results for the O(∞) φ4 model ob-
tained or reported in [26, 27, 32–34] and [35] can be
translated to the Bose gas case in a similar fashion. We V. SUMMARY AND CONCLUSIONS
give a few examples. First, the potential v(z), which is
symmetric with respect to reflections about the midplane
z = D/2, i.e., v(z) = v(D − z), behaves asymptotically We investigated fluctuation-induced forces in Bose
as gases confined to strips of thickness D near their Bose-
Einstein bulk condensation point. Both the cases of ideal
1 4t 56ζ(3) 2 and nonideal Bose gases have been considered. For con-
vd=3 (z, t, D) = − + 2 + t + ....
λth z1/|t|4z 2 π z π4 venience, we present here a brief overview of our results,
(4.93) putting them in context with previously published ones
Second, at bulk criticality t = 0, the second (“far”) and referencing our most important equations. We con-
boundary plane produces a leading correction ∼ D−3 so sider separately the parts dealing with ideal and nonideal
that Bose gases.
−1

512 DD z 3
 (i) Ideal Bose gas case. In [7] the residual grand po-
vd=3 (z, 0, D) = 1 − ∆ ∞,3 3 . (4.94) tential and the Casimir force have been determined at
λth zD 4z 2 π D
d = 3 in the form of double series for the cases of PBCs,
Here ∆DD
∞,3 denotes the Casimir amplitude
DDBCs, and NNBCs along the finite direction. There, it
has also been demonstrated that quantum effects con-
∆DD DD
∞,3 = Θ∞,3 (0) = −2 × 0.01077340685024782(1), tribute exponentially small corrections in the scaling
(4.95) regime near the bulk transition point. The double series

that follow from the results of this reference for the scal-
ing functions ΥP DD NN
d (xξ , xλ ) and Υd (xξ , xλ ) = Υd (xξ , xλ )
are listed in Eqs. (3.2) and (3.3). We have confirmed
them by means of alternative derivations.
Furthermore, we have generalized the above-mentioned
results to ABCs, DNBCs, and RBCs. The analog of
the double series (3.2) and (3.3) for ΥA d (xξ , xλ ) is given
in Eq. (3.2). Our result for ΥDN d (xξ , xλ ) is covered by
the one for ΥR d (xξ , xλ ) given in Eqs. (3.55) and (3.56)
as the special case (c1 , c2 ) = (∞, 0). Likewise, the lat-
ter equations apply to ΥDD NN
d (xξ , xλ ) and Υd (xξ , xλ ) if
(c1 , c2 ) = (∞, ∞) and (c1 , c2 ) = (0, 0), respectively. The
representation (3.56) of ΥR d provides an explicit decom-
position
P∞ into a leading classical contribution and a sum
ρ=1 of quantum corrections depending on the Matsub-
ara frequencies ωρ . These quantum corrections decay
exponentially with D on length scales that decrease as ρ
increases. At bulk criticality, the leading quantum cor-
rection decay ∼ e−D/l1 , where the length scale l1 ∝ λth is
half as big for free BC such as DDBs, NNBCs, DNBCs,
and RBCs as for PBC.
If the quantum corrections in the results of [7]
are dropped, the double series for ΥP d (xξ , xλ ) and
ΥDD NN
d (xξ , xλ ) = Υd (xξ , xλ ) must reduce to series for the
classical scaling functions of the free massive O(2) theory.
This was pointed out and verified in [18]. We have explic-
itly shown that the same holds true for ABCs, DNBCs,
and RBCs.
Summing the series of the scaling functions ΘBC with
3
BC = P, A, DD, NN, and DN of the classical free O(2)
theory, we have derived the closed exact analytic expres-
sions for ΘBC BC
3 (xξ ) and ϑ3 (xξ ) given in Eqs. (3.10a)–
(3.10d) and (3.12a)–(3.12d), respectively.
(i) Nonideal Bose gas case. We have considered two
distinct, but related, models for nonideal Bose gases on
a strip subject to different BC along the finite direc-
tion: the so-called imperfect Bose gas [22, 23] and an n-
component generalization of a standard Bose model with
short-range interactions. The first one, defined for PBC,
was investigated in [23]. There, the critical Casimir force
was computed right at the Bose-Einstein bulk transition
point in d = 3 dimensions, but the full scaling functions
near this transition not determined. The amplitude of
this force turned out to have twice the value it has for the
mean-spherical model [24, 25], which prompted the au-
thors to raise the question as to which universality class
applies to the fluctuation-induced forces of the imperfect
Bose gas.
We have shown that the imperfect Bose gas with PBC
corresponds to the n → ∞ limit of our n-component Bose
model with short-range interactions. It follows from the
general arguments discussed in Sec. II that the bulk crit-
ical behavior and the finite-size critical behavior of the
latter model on the strip are represented by the corre-
sponding classical O(2n) φ4 model. As a consequence,
the critical Casimir forces near the Bose-Einstein bulk
transition point of the imperfect Bose gas with PBCs
must be representative of the universality class of the
20
O(2n) φ4 model in the limit n → ∞. Since the O(n) φ4
model with PBCs in the limit n → ∞ belongs to the same
universality class as the mean spherical model, the bulk
critical and finite-size critical behaviors of the imperfect
Bose gas near the bulk transition point are represented
by the latter model up to a trivial factor of 2 in the
free energies and the Casimir force. We have explicitly
verified this by computing the scaling functions ΘP ∞,3 (x)
and ϑP ∞,3 (x) from the ∞-component Bose model, prov-
ing that they comply with the results of [24, 25] for the
mean spherical model (up to the mentioned trivial fac-
tor of 2). Our exact analytic result for ΘP ∞,3 (x) is given
in Eq. (4.55), and plots of the functions ΘP ∞,3 (x) and
P
ϑ∞,3 (x) are displayed in Fig. 4.
The equivalence of the imperfect Bose model and the
∞-component Bose model suggests a generalization of
the former for other BC, namely, free BC. We have in-
troduced such a generalized imperfect Bose model with
free BC along the z direction in Sec. IV D. Its bulk crit-
ical and finite-size critical behaviors near the bulk crit-
ical point are represented by the corresponding O(2∞)
φ4 with free BC. At d = 3, where RBCs with subcriti-
cal enhancement variables c1 > 0 and c2 > 0 turn into
DDBCs in the large length scale limit, one can there-
fore exploit the known exact results for the O(∞) φ4
model [26, 27, 29, 32–35] to obtain exact information
about the corresponding scaling functions ΘDD ∞,3 (x) and
ϑDD
∞,3 (x). A number of exact analytic properties, such
as the near-boundary behavior of the self-consistent po-
tential and the limiting behaviors of the scaling func-
tions for x → −∞, are presented in Eqs. (4.93), (4.94),
and (4.97)–(4.100). In order to benefit also from the nu-
merical results of [26] and [27], we have generalized the
discretization scheme of the discretized O(∞) φ4 model
called A in these references to the n-component Bose gas
and verified that the numerically computed scaling func-
tions correspond to one-half of those of the ∞-component
Bose gas, namely, ΘDD DD
∞,3 (x)/2 and ϑ∞,3 (x)/2.
The primary focus of our investigations here has been
put on fluctuation-induced forces in the scaling regime
of the bulk critical point. Accordingly, we have assumed
throughout this paper that the strip thickness D is much
larger than the thermal de Broglie wavelength λth . How-
ever, as T decreases at fixed given D, the thermal length
λth ultimately becomes much larger than the thickness
D. This suggests complementary studies of the asymp-
totic regime λth  D both for ideal and interacting Bose
gases. Rather than embarking on such a study, we re-
strict ourselves here to a few remarks.
An investigation of fluctuation-induced forces in the
asymptotic regime λth  D was made for the imper-
fect Bose gas on a strip with PBCs and 2 < d < 4 in
a recent paper [14]. In this case one can restrict one-
self to the k P = 0 mode of the operator −∂z2 because the
modes with eigenvalues (kνP )2 = (2πν/D)2 give sublead-
ing (exponentially decaying) corrections. The essence of
this approximation (used in [14]) is easily understood
 21

in the language of the coherent-state representation of where z is the fugacity.

the model. Written in terms of the (k = 0) field com- PWe now substitute the series representation Liν (x) =
RD ∞ −ν s
ponent Ψ(y, τ ) ≡ D−1/2 0 dz ψ(y, z, τ ), the effective s=1 s x for the polylogarithm in Eq. (A1) and
action Seff [Ψ∗ , Ψ] one obtains upon discarding the re- use Poisson’s summation formula (4.39) with D = 1 for
maining k P > 0 contributions to ψ(y, z, τ ) describes a f (k) = exp(−sk 2 /4πx2λ ). This gives the result stated in
(d − 1)-dimensional interacting Bose field theory with a Eq. (3.1).
coupling constant ů/D, where ů is the interaction con- To compute the scaling function ΥA d , one can use the
stant of the d-dimensional theory [37, 58]. Reference [14] generalized Poisson identity for theta functions
finds three regions of distinct asymptotic behaviors dis- ∞ ∞
tinguished by whether (|µ|/ů)/λ2th D  1 or  1 and the 2 2 2
X X
e−t(ν+a) = (π/t)1/2 e−π j /t
cos(2πaj)
sign of µ. Whether and to what extent these findings ν=−∞ j=−∞
might carry over to interacting Bose gases on a three- (A3)
dimensional strip with PBCs is not clear to us for two with t = πs/xλ2 and a = 1/2. A straightforward calcu-
reasons. The first is that the effective (d−1)-dimensional lation then yields Eq. (3.2).
interacting field theory that results upon making the re- For DDBCs, NNBCs, and DNBCs a more general form
placement ψ(y, z, τ ) → Ψ(y, τ ) appears to require at low of Poisson’s formula involving the cosine transform
temperatures a more sophisticated treatment than the
Z ∞
Hartree-type approximation to which the use of the im-
perfect Bose gas model corresponds. Renormalization fcos (x) ≡ dν f (ν) cos(xν). (A4)
0
group analyses of the low-temperature behavior of in-
teracting Bose gases such as [59] and [60] indicate this. can be used, namely,
(For further literature on such approaches, see the refer-
∞ ∞
ences of these papers and those of [37].) Furthermore, for X 1 X
d = 3, the corresponding effective two-dimensional inter- f (ν) = − f (0) + fcos (0) + 2 fcos (2πj), (A5)
ν=1
2 j=1
acting Bose gas has a low-temperature phase with quasi
long-range order, which the imperfect Bose gas misses.
which holds for non-negative, continuous, decreasing, and
Riemann integrable functions f on [0, ∞) [61].
It yields the Jacobi identity
ACKNOWLEDGMENTS
∞ ∞
X 2
/4πx2λ xλ 1 2xλ X −4πx2λ j 2 /s
We are grateful to M. Napiórkowski for informing us e−s(πν) = − + e (A6)
s1/2 2 s1/2 j=1
about his work on the imperfect Bose gas and informative ν=1
correspondence, and also to W. Zwerger for his helpful
comments. that was used in the calculation of [7] for DDBCs and
NNBCs at d = 3. The scaling functions ΥDD d (xλ , xξ ) and
ΥNNd (xλ , xξ ) can be computed for 2 < d 6= 3 along similar
Appendix A: Scaling functions of the ideal Bose gas lines by expanding the analog of the exponential in the
first line of Eq. (A1), integrating termwise, and using the
Jacobi identity (A6). From the first term on the right-
To compute ΥP P
d we use the fact that ϕs,d = 0, intro- hand side of Eq. (A6) the bulk term ϕb,d is recovered.
duce the variables xλ ≡ D/λth and xξ ≡ D/ξ, and set The second one yields surface contributions 2ϕD s,d (T, µ) =
D = 1 so that ΥBC BC
d (xλ , xξ ) = ϕres,d (T, µ, 1). After an −2ϕN s,d (T, µ) that are in accordance with Eqs. (3.29) and
integration by parts, one arrives at
(3.31), respectively. The last term on the right-hand side
Kd−1 of Eq. (A6) yields the expressions for the scaling functions
ΥP
d (xλ , xξ ) = − ΥDD NN
(d − 1)2πx2λ d (xλ , xξ ) and Υd (xλ , xξ ) given in Eq. (3.3).

Z ∞

Z ∞ We refrain from rewriting the double series expansion
X dk dp pd for ΥDNd one obtains from the analog of Eq. (A1) with the
×  −  h k2 +p2 +x2 i aid of the Poisson summation formula (A5) because we
k/2π∈Z −∞ 2π 0 exp ξ
−1
4πx2λ can get both ΥDN d as well as ΘDN
d upon setting (c1 , c2 ) =
(∞, 0) in the results for RBCs given in Eqs. (3.56) and
 
Z ∞  k2 +x2ξ 
d−1 
X dk  − 2
(3.58). Integrating by parts the corresponding expression
= −xλ − Li d+1 e 4πxλ . (A1)
−∞ 2π implied by Eq. (3.58) yields Eqs. (3.8a) and (3.8b).
2
k/2π∈Z

The subtracted term involving the k integral is the bulk

term Appendix B: Consistency of Eqs. (3.37) and (3.45)
1 2 2
ϕb,d (T, µ) = −λ−d
th Li d+2 (z), z = eβµ = e− 4π xξ /xλ , In this Appendix, we show the consistency of
2
(A2) Eqs. (3.37), (3.46), and (3.45) by rederiving the first of
 22

these equations fromR the last one. Equation (3.37) in- where
volves an integral dk . . . over an even function of k, 
c21 −κ2 (c1 −κ)(c2 −κ)

which converges for d < 2 and is defined by analytic 2κ 2κ e−κD
MD (κ, c1 , c2 ) =  
continuation for 2 ≤ d < 4. Upon changing to the inte- (c1 −κ)(c2 −κ) c22 −κ2
gration variable E = k 2 , we can rewrite it as a contour 2κ e−κD 2κ
integral along the contour C1 depicted in Fig. 3 so that (C2)
Eq. (3.37) becomes Going over to scaled variables then gives the result re-
ported in Eq. (3.58).
R (d−1)/2 √ An analogous procedure can be used to compute the
fs,d (ξ, c) E + ξ −2

dE 2c E
Z
= scaling function ΥR d (xλ , xξ , c1 , c2 ). Let S0∞ [ψ ∗ , ψ] be
Γ(1/2 − d/2) C1 8πE 4π E + c2 the coherent-state action of a free massive n-component
1 quantum theory defined by Eq. (2.20b) with V = Rd and
+ (4πξ 2 )(1−d)/2 . (B1)
4 consider the restricted partition function
√ Z
∞ ∗
Let us add and subtract i(E −c2 ) to the numerator 2c E ΞR,∞
0 (T, µ, c ,
1 2c ) ≡ DR [ψ ∗ , ψ] e−S0 [ψ ,ψ]/~ (C3)
of the fraction. The subtracted term yields a contribution
proportional to the residue at E = 0 of the associated with
integrand. It cancels the last term in Eq. (B1). The
remaining integral can be rewritten to obtain DR [ψ ∗ , ψ]
2
Y Y
= D[ψ ∗ , ψ] δ[(∂z + (−1)j cj )ψα (y, Dδj,2 , τ )]
R −∞ (d−1)/2 
fs,d (ξ, c) −E − ξ −2

dE
Z
= y,τ,α j=1
Γ(1/2 − d/2) −1/ξ 2 4πE 4π
×δ[(∂z + (−1)j cj )ψα∗ (y, Dδj,2 , τ )] .

(C4)
E 1/2 − ic

× 1/2 sin[π(d − 1)/2]
E + ic Here the δ functions ensure that the fields ψ(y, z, τ ) and
Kd−1 ψ ∗ (y, z, τ ) satisfy the RBCs (1.15). Representing these
= cd−1 Jd (ξc). (B2) δ functions by means of two pairs of n-component auxil-
Γ[(1 − d)/2]
iary fields χj (y, τ ) and χ∗j (y, τ ) located on the respective
The result given in the last line is equivalent to planes z = Dδj,2 , we arrive at
Eqs. (3.37) and (3.46). To
p get it we transformed to the
Z
∞ ∗

integration variable p = −E − 1/ξ 2 . ΞR,∞

0 (T, µ, c ,
1 2c ) = e−S0 [ψ ,ψ]/~
χ∗ ,χ,ψ ∗ ψ
2
Y  Z β~ Z
dd−1 y χ∗j · ∂z + (−1)j cj ψ j


× exp i dτ
Appendix C: Scaling functions of the residual free j=1 0
energy and residual grand potential for Robin 
j ∗


boundary conditions + χj · ∂z + (−1) cj ψ j , (C5)
R
Here, we complete our calculations of the scaling func- where χ∗ ,χ,ψ∗ ψ is a short-hand for the functional inte-
tions ΘR R
d (xξ , c1 , c2 ) and Υd (xλ , xξ , c1 , c2 ), establishing grals D[χ∗ , χ] D[ψ ∗ , ψ] and |j means that z has been
R R
the results given in Eqs. (3.58) and (3.56). We first de- set to Dδj,2 .
scribe an alternative, somewhat easier way of computing The action is quadratic in the fields ψ ∗ . .R. , χ. We
these functions. can first perform the functional integration ψ∗ ,ψ and
The function ΘR d was computed for xξ = 0 in [11, 20], a
R
subsequently χ∗ ,χ . The integrand of the latter integral
value for which it reduces to the scale-dependent Casimir is a Gaussian involving the matrix kernel
amplitude of Eq. (3.59). The calculation used in these
references can be extended in a straightforward fashion ←− ←− !
(∂n − c1 )1 |g (b) |1 (∂n − c1 ) (∂n − c1 )1 |g (b) |2 (∂n0 − c2 )
to the noncritical case ξ∞ > 0 to derive the result given ←− ←−
in Eq. (3.58). A convenient alternative way is to consider (∂n − c2 )2 |g (b) |1 (∂n0 − c1 ) (∂n − c2 )2 |g (b) |2 (∂n0 − c2 )
an O(2n) massive free field theory in the infinite space 2m
= MD [κp,ρ (ξ, λth ), c1 , c2 ], (C6)
Rd , impose the boundary conditions (1.15) via δ func- ~
tions, represent these δ functions as integrals over aux-
where ∂n means the inner normal, i.e., ∂n |1 = ∂z |z=0
iliary fields χ1 (y) and χ2 (y) with support on the planes
and ∂n |n = −∂z |z=D . The function g (b) denotes the (free
z = 0 and D, respectively, and integrate out φ (see, e,g.,
bulk) propagator
[20, 62, 63]). This gives a Gaussian free energy from
which we subtract its value for D = ∞, obtaining (b)
hψα (y, z, τ )ψβ∗ (0, 0, 0)i0
Z (d−1) ∞
Z
dd−1 p

MD [κp (ξ), c1 , c2 ])
 1 X (b)
n ln det , (C1) = δαβ eip·y g (p, ρ, z) e−iωρ τ ,(C7)
(2π)d−1 M∞ [κp (ξ), c1 , c2 ] p β~ ρ=−∞
 23

associated with the action S0∞ and is given by What remains to show is the equivalence of Eq. (3.55)
with Eq. (3.56). We use the fact that Tres (z; 1, c1 , c2 ) can
2m 1 be written as
g (b) (p, ρ, z) = e−κp,ρ (ξ,λth )|z| . (C8)
~ 2κp,ρ (ξ, λth )
√
Rc1 ,c2 ( ζ)
Tres (ζ; 1, c1 , c2 ) = ∂ζ ln (C10)
R
Performing the Gaussian integral χ∗ ,χ yields a (0) √
Rc1 ,c2 ( ζ)
determinant for the ratio Rof the partition func-
∞
tions ΞR,∞
0 (T, µ, c1 , c2 ) = DR [ψ ∗ , ψ] e−S0 /~ and
∞ with
Ξ∞ D[ψ ∗ , ψ] e−S0 /~ . The value of its loga-
R
0 (T, µ) =
rithm at D = ∞ is easily subtracted. One thus arrives √ √
at
p ( ζ + ic1 )( ζ + ic2 ) −i√ζ
Rc(0)
1 ,c2
( ζ) = √ e , (C11)
2i ζ
ϕR
res,d (T, µ, c1 , c2 )
X Z ∞ so that
= D1−d Kd dp pd−2 gc1 ,c2 [κp,ρ (1/xξ , 1/xλ )].
ρ 0 √ √ √
Rc1 ,c2 ( ζ) ( ζ − ic1 )( ζ − ic2 ) i2√ζ
(C9) (0) √ =1− √ √ e . (C12)
Rc1 ,c2 ( ζ) ( ζ + ic1 )( ζ + ic2 )
The ρ = 0 contribution yields the limiting classical be-
havior. The contributions for ρ = ±|ρ| =
6 0 yield the sum Using these results to express the function Tres in
of quantum corrections resulting from the second term in Eq. (3.55), one can
√ integrate by parts and transform to
the curly brackets of Eq. (3.56). the variable p = −u. One thus arrives at Eq. (3.56).

[1] H. B. G. Casimir, Proc. K. Ned. Akad. Wet. 51, 793
(1948).
[2] M. Krech, Casimir Effect in Critical Systems (World Sci-
entific, Singapore, 1994).
[3] J. G. Brankov, D. M. Dantchev, and N. S. Tonchev,
Theory of Critical Phenomena in Finite-Size Systems —
Scaling and Quantum Effects (World Scientific, Singa-
pore, 2000).
[4] A. Gambassi, J. Phys.: Conference Series 161, 012037
(2009).
[5] J. A. Hertz, Phys. Rev. B 14, 1165 (1976).
[6] S. Sachdev, Quantum Phase Transitions, 2nd ed. (Cam-
bridge University Press, Cambridge (UK), 2011).
[7] P. A. Martin and V. A. Zagrebnov, Europhys. Lett. 73,
15 (2006).
[8] For a more recent investigation of the (d = 3)-
dimensional ideal Bose gas case, both above and below
the bulk critical temperature Tc , see [9].
[9] S. Biswas, Journal of Physics A: Mathematical and The-
oretical 40, 9969 (2007).
[10] H. W. Diehl, in Phase Transitions and Critical Phenom-
ena, Vol. 10, edited by C. Domb and J. L. Lebowitz (Aca-
demic, London, 1986) pp. 75–267.
[11] F. M. Schmidt and H. W. Diehl, Phys. Rev. Lett. 101,
100601 (2008).
[12] H. W. Diehl and F. M. Schmidt, New Journal of Physics
13, 123025 (2011).
[13] Note that ϕb,d and ϕBC s,d are defined through appropriate
limits D → ∞ at fixed λth . If λth  D, a limit recently
considered in Ref. [14], then the decomposition looses its
significance.
[14] P. Jakubczyk, M. Napiórkowski, and T. Sȩk, EPL (Eu-
rophysics Letters) 113, 30006 (2016).
[15] J. D. Gunton and M. J. Buckingham, Phys. Rev. 166,
152 (1968).
[16] In the case of interacting Bose gases, there are additional
length scales associated with the interaction; see, e.g.,
Ref. [17]. Likewise, additional lengths related to surface
properties, such as 1/c1 and 1/c2 , come into play if one
considers films subject to Robin boundary conditions.
[17] P. B. Weichman, M. Rasolt, M. E. Fisher, and M. J.
Stephen, Phys. Rev. B 33, 4632 (1986).
[18] A. Gambassi and S. Dietrich, Europhys. Lett. 74, 754
(2006).
[19] M. Krech and S. Dietrich, Phys. Rev. A 46, 1886 (1992).
[20] T. Emig, N. Graham, R. L. Jaffe, and M. Kardar, Phys.
Rev. D 77, 025005 (2008).
[21] F. M. Schmidt, Der thermodynamische Casimir-Effekt
mit symmetrieerhaltenden und symmetriebrechenden
Randbedingungen, Doktorarbeit, Universität Duisburg-
Essen, Duisburg (2014).
[22] B. Davies, J. Math. Phys. 13, 1324 (1972).
[23] M. Napiórkowski, P. Jakubczyk, and K. Nowak, Journal
of Statistical Mechanics: Theory and Experiment 2013,
P06015 (2013).
[24] D. Danchev, Phys. Rev. E 53, 2104 (1996).
[25] D. M. Danchev, Phys. Rev. E 58, 1455 (1998).
[26] H. W. Diehl, D. Grüneberg, M. Hasenbusch, A. Hucht,
S. B. Rutkevich, and F. M. Schmidt, EPL (Europhysics
Letters) 100, 10004 (2012), arXiv:1205.6613.
[27] H. W. Diehl, D. Grüneberg, M. Hasenbusch, A. Hucht,
S. B. Rutkevich, and F. M. Schmidt, Phys. Rev. E 89,
062123 (2014), arXiv:1405.5787.
[28] D. Dantchev, J. Bergknoff, and J. Rudnick, Phys. Rev.
E 89, 042116 (2014).
[29] H. W. Diehl, D. Grüneberg, M. Hasenbusch, A. Hucht,
S. B. Rutkevich, and F. M. Schmidt, Phys. Rev. E 91,
026101 (2015), arXiv:1405.5787.
[30] A. J. Bray and M. A. Moore, Phys. Rev. Lett. 38, 735
(1977).
[31] A. J. Bray and M. A. Moore, J. Phys. A 10, 1927 (1977).
[32] H. W. Diehl and S. B. Rutkevich, Journal of Physics

A: Mathematical and Theoretical 47, 145004 (2014),
arXiv:1401.1357.
[33] S. B. Rutkevich and H. W. Diehl, Phys. Rev. E 91,
062114 (2015).
[34] S. B. Rutkevich and H. W. Diehl, Journal of Physics A:
Mathematical and Theoretical 48, 375201 (2015).
[35] H. W. Diehl and S. B. Rutkevich, TMF 190, 325 (2017),
Engl. transl.: Theoretical and Mathematical Physics
190(2), 279–294 (2017); arXiv:1512.05892.
[36] This function differs from the function Rc1 ,c1 (k) used in
Refs. [11], [12], and [21]. Our different choice here serves
to avoid a zero at k = 0 whenever c1 and c2 do not both
vanish.
[37] I. Bloch, J. Dalibard, and W. Zwerger, Rev. Mod. Phys.
80, 885 (2008).
[38] P. Jakubczyk and M. Napiórkowski, Phys. Rev. B 87,
165439 (2013).
[39] M. Napiórkowski and J. Piasecki, Phys. Rev. E 84,
061105 (2011).
[40] M. Kac, G. E. Uhlenbeck, and P. C. Hem-
mer, Journal of Mathematical Physics 4, 216 (1963),
http://dx.doi.org/10.1063/1.1703946.
[41] U. C. Täuber, Critical Dynamics. A Field Theory Ap-
proach to Equilibrium and Non-Equilibrium Scaling be-
havior (Cambridge University Press, University Printing
House, Cambridge CB2 8BS, United Kingdom, 2014).
[42] It has been known for long that appropriate classical
models describing the critical behavior of quantum mod-
els near critical temperatures Tc > 0 result upon replace-
ment of ψ by its zero-matsubara frequency component;
see, e.g., [5]. The fact that the classical 2n-component
φ4 model results in this manner was used, for example,
in the bulk case to compute the interaction-induced shift
of the bulk critical point by means of a 1/n expansion
applied to the φ4 theory; see, e.g., [43], [44], and their
references.
[43] G. Baym, J.-P. Blaizot, and J. Zinn-Justin, Europhys.
Lett. 49, 150 (2000).
[44] B. Kastening, Phys. Rev. A 69, 043613 (2004).
[45] DLMF, “NIST Digital Library of Mathematical Func-
tions,” http://dlmf.nist.gov/, Release 1.0.8 of 2014-04-
25, online companion to [46].
[46] F. W. J. Olver, D. W. Lozier, R. F. Boisvert, and C. W.
Clark, eds., NIST Handbook of Mathematical Functions
(Cambridge University Press, New York, NY, 2010) print
24
companion to [45].
[47] K. Symanzik, Nuclear Physics B 190, 1 (1981).
[48] M. E. Fisher, Phys. Rev. 176, 257 (1968).
[49] The two expansion terms given for d = 3 can be obtained
by expanding the two expansion terms given for d 6=3 to
order x ln x. The pole ∝ x2 /(d−3) that the Laurent series
of the x(d−1)/2 term yields is of order x2 .
[50] Wolfram Research, Computer code Mathematica, ver-
sion 11.
[51] We use dimensional regularization for the classcial the-
ory. As is expounded in Ref. [27], the value ∞ may be
viewed as a fixed-point value of g. Setting g = ∞ elim-
inates the Wegner-type corrections to scaling associated
with deviations of g from its fixed-point value.
[52] D. Dantchev, H. W. Diehl, and D. Grüneberg, Phys.
Rev. E 73, 016131 (2006).
[53] Note that the second line of Eq. (4.72)
√ of Ref. [52] con-
tains a misprint: the right-most e ř∞ /2 should be re-
√
placed by e ř∞ .
[54] D. Dantchev and D. Grüneberg, Phys. Rev. E 79, 041103
(2009).
[55] H. W. Diehl and H. Chamati, Phys. Rev. B 79, 104301
(2009).
[56] H. Chamati, Journal of Physics A: Mathematical and
Theoretical 41, 375002 (2008).
[57] The result given here follows from the result given in
Eq. (A3) of [27] for the integral Id (r = 1/ξ 2 ) by multi-
plying it with −Ad−1 and the use of a standard transfor-
mation formula for the hypergeometric function.
[58] As can be seen from Eq. (2.11), the coupling constant
ů of the (d = 3)-dimensional Bose theory is linear in the
s-wave scattering length as . its (d = 2)-dimensional coun-
terpart vanishes (see Ref. [37] and its references). In the
case of a strictly two-dimensional interacting Bose gas,
one would have to choose an energy-dependent interac-
tion constant [37].
[59] F. Pistolesi, C. Castellani, C. Di Castro, and G. C. Stri-
nati, Phys. Rev. B 69, 024513 (2004).
[60] S. Floerchinger and C. Wetterich, Phys. Rev. A 79,
013601 (2009).
[61] T. M. Apostol, Mathematical Analysis (Addison-Wesley
Publishing Company, Reading, MA, 1957).
[62] H. Li and M. Kardar, Phys. Rev. A 46, 6490 (1992).
[63] M. Burgsmüller, H. W. Diehl, and M. A. Shpot,
J. Stat. Mech.: Theor. Exp. 2010, P11020 (2010),
[arXiv:1008.4241].