Local zeta regularization

and the scalar Casimir effect IV.

The case of a rectangular box

Davide Fermi a , Livio Pizzocchero b (1 )

arXiv:1505.03276v4 [math-ph] 6 Aug 2015

a
Dipartimento di Matematica, Università di Milano
Via C. Saldini 50, I-20133 Milano, Italy
e–mail: [email protected]
b
Dipartimento di Matematica, Università di Milano
Via C. Saldini 50, I-20133 Milano, Italy
and Istituto Nazionale di Fisica Nucleare, Sezione di Milano, Italy
e–mail: [email protected]

Abstract
Applying the general framework for local zeta regularization proposed in
Part I of this series of papers, we compute the renormalized vacuum expecta-
tion value of several observables (in particular, of the stress-energy tensor and
of the total energy) for a massless scalar field confined within a rectangular
box of arbitrary dimension.

Keywords: Local Casimir effect, renormalization, zeta regularization.

AMS Subject classifications: 81T55, 83C47.
PACS: 03.70.+k, 11.10.Gh, 41.20.Cv .

1
Corresponding author
Contents

1 Introduction. 3

2 A summary of results from Part I 5

2.1 General setting. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Zeta regularization. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.3 Conformal and non-conformal parts of the stress-energy VEV. . . . . 7
2.4 Integral kernels. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.5 The Dirichlet kernel and its relations with the stress-energy VEV. . . 7
2.6 The heat kernel. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.7 The Dirichlet kernel as Mellin transform of the heat kernel. . . . . . . 9
2.8 The case of product domains. Factorization of the heat kernel. . . . . 9
2.9 Pressure on the boundary. . . . . . . . . . . . . . . . . . . . . . . . . 10
2.10 The total energy. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.11 Forces on the boundary. . . . . . . . . . . . . . . . . . . . . . . . . . 11

3 The case of a massless field inside a box 11

3.1 Introducing the problem. . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.2 The heat kernel. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.2.1 First representation for the heat kernel (useful for large t). . . 12
3.2.2 Second representation for the heat kernel (useful for small t). . 13
3.2.3 Considerations on the sign of some coefficients. . . . . . . . . 14
3.3 The Dirichlet kernel. . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.3.1 Series expansion and analytic continuation of Ds(>) (x, y). . . 16
3.3.2 Series expansion and analytic continuation of Ds(<) (x, y). . . 16
3.3.3 Conclusions for the Dirichlet kernel. . . . . . . . . . . . . . . . 19
3.4 Convergence and remainder estimates for the series in Eq.s (3.30)
(3.31) and (3.36) (3.37). . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.4.1 Estimates for the series (3.30) (3.31). . . . . . . . . . . . . . . 20
3.4.2 Estimates for the series (3.36) (3.37). . . . . . . . . . . . . . . 21
3.5 The stress-energy tensor. . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.6 The pressure on the boundary. . . . . . . . . . . . . . . . . . . . . . . 22
3.6.1 Equivalence of prescriptions (3.62) (3.63). . . . . . . . . . . . 23
3.6.2 Non-integrable behaviour of the renormalized pressure near
the edges. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.7 The total energy. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.7.1 Series expansions and analytic continuations for E u,(>) , E u,(<) . 27
3.7.2 Estimates for the series (3.79) (3.80). . . . . . . . . . . . . . . 28
3.8 The total force on a side of the box. . . . . . . . . . . . . . . . . . . . 29

1
 u,(>) u,(<)
3.8.1 Series expansions and analytic continuations for F1,0 , F1,0 . 30
3.8.2 Estimates for the series (3.92) (3.93). . . . . . . . . . . . . . . 31
3.9 Scaling considerations. . . . . . . . . . . . . . . . . . . . . . . . . . . 33

4 The previous results in spatial dimension d ∈ {1, 2} 34

4.1 Case d = 1 (the segment). . . . . . . . . . . . . . . . . . . . . . . . . 34
4.2 Case d = 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

A Appendix. Absolute convergence and remainder bounds for the

series of subsections 3.3. 47
A.1 Preliminary estimates. . . . . . . . . . . . . . . . . . . . . . . . . . . 47
A.2 The series (A.1); connections with the expansions (3.30) (3.31) for
Ds(>) and its derivatives. . . . . . . . . . . . . . . . . . . . . . . . . . 49
A.3 The series (A.2): connections with the expansions (3.36) (3.37) for
Ds(<) and its derivatives. . . . . . . . . . . . . . . . . . . . . . . . . . 51
(d)
A.4 Asymptotics for HN . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

B Appendix. The expansions of subsection 3.7 for the bulk energy 53

B.1 Derivation of the expansion (3.79). . . . . . . . . . . . . . . . . . . . 53
B.2 Derivation of the expansion (3.80). . . . . . . . . . . . . . . . . . . . 54
B.3 Concluding the previous argument: derivation of Eq. (B.7). . . . . . . 54

C Appendix. Absolute convergence and remainder bounds for the

series of subsection 3.7. 56
u,(>)
C.1 The series (C.1); connections with the expansion (3.79) of E . . . 56
u,(<)
C.2 The series (C.2); connections with the expansion (3.80) of E . . . 57

2
1 Introduction.
In Part I of this series [12, 13, 14] we have considered in general local (and global)
zeta regularization for a quantized neutral scalar field in a d-dimensional spatial
domain Ω, in the environment of (d + 1)-dimensional Minkowski spacetime.
In the present Part IV we consider a massless field confined within a d-dimensional
rectangular domain Ω = (0, a1 ) × ... × (0, ad ), with Dirichlet boundary conditions;
applying the general scheme of Part I we renormalize several observables, namely:
the vacuum expectation value (VEV) of the stress-energy tensor, the pressure on
boundary points, the total energy VEV and of the total force acting on any side of the
box. All these renormalized observables are represented as sums of series converging
with exponential speed, for which we give quantitative remainder estimates. Our
results hold for an arbitrary spatial dimension d; we subsequently specialize them
to the subcase (d = 1 and) d = 2.
Let us make a comparison with the existing literature on the subject. The total
energy and the forces on the sides of a rectangular box have been discussed in a
lot of works, some of them using zeta regularization; here we only mention some
of them. The foremost computation was performed by Lukosz [20, 21] for the elec-
tromagnetic field, by means of exponential regularization and Abel-Plana formula;
the same technique was used by Mamaev and Trunov [22, 23] (also see [24, 25]) to
discuss, amongst other models, the case of a conformal scalar field (2 ). Alternative
derivations of the total energy for a scalar field based on global zeta regularization
were given by Ruggerio, Vilanni and Zimerman [31, 32] in two and three spatial
dimensions, and by Ambjørn and Wolfram [3] (see also [4] for the electromagnetic
case) in the case of a multidimensional rectangular cavity for several boundary con-
ditions. The same configurations were later re-examined by X. Li, Cheng, J. Li and
Zhai [19] by means of a zeta strategy, and by A. Edery [8, 9] using a so called “mul-
tidimensional cut-off technique”. Let us also cite the papers by Estrada, Fulling et
al. [15, 16], on which we return later in this Introduction. Finally, let us mention
the monographies of Elizalde et al. [10, 11] and Bordag et al. [5]; these can be taken
as standard references for the study of global aspects.
In the present paper we consider global observables, such as the total energy and
forces, mainly to complete the analysis of local aspects (i.e., the VEV of the stress-
energy tensor), which occupy most of our analysis. Our series representations for
these global observables are different, but equivalent to the ones of [5] (which appear
to converge exponentially, like ours).
Concerning local aspects in the previous literature for a scalar field in a rectangu-
lar box let us first mention the two seminal papers [1] and [2] by Actor. In these
2
These authors also consider the cases of electromagnetic, (massless) gluon and spinor fields in
(3 + 1)-dimensional Minkowski spacetime; furthermore, they derive the renormalized average over
the spatial domain of the stress-energy VEV.

3
papers d = 3, the framework is Euclidean and the author renormalizes by analytic
continuation the effective Lagrangian density and the VEV h0|φb 2 (x)|0i. The infor-
mation contained in these works is equivalent, in the language of our papers, to the
specification of the Dirichlet function Ds (x, y) := hδx |(−∆)s δy i along the diagonal
y = x (with ∆ the Laplacian and x, y ∈ Ω), including its analytic continuation with
respect to the complex parameter s. However, the VEV of the stress-energy tensor
depends on Ds (x, x) as well as on the derivatives of Ds (x, y) with respect to x and
y, at points of the diagonal (see Part I of the present series). For this reason the
results of [1, 2] are not sufficient to determine the stress-energy VEV, which in fact
is not mentioned therein.
The renormalized VEV of the stress-energy tensor is derived in a work of Svaiter et
al. [30]; the latter employs, again, analytic continuation methods in a formulation
closely related to the approach of Actor. However, in [30] the authors consider an
infinite rectangular waveguide, rather than a box; more precisely, it is assumed d = 3
and the spatial domain is Ω = (0, a1 ) × (0, a2 ) × R. Apart from the domain, there
are other differences bewteen the approach of Svaiter et al. and ours; in particular,
the methods employed in [30] ultimately yield a representation of the stress-energy
VEV via series converging with polynomial speed. Even though this approach could
be extended to treat a box domain (Ω = (0, a1 ) × (0, a2 ) × (0, a3 )), it would most
likely yield similar series representations, whose polynomial convergence would be
slower compared to the exponential convergence of ours.
To go on, let us return to the already mentioned works by Estrada, Fulling et al.
[15, 16]. Therein the configuration of a 2-dimensional rectangular box is analysed for
both Dirichlet and Neumann boundary conditions (along with some related models,
such as the Casimir piston and the so-called “Casimir pistol”). The cited works
introduce a regularized version of the stress-energy VEV, based on an exponential
cutoff; this can be expressed in terms of the so-called cylinder kernel, which was
also considered for other reasons in Parts I and II of this series. The regularized
stress-energy VEV is computed by the method of images, allowing to express it as
a sum over infinitely many optical paths. By integration of the above VEV, the
authors of [15, 16] also obtain the regularized total energy and the regularized force
acting on a side of the boundary. Their position of principle is that the theory with
a cutoff is a more realistic description of the physical system under investigation;
nonetheless, they also point out that the VEV of the observables for the system under
analysis can be renormalized retaining only their regular parts with respect to the
cut-off (an idea somehow related to what we call the “extended zeta approach”).
Renormalization along these lines is carried out for global observables like the total
energy, and hinted at for local observables (namely, for the energy density).
The present paper is organized as follows. In Section 2, as in the other papers of
this series, we report a summary of results from Part I to be used in the present
work. Section 3 and the related Appendices A, B, C are the core the paper; therein

4
we treat the box configuration Ω = (0, a1 )×...×(0, ad ) in arbitrary spatial dimension
d, for Dirichlet boundary conditions. Our starting point is the heat kernel hδx |e−tA δy i
for which we derive two different series representations capturing, respectively, the
behavior for small and large t. As emphasized in Part I, the heat kernel determines
an integral representation of the Dirichlet function Ds (x, y) := hδx |(−∆)s δy i, which
can be used to construct the analytic continuation in s of the latter. Combining
these general facts with the series representation for the heat kernel cited above, we
ultimately produce series expansions for the analytic continuations of Ds (x, y) and
its derivatives, with the previously mentioned exponential speed of convergence;
these determine all the local or global renormalized observables indicated before,
from the stress-energy VEV to the force on each side of the box.
In Section 4, the previous general results are specialized to the cases with d ∈ {1, 2}.
For d = 1, we recover from a different viewpoint the results already obtained in
Section 6 of Part I where, as a first example of our general formalism, we discussed
a massless scalar field on a segment (0, a) (see this section of I for some references on
this case, especially [26]). For d = 2, working on Ω = (0, a1 ) × (0, a2 ) and using the
previously mentioned series expansions, we produce several graphs; some of them
refer to the components of the stress-energy VEV for some choices of a1 , a2 , while
the others are about the further observables in which we are interested.
Before moving on, let us mention that many of the results presented in this paper
have been derived with the aid of the software Mathematica for both symbolic and
numerical computations.

2 A summary of results from Part I

2.1 General setting. Throughout the paper we use natural units, so that

c=1, ~=1. (2.1)

Our approach works in (d + 1)-dimensional Minkowski spacetime, which is identified

with Rd+1 using a set of inertial coordinates

x = (xµ )µ=0,1,...,d ≡ (x0 , x) ≡ (t, x) ; (2.2)

the Minkowski metric is (ηµν ) = diag(−1, 1, ... , 1) . We fix a spatial domain Ω ⊂ Rd

and a background static potential V : Ω → R. We consider a quantized neutral, scalar
field φb : R×Ω → Lsa (F) (F is the Fock space and Lsa (F) are the selfadjoint operators
on it); suitable boundary conditions are prescribed on ∂Ω. The field equation reads

0 = (−∂tt + ∆ − V (x))φ(x,
b t) (2.3)
Pd
(∆ := i=1 ∂ii is the d-dimensional Laplacian). We put

A := −∆ + V , (2.4)

5
keeping into account the boundary conditions
R on ∂Ω, and consider the Hilbert space
2
L (Ω) with inner product hf |gi := Ω dx f (x)g(x). We assume A to be selfadjoint
in L2 (Ω) and strictly positive (i.e., with spectrum σ(A) ⊂ [ε2 , +∞) for some ε > 0).
We often refer to a complete orthonormal set (Fk )k∈K of (proper or improper) eigen-
functions of A with eigenvalues (ωk2 )k∈K (ωk > ε for all k ∈ K). Thus
Fk : Ω → C; AFk = ωk2 Fk ;
(2.5)
hFk |Fh i = δ(k, h) for all k, h ∈ K .
R
The labels k ∈ K can include both discrete and continuous parameters; K dk indi-
cates summation over all labels and δ(h, k) is the Dirac delta function on K.
We expand the field φb in terms of destruction and creation operators corresponding
to the above eigenfunctions, and assume the canonical commutation relations; |0i ∈
F is the vacuum state and VEV stands for “vacuum expectation value”.
The quantized stress-energy tensor reads (ξ ∈ R is a parameter)
 
1
Tµν := (1 − 2ξ) ∂µ φ ◦ ∂ν φ −
b b b − 2ξ ηµν (∂ λ φb ∂λ φb + V φb2 ) − 2ξ φb ◦ ∂µν φb ; (2.6)
2
in the above we put A b◦B b := (1/2)(AbB
b+B b A)
b for all A,
bB b ∈ Lsa (F), and all the
bilinear terms in the field are evaluated on the diagonal (e.g., ∂µ φb ◦ ∂ν φb indicates
the map x 7→ ∂µ φ(x)
b ◦ ∂ν φ(x)).
b The VEV h0|Tbµν |0i is typically divergent.
2.2 Zeta regularization. The zeta-regularized field operator is

φbu := (κ−2 A)−u/4 φb , (2.7)

where A is the operator (2.4), u ∈ C and κ > 0 is a “mass scale” parameter; note
that φbu |u=0 = φ,
b at least formally. The zeta regularized stress-energy tensor is
  
1 
Tµν := (1−2ξ)∂µ φ ◦ ∂ν φ − −2ξ ηµν ∂ λ φbu ∂λ φbu +V (φbu )2 − 2ξ φbu ◦ ∂µν φbu . (2.8)
bu bu bu
2
u
The VEV h0|Tbµν |0i is well defined for <u large enough (see the forthcoming subsec-
tion 2.5); moreover, in the region of definition it is an analytic function of u. The
same can be said of many related observables (including global objects, such as the
total energy VEV).
For any one of these observables, let us denote with F(u) its zeta-regularized version
and assume this to be analytic for u in a suitable domain U0 ⊂ C. The zeta approach
to renormalization can be formulated in two versions.
i) Restricted version. Assume the map U0 → C, u 7→ F(u) to admit an analytic
continuation (indicated with the same notation) to an open subset U ⊂ C with
U 3 0 ; then we define the renormalized observables as
Fren := F(0) . (2.9)

6
ii) Extended version. Assume that there exists an open subset U ⊂ C with U0 ⊂ U,
such that 0 ∈ U and the map u ∈ U0 7→ F(u) has an analytic continuation
P+∞to U \{0}
(still denoted with F). Starting from the Laurent expansion F(u) = k=−∞ Fk uk ,
we introduce the regular part (RP F)(u) := +∞ k
P
k=0 Fk u and define

Fren := (RP F)(0) . (2.10)

Of course, if F is regular at u = 0 the defnitions (2.9) (2.10) coincide.

Differently from the other papers of this series, the observables considered in the
present Part IV will never exhibit singularities at u = 0; thus, we will never need to
use the prescription (ii).
In the sequel we will often refer to the stress-energy VEV, for which the prescription
(i) gives
u
h0|Tbµν (x)|0iren := h0|Tbµν (x)|0i . (2.11)
u=0

2.3 Conformal and non-conformal parts of the stress-energy VEV. These

are indicated by the superscripts (♦) and (), respectively; they are defined by
(♦) ()
h0|Tbµν |0iren = h0|Tbµν |0iren + (ξ −ξd ) h0|Tbµν |0iren , (2.12)

where we are considering for the parameter ξ the critical value

d−1
ξd := . (2.13)
4d
2.4 Integral kernels. If B is a linear operator in L2 (Ω), its integral kernel is the
(generalized) function (x, y) ∈ Ω×Ω 7→ B(x, y) := hδx |BR δy i (δx is the Dirac delta
at x). The trace of B, assuming it exists, fulfills Tr B = Ω dx B(x, x) .
In the following subsections A is a strictly positive selfadjoint operator in L2 (Ω), with
a complete orthonormal set of eigenfunctions as in Eq. (2.5). In typical applications,
A is the operator (2.4).
2.5 The Dirichlet kernel and its relations with the stress-energy VEV.
For (suitable) s ∈ C, the s-th Dirichlet kernel of A is
Z
−s dk
Ds (x, y) := A (x, y) = 2s
Fk (x)Fk (y) . (2.14)
K ωk

If Ω is a bounded domain with smooth boundary ∂Ω and A = −∆ + V (with V a

smooth potential) is a strictly positive operator on L2 (Ω), considering the closure
Ω := Ω ∪ ∂Ω one can show that the map Ds ( , ) : Ω × Ω → C, (x, y) 7→ Ds (x, y)
is continuous along with all its partial derivatives up to order j ∈ N, for all s ∈ C

7
with <s > d/2+j/2 ; moreover, the eigenfunction expansion in Eq. (2.14) converges
absolutely and uniformly on Ω with all the derivatives up order j for <s > d + j/2.
Recalling Eq. (2.8), the regularized stress-energy VEV can be expressed as follows:
    
u u 1 1 x`
h0|T00 (x)|0i = κ
b +ξ D u−1 (x, y)+ −ξ (∂ ∂y` +V (x))D u+1 (x, y) , (2.15)
4 2 4 2
y=x
u u
h0|Tb0j (x)|0i = h0|Tbj0 (x)|0i = 0 , (2.16)
h0|Tbiju (x)|0i = h0|Tbjiu (x)|0i =

1   `

= κu − ξ δij D u−1 (x, y) − (∂ x ∂y` +V (x))D u+1 (x, y) +
4 2 2 (2.17)
  
1 
+ − ξ ∂xi yj − ξ ∂xi xj D u+1 (x, y)
2 2
y=x
u u
(h0|Tbµν (x)|0i is short for h0|Tbµν (t, x)|0i; indeed, the VEV does not depend on t).
u
Of course, the map Ω → C, x 7→ h0|Tbµν (x)|0i possesses the same regularity as the
functions x ∈ Ω 7→ D u±1 (x, x), ∂zw D u+1 (x, x) (z, w any two spatial variables); so,
2 2
u
due to the previously mentioned results, x 7→ h0|Tbµν (x)|0i is continuous on Ω for
<u > d + 1.
Assume that D u−1 and ∂zw D u+1 (for z, w any two spatial variables) have analytic
2 2
continuations regular at u = 0 (which happens for the configuration considered in
this paper); then, we can define
D± 1 (x, y) := D u±1 (x, y) , (2.18)
2 2 u=0

∂zw D 1 (x, y) := ∂zw D u+1 (x, y) . (2.19)

2 2 u=0
u
The renormalized stress-energy VEV h0|Tbµν (x)|0iren := h0|Tbµν (x)|0i|u=0 (see Eq.
(2.11)) can be expressed as
    
1 1 x `
h0|Tb00 (x)|0iren = +ξ D− 1 (x, y)+ −ξ (∂ ∂y` +V (x))D+ 1 (x, y) , (2.20)
4 2 4 2
y=x

h0|Tb0j (x)|0iren = h0|Tbj0 (x)|0iren = 0 , (2.21)

h0|Tbij (x)|0iren = h0|Tbji (x)|0iren =

1  
x`

= − ξ δij D− 1 (x, y) − (∂ ∂y` +V (x))D+ 1 (x, y) +
4 2 2 (2.22)
  
1 
+ − ξ ∂xi yj − ξ ∂xi xj D+ 1 (x, y) .
2 2
y=x
(More generally, in presence of a singularity at u = 0 we should consider the regular
parts of the functions κu D u−1 , κu ∂zw D u+1 ).
2 2

8
2.6 The heat kernel. For t ∈ [0, +∞), this is given by
Z
−tA 2
K(t ; x, y) := e (x, y) = dk e−t ωk Fk (x)Fk (y) . (2.23)
K

If Ω is a bounded with smooth boundary ∂Ω and A = −∆+V (V smooth) is strictly

positive, the map K(t ; , ) : Ω × Ω → C, (x, y) 7→ K(t ; x, y) is continuous along
with all its partial derivatives of any order for all t > 0 ; moreover, the eigenfunction
expansion in Eq. (2.23) converges absolutely and uniformly (the same holds for all
the corresponding derivatives).
2.7 The Dirichlet kernel as Mellin transform of the heat kernel. For suit-
able values of s ∈ C (see Part I), there holds
Z +∞
1
Ds (x, y) = dt ts−1 K(t ; x, y) . (2.24)
Γ(s) 0

2.8 The case of product domains. Factorization of the heat kernel. Let
A := −∆ + V and consider the case where

Ω = Ω1 × Ω2 3 x = (x1 , x2 ) , y = (y1 , y2 ) , (2.25)

V (x) = V1 (x1 ) + V2 (x2 ) (2.26)

(Ωa ⊂ Rda is an open subset, for a ∈ {1, 2}; d1 + d2 = d); assume the boundary
conditions on ∂Ω to arise from suitable boundary conditions prescribed separately
on ∂Ω1 and ∂Ω2 so that, for a = 1, 2, the operators

Aa := −∆a + V (xa ) (2.27)

(with ∆a the Laplacian on Ωa ) are selfadjoint and strictly positive in L2 (Ωa ). Then,
the Hilbert space L2 (Ω) and the operator A can be represented as

L2 (Ω) = L2 (Ω1 ) ⊗ L2 (Ω2 ) , A = A1 ⊗ 1 + 1 ⊗ A2 . (2.28)

This implies, amongst else, that the heat kernels K(t ; x, y) := etA (x, y), Ka (t ; xa , ya )
:= etAa (xa , ya ) (a = 1, 2) are related by

K(t ; x, y) = K1 (t ; x1 , y1 ) K2 (t ; x1 , x2 ) . (2.29)

Similarly, writing K(t), Ka (t) (a ∈ {1, 2}) for the heat traces of A and Aa (a ∈
{1, 2}), respectively, we have

K(t) = K1 (t) K2 (t) . (2.30)

The above considerations have obvious extensions to product domains with more
than two factors.

9
2.9 Pressure on the boundary. This is the force per unit area produced by the
quantized field inside Ω at a point x ∈ ∂Ω. We first consider, for <u large, the
regularized pressure pu (x) with components

pui (x) := h0|Tbiju (x)|0i nj (x) ; (2.31)

here and in the remainder of this paper, n(x) ≡ (ni (x)) is the unit outher normal
at x ∈ ∂Ω. For Dirichlet boundary conditions the above definition implies
  
u u 1 x` 1
pi (x) = κ − δij ∂ ∂y` + ∂xi yj D u+1 (x, y) nj (x) . (2.32)
4 2 2
y=x

We can define the renormalized pressure by analytic continuation as

pren u
i (x) := pi (x) (2.33)
u=0

(more generally, one should take the regular part if u = 0 is a singular point; this
will never happen in the present paper). Alternatively, we could put
 
0
pi (x) := 0 lim0 h0|Tij (x )|0iren nj (x) .
ren b (2.34)
x ∈Ω,x →x

Prescriptions (2.33) (2.34) do not always agree (for a counterexample, see Section
5 of Part II [13] of this series of papers). In Part I we conjectured that the two
approaches agree when both of them give a finite result; this conjecture is true in
the present case of a box as shown in subsection 3.6).
Contrary to the previous works of this series [12, 13, 14], here we first discuss the
pressure and next pass to the total energy; we make this choice because quite dif-
ferent computation techniques are employed for the stress-energy VEV and for the
pressure, on the one hand, and for the total energy and integrated force, on the
other hand.
2.10 The total energy. The zeta-regularized total energy is
Z
u u
E := dx h0|Tb00 (x)|0i = E u + B u ; (2.35)
Ω

the second equality is proved after defining the regularized bulk and boundary en-
ergies, which are
κu κu
Z
1−u
u
E := dx D (x, x) =
u−1 Tr A 2 , (2.36)
2 Ω 2 2
 Z
u u 1 ∂
B := κ −ξ da(x) D u+1 (x, y) . (2.37)
4 ∂Ω ∂ny 2
y=x

10
One has B u = 0 for Ω bounded and either Dirichlet or Neummann boundary con-
ditions on ∂Ω.
Assuming the functions (2.36) (2.37) to be finite and analytic for suitable u ∈ C,
we define the renormalized energies by the restricted (or generalized) zeta approach;
for example, we put
E ren := E u (2.38)
u=0
(again, if a singularity appears at u = 0 we should take the regular part).
2.11 Forces on the boundary. Let O ⊂ ∂Ω; we consider (for <u large) the
regularized integrated force acting on O, i.e.,
Z
u
FO := da(x) pu (x) (2.39)
O
u
(p (x) is the regularized pressure of Eq. (2.31)). We can define the renormalized
total force on O as
Fren
O := FO
u
(2.40)
u=0
(once more, the regular part should be taken if u = 0 is a singular point, a variation
which we shall never employ in the present paper). Alternatively, we could put
Z
ren
FO := da(x) pren (x) , (2.41)
O

(pren (x) is the renormalized pressure, defined according to either Eq. (2.33) or Eq.
(2.34)). In general the alternatives (2.40) (2.41) give different results.

3 The case of a massless field inside a box

3.1 Introducing the problem. In the present section we analyse the model of a
massless scalar field confined within a d-dimensional box, with no external potential;
more precisely, we assume

Ω = ×di=1 (0, ai ) with ai > 0 for i ∈ {1, ..., d} , V =0. (3.1)

The boundary ∂Ω of the spatial domain is composed by the sides

πp,λ := {x ∈ Rd | xp = λ ap , xi ∈ [0, ai ] for i 6= p}
(3.2)
for p ∈ {1, ..., d}, λ ∈ {0, 1} ;
for the sake of simplicity, we restrict attention to the case where the field fulfills
Dirichlet boundary conditions on each one of these sides, meaning that
b x) = 0 for all t ∈ R, x ∈ πp,λ (p ∈ {1, ..., d}, λ ∈ {0, 1}) .
φ(t, (3.3)

11
As a matter of fact, all the results to be reported in the following could be generalized
to the case of Neumann or periodic boundary conditions, possibly including cases
where different boundary conditions are prescribed on different sides of the box;
moreover, the methods to be presented could be adapted with little effort to deal
with the cases of a massive scalar field (V = m2 ) and of a slab configuration (see
Part I) where
Ω = Ω1 × Rd2 , Ω1 = ×di=1
1
(0, ai ) ⊂ Rd1 (d1 + d2 = d) . (3.4)
None of these generalizations will be considered in the present paper.
3.2 The heat kernel. Similarly to the model with a harmonic potential consid-
ered in our previous work [14], in the present setting we are dealing with a product
domain configuration, in the sense of subsection 2.8. Working in standard Cartesian
coordinates (xi )i=1,...,d , the Hilbert space and the fundamental operator A := −∆
can berepresented as
O d
L2 (Ω) = L2 (0, ai ) , (3.5)
i=1

A = A1 ⊗ 1 ⊗ ... ⊗ 1 + ... + 1 ⊗ ... ⊗ 1 ⊗ Ad ,

d2 (3.6)
Ai := − 2 on L2 (0, ai ) (i ∈ {1, ..., d}) ;
dx
for each operator Ai we assume the induced Dirichlet boundary conditions in xi = 0
and xi = ai .
According to the general considerations of subsection 2.8 (see, in particular, Eq.
(2.29)), in this situation the heat kernel associated to A factorizes; more precisely,
we have
d
Y
K(t ; x, y) = Ki (t ; xi , y i ) , (3.7)
i=1
where, for i ∈ {1, ..., d}, Ki indicates the heat kernel of Ai . Hereafter we compute
the heat kernels Ki , giving for all of them two distinct representations; these are
suited to describe, respectively, the behaviour of the kernels Ki for small and large t
(in a sense to be made more precise in the following) (3 ). Clearly, each one of these
representations yields, in turn, alternative expressions for the total heat kernel (3.7).
3.2.1 First representation for the heat kernel (useful for large t). Let us
begin noting that, for any i ∈ {1, ..., d}, a complete orthonormal set of eigenfunctions
(Fki )ki ∈Ki for Ai in L2 (0, ai ), with corresponding eigenvalues (ωk2i )ki ∈Ki , is given by
r  
i 2 i 2 2 ni π
Fki (x ) := sin(ki x ) , ωki := ki for ki ∈ Ki ≡ ni = 1, 2, 3, ... . (3.8)
ai ai
3
For further information about approximate evaluations of the heat kernel for a box configura-
tion, see, e.g., [7, 17, 18].

12
Using the eigenfunction expansion (2.23), we obtain for the 1-dimensional heat kernel
Ki the following expression (4 ):
+∞ 2 2
i 2 X − nai 2π t  ni π i   ni π i 
i
Ki (t ; x , y ) = e i sin y sin x . (3.9)
ai n =1 ai ai
i

The above relation, along with Eq. (3.7), yields

2d X 2
K(t ; x, y) = e−ωn t Cn (x, y) , (3.10)
a1 ... ad d
n∈N

where, for the sake of brevity, we put

N := {1, 2, 3, ...} , (3.11)
d d
n2 π 2
   
X
i
Y ni π i ni π i
ωn2 := , Cn (x, y) := sin y sin x for n := (ni )i=1,...,d .
i=1
a2i i=1
a i ai

Eq. (3.10) is easily seen to give a large t expansion for the heat kernel K(t ; x, y) of
A; with this, we mean that the series over n ∈ Nd in the cited equation is mainly
determined by the terms corresponding to small values of ni , for i ∈ {1, ..., d}.
3.2.2 Second representation for the heat kernel (useful for small t). Let
us move on and note that, with little effort (namely, writing the sines in terms of
complex exponentials), we can rephrase Eq. (3.9) as
+∞ 2 2
i 1 X − nai 2π t h i nai π (xi −yi )
i
n π
i i (xi +y i )
i
Ki (t ; x , y ) = e i e i − e ai . (3.12)
2ai n =−∞
i

Using the Poisson summation formula (5 ) and noting that

Z +∞ 2
− π2 t z 2 i π ((xi ±y i )−2ai hi )z ai (2ai hi −(xi ±y i ))2
dz e a
i e ai = √ e− 4t

−∞ πt (3.13)
for hi ∈ Z with Z := {0, ±1, ±2, ...} ,
4
Let us mention that the series in Eq. (3.9) could be explicitly evaluated to yield
−(xi−y i )2 −(xi+y i )2
i i e 4t
 ai (xi −y i ) − a2i  e 4t  ai (xi +y i ) − a2i 
Ki (t ; x , y ) = √ θ3 −i ,e t − √ θ3 − i ,e t ,
4πt 2t 4πt 2t
where θ3 ( , ) denotes one of the Jacobi elliptic theta functions; see [6] for more details.
5
The Poisson summation formula states that, for any sufficiently regular function f : R → C,
there holds
+∞
X +∞
X Z +∞
f (n) = fˆ(h) , where fˆ(h) := dz e−2iπhz f (z)
n=−∞ h=−∞ −∞

(fˆ is, essentially, the Fourier transform of f ). Eq. (3.13) gives fˆ in the case we are considering.

13
it follows from Eq. (3.12) that (6 )
+∞ h
i i 1 X (2a h −(xi −y i ))2
− i i 4t
(2a h −(xi +y i ))2
− i i 4t
i
Ki (t ; x , y ) = √ e −e . (3.14)
4πt hi =−∞

The above identity can be rephrased as follows:

1 X 1 2 i i 2
Ki (t ; xi , y i ) = √ δli e− t ai (hi −Uli (x ,y )) (3.15)
4πt h ∈Z, l ∈{1,2}
i i

where
(
 xi −y i
1 for li = 1 i i 2ai
for li = 1
δli := , Uli (x , y ) := xi +y i . (3.16)
−1 for li = 2 for li = 2
2ai

Eq. (3.15), along with Eq. (3.7), allows us to infer for the heat kernel the alternative
representation
1 X 1
K(t ; x, y) = δl e− t bhl (x,y) , (3.17)
(4πt)d/2
h∈Zd , l∈{1,2}d

where, for simplicity of notation, we have put (recall Eq. (3.16))

d
Y
h := (hi )i=1,...,d , l := (li )i=1,...,d , δl := δli ,
i=1
d
(3.18)
X
bhl (x, y) := a2i (hi −Uli (xi , y i ))2 .
i=1

Notice that, for small t, the sum of the series appearing in Eq. (3.17) is mainly
determined by the terms corresponding to small values of |hi |, for i ∈ {1, ..., d}; thus,
the mentioned equation yields a small t expansion for the heat kernel K(t ; x, y) .
3.2.3 Considerations on the sign of some coefficients. Before moving on,
let us emphasize a number of facts on the expansions (3.10) (3.17); we will resort
to them in the following subsections, when performing the analytic continuation of
the Dirichlet kernel and of its derivatives.
6
The very same result of Eq. (3.14) could be obtained via the method of reflections, starting
with the heat kernel associated to the laplacian −∂x1 x1 on R , i.e.
1 |x−y|2
− 4t
K(t ; x, y) = e .
(4πt)d/2

14
i) On the one hand, we have

ωn2 > 0 for all n ∈ Nd . (3.19)

ii) On the other hand, notice that

bhl (x, y) > 0 for all h ∈ Zd , l ∈ {1, 2}d , x, y ∈ Ω . (3.20)

In particular, since   1 1
i i − 2 , 2 for li = 1
Uli (x , y ) ∈ (3.21)
[0, 1] for li = 2
(see the definition of Uli in Eq. (3.16)), it follows that

bhl (x, y) = 0 ⇔

 hi = 0, li = 1, y i = xi ∈ [0, ai ] (3.22)
⇔ for each i ∈ {1, ..., d} , one has or hi = 0, li = 2, y i = xi = 0 ;
or hi = 1, li = 2, y i = xi = ai


let us stress that this implies, in particular, bhl (x, y) 6= 0 for x 6= y.

3.3 The Dirichlet kernel. Consider the integral representation (2.24) for the
Dirichlet kernel Ds in terms of the heat kernel K of A . We are going to construct
the analytic continuation of Ds in the style of Minakshisundaram (see [27]); to this
purpose, let us fix arbitrarily
T ∈ (0, +∞) (3.23)
and re-express the cited integral representation as

Ds (x, y) = Ds(>) (x, y) + Ds(<) (x, y) , where (3.24)

Z +∞
(>) 1
Ds (x, y) := dt ts−1 K(t ; x, y) , (3.25)
Γ(s) T
Z T
(<) 1
Ds (x, y) := dt ts−1 K(t ; x, y) (3.26)
Γ(s) 0
(>) (<)
(notice that Ds and Ds depend on T , but their sum Ds does not!). The idea
we are going to pursue in the following is to substitute into Eq.s (3.25) (3.26),
respectively, the large and small t expansions (3.10) (3.17) for the heat kernel K (7 ).
7
In these manipulations (and in some related computations) we often take for granted that
certain series can be integrated or differentiated term by term. In all cases under analysis, rigorous
justifications could be given using the Lebesgue dominated convergence theorem or the Fubini-
Tonelli theorem, but we will not go into the details; the estimates on the series in Appendices A
and C could be connected to such rigorous proofs.

15
3.3.1 Series expansion and analytic continuation of Ds(>) (x, y). Using Eq.s
(3.10) (3.25), we readily infer
Z +∞
2d X 2
(>)
Ds (x, y) = Cn (x, y) dt ts−1 e−ωn t . (3.27)
a1 ...ad Γ(s) d T
n∈N

Concerning the integral over (T, +∞), via the change of variable τ := ωn2 t (recall
that ωn2 > 0 for all n ∈ Nd ; see Eq. (3.19)), we obtain
Z +∞ Z +∞
s−1 −ωn 2t −2s
dt t e = ωn dτ τ s−1 e−τ = ωn−2s Γ(s, ωn2 T ) for s ∈ C , (3.28)
T 2T
ωn

where the last passage contains the upper incomplete gamma function (see [29],
p.174, Eq.8.2.2)
Z +∞
Γ(s, z) := dw e−w ws−1 (s ∈ C, z ∈ (0, +∞)) . (3.29)
z

Summing up, we have

2d X
Ds(>) (x, y) = ωn−2s Γ(s, ωn2 T ) Cn (x, y) . (3.30)
a1 ...ad Γ(s) d
n∈N

The above expression can be used to evaluate derivatives of any order of the function
(>)
Ds (x, y); for example, for any pair of spatial variables z, w, we obtain
2d X
∂zw Ds(>) (x, y) = ωn−2s Γ(s, ωn2 T ) ∂zw Cn (x, y) . (3.31)
a1 ...ad Γ(s) d
n∈N

Let us anticipate that the series in the right-hand sides of Eq.s (3.30) (3.31) converge
for all s ∈ C, even for y = x (see subsection 3.4 and Appendices A, C for more
details); so, Eq.s (3.30) (3.31) yield automatically the analytic continuations of the
(>) (>)
maps s 7→ Ds (x, y), ∂zw Ds (x, y) to the whole complex plane.
3.3.2 Series expansion and analytic continuation of Ds(<) (x, y). Proceeding
(>)
similarly to what we did for the function Ds (x, y), we can use Eq.s (3.17) (3.26)
to deduce
Z T
1 X d 1
(<)
Ds (x, y) = d/2
δl dt ts− 2 −1 e− t bhl (x,y) . (3.32)
(4π) Γ(s) d d 0
h∈Z , l∈{1,2}

To go on, for β > 0 and s ∈ C satisfying the conditions in the forthcoming Eq.
(3.34), let us introduce the function
Z 1
β
Ps (β) := dτ τ s−1 e− τ ; (3.33)
0

16
it is easy to check that

s−1

for β = 0, <s > 0
Ps (β) = s (3.34)
β Γ(−s, β) for β > 0

where, again, Γ( , ) denotes the upper incomplete gamma function. Starting from
Eq. (3.33), we find as well that

∂β` Ps (β) = (−1)` Ps−` (β) for all ` ∈ {0, 1, 2, ...} . (3.35)

Let us return to Eq. (3.32) (recalling that bhl (x, y) > 0 for all h ∈ Zd , l ∈ {1, 2}d ;
see Eq. (3.20)) and make therein the change of variable τ := t/T ; comparing with
the definition (3.33) of P• ( ), we get
d
T s− 2
 
(<)
X bhl (x, y)
Ds (x, y) = δl Ps− d . (3.36)
(4π)d/2 Γ(s) d d
2 T
h∈Z , l∈{1,2}

The above result can be used, along with Eq. (3.35), to infer analogous representa-
(<)
tions for the derivatives of any order of Ds ; for example, if z, w are any two spatial
variables, derivating term by term Eq. (3.36) we obtain

∂zw Ds(<) (x, y) = (3.37)

d
T s− 2 −2
     
X bhl bhl
δl Ps− d −2 ∂z bhl ∂w bhl − T Ps− d −1 ∂zw bhl (x, y) .
(4π)d/2 Γ(s) 2 T 2 T
h∈Zd , l∈{1,2} d

Let us point out that, due to the results reported in Eq. (3.22), we have

bhl (x, y) = 0 only for y = x and for

(3.38)
a finite number of terms in the series of Eq.s (3.36) (3.37) .

The above mentioned terms of Eq.s (3.36) (3.37) deserve special attention, and we
must use the first equality in (3.34) to evaluate them; on the contrary, for the in-
finitely many terms with bhl (x, y) > 0 , the second line in Eq. (3.34) gives expressions
in terms of upper incomplete gamma functions. In this way we obtain
 
d
T s− 2  X 
Ds(<) (x, y) =  δl 
+
(4π)d/2 Γ(s)(s − d2 ) 
h∈Zd , l∈{1,2}d
s.t. bhl (x,y)=0 (3.39)
 !
1 X s− d2 d bhl
+ δl bhl Γ −s , (x, y) .
(4π)d/2 Γ(s) 2 T
h∈Zd , l∈{1,2}d
s.t. bhl (x,y)>0

17
Let us repeat that the first sum in the above expression contains finitely many terms.
Notice that the term in the first line of Eq. (3.39) is related to the first equality in
Eq. (3.34) which, in principle, would require <s > d/2 ; however, this term makes
sense for all complex s except s = d/2 , where a simple pole appears. Moreover, the
series in the second line of Eq.s (3.39) can be proved to converge for any complex
s; we defer a comprehensive discussion of this statement to subsection 3.4 (see also
Appendix A).
In view of the above remarks, Eq. (3.39) gives automatically the analytic continua-
(<)
tion of Ds (x, y) to a meromorphic function of s on the whole complex plane, with
a simple pole singularity only at
s = d/2 (3.40)
for x, y such that the first sum in Eq. (3.41) is non-empty; this happens only for
y = x due to Eq. (3.22).
(<)
A similar analysis can be made for the derivatives of Ds . For example, if z, w are
any two spatial variables, we obtain the following expression from Eq. (3.37):
!
s− d2 −1
T X
∂zw Ds(<) (x, y) = − δl ∂zw bhl (x, y) +
(4π)d/2 Γ(s)(s − d2 − 1) d d h∈Z , l∈{1,2}
s.t. bhl (x,y)=0
   
1 X s− d2 −2 d bhl
+ δl bhl Γ +2−s , ∂z bhl ∂w bhl + (3.41)
(4π)d/2 Γ(s) 2 T
h∈Zd , l∈{1,2} d
s.t. bhl (x,y)>0
  
d bhl
−Γ +1−s , bhl ∂zw bhl (x, y) .
2 T

Again, the first of the above two sums is made of finitely many terms; besides,
contrary to what one could expect from Eq. (3.37), this sum contains no term with
the first order derivatives ∂w bhl (x, y), ∂z bhl (x, y) because they vanish if bhl (x, y) =
0. By considerations analogous to those made above for Eq. (3.39) (based on
the convergence of the second sum for all s ∈ C), we infer that Eq. (3.41) gives
(<)
automatically the analytic continuation of ∂zw Ds (x, y) to a meromorphic function
of s on the whole complex plane, with a simple pole singularity only for y = x at

s = d/2 + 1 . (3.42)

Let us stress that Eq.s (3.39) (3.41) are just the original Eq.s (3.36) (3.37), rewritten
separating the terms with bhl (x, y) = 0 for a better understanding of the behaviour
with respect to s . In the sequel, even when considering meromorphic continuations,
we will always refer to the more concise representations (3.36) (3.37).

18
3.3.3 Conclusions for the Dirichlet kernel. Using Eq. (3.24) and expressions
(>) (<)
(3.30) (3.36) for the functions Ds , Ds , respectively, we obtain the analytic con-
tinuation of the full Dirichlet kernel Ds (x, y) to a meromorphic function on the
whole complex plane. Similar results hold for the derivatives of the Dirichlet kernel
(see Eq.s (3.31) (3.37)).
The only singularity of Ds (x, y) is a simple pole for

y = x and s = d/2 , (3.43)

while ∂zw Ds (x, y) (for any pair of spatial variables z, w) has a simple pole for

y = x and s = d/2 + 1 . (3.44)

In particular the analytic continuation of D u−1 (x, y)|y=x and ∂zw D u+1 (x, y)|y=x ,
2 2
required for the evaluation of the regularized stress-energy VEV and pressure, are
both regular at u = 0 .
3.4 Convergence and remainder estimates for the series in Eq.s (3.30)
(3.31) and (3.36) (3.37). This subject is discussed in more detail in Appendix
A; here we only report the main results. In the mentioned Appendix we show that
the series cited in the title of this subsection are absolutely convergent; moreover,
we derive fully quantitative remainder estimates when these series are approximated
by finite sums.
To report here these estimates we need some notations, introduced hereafter. First
of all we put

a := min {ai } , A := max {ai } ;

i∈{1,...,d} i∈{1,...,d}
!1/2
d
X (3.45)
|z| := zi2 for z = n ∈ Nd or z = h ∈ Zd .
i=1
√
Besides, for N ∈ (2 d, +∞), α ∈ (0, 1), β ∈ (0, +∞), σ, ρ ∈ R, we set
(d)
HN (α, β; σ, ρ) := (3.46)
√ !d−1
π d/2 √ 2
 
N− d 2 d+ρ
d+ρ
√ Γ(σ, (1−α)βN ) Γ ; α β(N −2 d) ;
(1−α)σ (α β) 2 Γ( d2 ) N −2 d 2
it can be shown that there holds the asymptotic expansion

(d) π d/2 β σ−2 e−4α βd −βN (N −4αβ √d) 2σ+ρ+d−4

HN (α, β; σ, ρ) = e N (1 + O(N −1 ))
α(1−α) Γ( d2 ) (3.47)
for N → +∞ .

19
Finally, we put
 √ !!2σ √ !!2σ 
(d) d d
Ca,A (σ, N ) := max  a 1− , A 1+ . (3.48)
N N

Having introduced the above notations, in the next two subsections we report the
(>) (<)
remainder estimates of Appendix A for the series expansions of Ds , Ds and
(d)
of their derivatives. In all cases the remainder is controlled by the function HN ;
due to the exponential decay of this function for large N (see Eq. (3.47)), good
approximations of all the series under investigation can be obtained by just summing
the first few terms.
3.4.1 Estimates for the series (3.30) (3.31). Consider any s ∈ C; keeping in
mind Eq. (3.30), for any N ∈ (0, +∞) let us write
(>) (>)
Ds(>) (x, y) = Ds,N (x, y) + Rs,N (x, y) ,
(>) 2d X
Ds,N (x, y) := ωn−2s Γ(s, ωn2 T ) Cn (x, y) ,
a1 ...ad Γ(s) d (3.49)
n∈N , |n|6N

(>) 2d X
Rs,N (x, y) := ωn−2s Γ(s, ωn2 T ) Cn (x, y) .
a1 ...ad Γ(s)
n∈Nd , |n|>N

This equation implies a similar representation for the (analytically continued) deriva-
(>) (>) (>)
tives of Ds in terms of the derivatives of Ds,N and Rs,N . For the remainder function
(>)
Rs,N (x, y) and for its derivatives with respect to any two spatial variables z, w, we
have the following uniform estimates:

max(a2<s , A2<s ) (d) π2 T

 
(>)
Rs,N (x, y) 6 H α, 2 ; <s, −2<s
a1 ...ad π 2<s |Γ(s)| N A
r (3.50)
√ √ A |<s|
for either <s > 0, N > 2 d or <s < 0, N > 2 d+ ;
π αT
max(a2<s , A2<s ) π2 T
 
(>) (d)
∂zw Rs,N (x, y) 6 H α, 2 ; <s, 2(1−<s)
a1 ...ad a2 π 2(<s−1) |Γ(s)| N A
r (3.51)
√ √ A (1−<s)
for either <s > 1, N > 2 d or <s < 1, N > 2 d+ .
π αT
In the above, α is a parameter that can be freely chosen in (0, 1); of course, the best
choice is the one minimizing the right-hand sides of Eq.s (3.50) and (3.51), which
depends on the other parameters (e.g., N, T ) involved in these considerations.

20
3.4.2 Estimates for the series (3.36) (3.37). Let s ∈ C, and exclude the case
(3.43); keeping in mind Eq. (3.36), for any N ∈ (0, +∞) we put
(<) (<)
Ds(<) (x, y) = Ds,N (x, y) + Rs,N (x, y) ,
d
T s− 2
 
(<)
X bhl (x, y)
Ds,N (x, y) := δl Ps− d ,
(4π)d/2 Γ(s) d d
2 T (3.52)
h∈Z , |h|6N, l∈{1,2}
d
T s− 2
 
(<)
X bhl (x, y)
Rs,N (x, y) := δl Ps− d .
(4π)d/2 Γ(s) d d
2 T
h∈Z , |h|>N, l∈{1,2}

The above equation can be used to derive similar representations for the (analyt-
(<) (<) (<)
ically continued) derivatives of Ds in terms of the derivatives of Ds,N and Rs,N ,
(<)
with the exclusion of the case (3.44) . For the remainder function Rs,N and for its
derivatives with respect to any two spatial variables z i , wj (i, j ∈ {1, ..., d}), we have
the following uniform estimates:
√
(d)
!
(<) Ca,A (<s − d2 , N ) (d) a2 (1− Nd )2 d
Rs,N (x, y) 6 HN α, ; −<s, 2<s−d
π d/2 |Γ(s)| T 2
s (3.53)
d √ d √ 1 (<s− d
)T
2
for either <s 6 , N > 2 d or <s > , N > 3 d+ ;
2 2 a α
(<)
∂zi wj Rs,N (x, y) 6 (3.54)
 √ !2   2
√
d 2
!
1 d (d) d (d) a (1− ) d
 1+ A2 Ca,A <s− −2, N HN α, N
; +2−<s, 2<s−d−2
π d/2 |Γ(s)| N 2 T 2
√ !
  2 d 2
1 (d) d (d) a (1− N ) d
+ δij Ca,A <s− −1, N HN α, ; +1−<s, 2<s−d−2 
2 2 T 2
s
d √ d √ 1 (<s− d2 −1)T
for either <s 6 +1, N > 2 d or <s > +1, N > 3 d+ .
2 2 a α
Again, the parameter α can be freely taken in (0, 1) and it is conventient to choose
for it the value minimizing the right-hand sides of Eq.s (3.53) (3.54), keeping into
account the choices made for the other parameters (in particular, N ).
3.5 The stress-energy tensor. Consider the representations deduced in subsec-
tion 3.3 for the analytic continuations of the Dirichlet kernel and of its derivatives.

21
Resorting to Eq.s (2.15-2.17), we obtain the following expressions for the components
of the regularized VEV of the stress-energy tensor:
u u,(>) u,(<)
h0|Tbµν (x)|0i = Tµν (x) + Tµν (x) , (3.55)
u,(•)
where, for • equal to > or <, Tµν (x) has the espression corresponding to Eq.s
(•)
(2.15-2.17), with Ds replaced by Ds . Thus, for i, j ∈ {1, ..., d}, we have
    
u,(•) u 1 (•) 1 x` (•)
T00 (x) = κ + ξ D u−1 (x, y) + − ξ ∂ ∂y` D u+1 (x, y) , (3.56)
4 2 4 2
y=x

u,(•) u,(•)
T0j (x) = Tj0 (x) = 0 , (3.57)
  
u,(•) 1 (•) ` (•)

Tij (x) = κu − ξ δij D u−1 (x, y) − ∂ x ∂y` D u+1 (x, y) +
4 2 2
   (3.58)
1 
(•)
+ − ξ ∂xi yj − ξ ∂xi xj D u+1 (x, y) .
2 2
y=x

To proceed notice that, concerning the analyticity of the above functions, there hold
considerations analogous to those presented in subsection 3.3.3 for the Dirichlet
kernel and its derivatives; in particular, it appears that u = 0 is a regular point
for each component of the regularized stress-energy VEV so that, according to the
restricted version of the zeta approach, we can simply put
u
h0|Tbµν (x)|0iren := h0|Tbµν (x)|0i . (3.59)
u=0

In the forthcoming subsections 4.1 and 4.2, dealing with the cases d = 1 and d = 2,
we will use approximate expressions for all the components of h0|Tbµν (x)|0iren ob-
(•)
tained replacing each Dirichlet function Ds in Eq.s (3.56-3.58) with the truncations
(•)
Ds,N of a fixed (sufficiently large) order N , given by Eq.s (3.49) (3.52). Let us re-
call that we have explicit remainder bounds for these truncations (see Eq.s (3.50)
(3.51) and Eq.s (3.53) (3.54)); these will allow us to infer error estimates for the
approximate expressions of h0|Tbµν (x)|0iren described above.
3.6 The pressure on the boundary. We refer to the general discussion of sub-
section 2.9; so, we have two alternative definitions for the renormalized pressure on
each of the sides πp,λ of the box (p ∈ {1, ..., d}, λ ∈ {0, 1}; see Eq. (3.2)).
Let x be any point interior to one of the sides πp,λ ; let us stress that we exclude x
to be on an edge of the box (i.e., on the intersection of two or more sides), where
the outer normal is ill-defined. As an example, let us assume x to be an inner point
of the side π1,0 , so that the unit outer normal at x is n(x) = (−1, 0, ..., 0). We first
consider the regularized pressure
pui (x) := h0|Tbiju (x)|0i nj (x) = − h0|Tbi1u (x)|0i ; (3.60)

22
this can be expressed using the general rule (2.32) for the case of Dirichlet boundary
conditions which, in the present case, gives (8 )
κu
pui (x) = − δi1 ∂x1 y1 D u+1 (x, y) =
4 2 y=x
(3.61)
κu
 
(>) (<)
= − δi1 ∂x1 y1 D u+1 (x, y) + D u+1 (x, y) .
4 2 y=x 2 y=x

The equality in the second line of the above equation, involving the derivatives of
(•)
the Dirichlet functions D u+1 , follows from Eq. (3.24) and will be useful for later
2
purposes.
On the one hand, we can put

pren u
i (x) := pi (x) = − h0|Tbi1u (x)|0i =
u=0 u=0
δi1
  (3.62)
(>) (<)
=− ∂x1 y1 D1/2 (x, y) + ∂x1 y1 D1/2 (x, y) .
4 y=x y=x

On the other hand, we have the alternative definition

   
ren 0 j 0
pi (x) := 0 lim0 h0|Tij (x )|0iren n (x) = − 0 lim0
b h0|Ti1 (x )|0iren . (3.63)
b
x ∈Ω, x →x x ∈Ω, x →x

In the next two paragraphs we show the equivalence of the alternative prescriptions
(3.62) (3.63) and the non-integrable divergence near the edges of the renormalized
pressure (evaluated, equivalently, according to either of the two cited prescriptions).
Before proceeding to the proof of the above statements, let us anticipate that in
subsections 4.1 and 4.2 (dealing with the cases d = 1 and d = 2, respectively) we
will evaluate the pressure starting from Eq. (3.62) and substituting the functions
(•) (•)
D1/2 therein with the truncations D1/2,N of a sufficiently large order N ; the errors
of these approximants will be evaluated using Eq.s (3.50) (3.51) and (3.53) (3.54).
3.6.1 Equivalence of prescriptions (3.62) (3.63). The proof of this equiva-
lence, given hereafter, uses arguments similar to the ones proposed in subsection 4.6
of Part II, where an analogous statement was derived for a domain Ω bounded by
orthogonal hyper-planes.
8
In the application of Eq. (2.32) to the present case, we use the previous expression for n(x)
and the fact that

∂xi yj Ds (x, y) =0 for all i, j ∈ {1, ..., d} such that i 6= 1 or j 6= 1 ;

y=x

this follows straightforwardly from the Dirichlet conditions prescribed on the boundary of Ω and
from the eigenfunction expansion in Eq. (2.14), taking into account the factorized structure of the
eigenfunctions in our case.

23
Let us consider again an inner point x = (0, x2 , ..., xd ) of the side π1,0 . When
expressing h0|Tbi1u (x)|0i|u=0 (see Eq. (3.62)) or limx0 →x h0|Tbi1 (x0 )|0iren (see Eq. (3.63))
in terms of the Dirichlet kernel and of its series expansions, there is only one type of
potentially troublesome terms, which could give different contributions in the two
u,(<)
cases. These are terms arising from the summand Ti1 (x) in Eq. (3.58), when we
(<)
use for Ds and its derivatives the series expansions (3.36) (3.37); more precisely,
potential troubles could arise from terms in the cited expansions with

bhl (x, x) = 0 and bhl (x0 , x0 ) 6= 0 for x0 ∈ Ω , (3.64)

corresponding to the choice

h1 = 0, l1 = 2 and hi = 0, li = 1 for i 6= 1 . (3.65)

u,(<)
Consider the expression for Ti1 (x0 ) obtained using Eq. (3.58) and Eq.s (3.36)
(3.37), where x0 is either a point in the interior of the spatial domain Ω or the
boundary point x under consideration; the contribution to this expression from
problematic terms, with (h, l) as in Eq. (3.65), is
  u u−d−1  
1 κ T 2 u−d−1 1
δi1 −ξ d/2 u+1 P u−d−1 (z)+z P u−d−3 (z) ≡ f (u, x0 ) . (3.66)
4 (4π) Γ( 2 ) 2 2 2
z=
(x 0 1 )2
T

To show the equivalence between prescriptions (3.62) (3.63) at the point x in con-
sideration, we must show that
 
01
f (u, 0) = 1 lim 1 f (u, x ) (3.67)
u=0 x0 ∈(0,a1 ), x0 →0 u=0

where |u=0 indicates, as usual, the analytic continuation at u = 0 . In order to check

the equality in Eq. (3.67) we first notice that
u−d−1
κu T 2
   
1 u−d−1
f (u, 0) = δi1 −ξ P u−d−1 (0) =
4 (4π)d/2 Γ( u+1
2
) 2 2

u−d−1
(3.68)
κu T 2
 
1
= δi1 −ξ
4 (4π)d/2 Γ( u+1
2
)

(the second equality relies on Eq. (3.34) for Ps (0)). The last expression in (3.68)
gives the analytic continuation of f (u, 0) to the whole complex plane; in particular,
when evaluated at u = 0, it yields
 
1 2
f (u, 0) = δi1 −ξ d+1 . (3.69)
u=0 4 (4π T ) 2

24
On the other hand, returning to Eq. (3.66) we see that, for any x0 1 ∈ (0, a1 ),
   
01 1 2 d+1
f (u, x ) = δi1 −ξ d+1 − P− d+1 (z)+z P− d+3 (z) . (3.70)
u=0 4 (4π T ) 2 2 2 2
z=(x0 1 )2 /T

Resorting to the second equality in Eq. (3.34) for the functions P• (z) and using a re-
cursive relation for the upper incomplete gamma function (see [29], p.178, Eq.8.8.2),
we obtain   1
01 1 2 (x0 )2
− T
f (u, x ) = δi1 −ξ d+1 e ; (3.71)
u=0 4 (4π T ) 2
comparing with Eq. (3.69), we immediately obtain the desired relation (3.67), that
is f (u, 0)|u=0 = limx0 1 ∈(0,a1 ), x0 1 →0 f (u, x0 1 )|u=0 .

This concludes our analysis of the renormalized pressure at points in the interior of
π1,0 ; needless to say, the very same results also hold for the pressure acting on any
other of the sides delimiting Ω .
3.6.2 Non-integrable behaviour of the renormalized pressure near the
edges. Let us first remark that at points on the edges of the box (i.e., on the
corners which appear whenever d > 1) the outer normal and, consequently, the
pressure are both ill-defined.
In this paragraph we discuss the behaviour of the pressure at points in the neighbor-
hood of the edges; more precisely, we show that the renormalized pressure evaluated
at inner points of one side diverges in a non-integrable manner when moving towards
anyone of the edges.
In order to fix our ideas, let us consider the renormalized pressure pren i on the side
π1,0 ; in the following we discuss the behaviour of this quantity in a neighborhood of
the corner placed at x = 0, i.e., at the intersection of all the sides πp,0 (p ∈ {1, ..., d}).
To this purpose, let us consider the expression (3.62) for pren i ; by considerations
analogous to those of the previous paragraph, we see that contributions diverging
(<)
for x → 0 arise from ∂x1 y1 D1/2 when we use for it the series expansion (3.37). More
precisely, these contributions arise from the terms for which

bhl (0, 0) = 0 and bhl (x, x) 6= 0 for x interior to π1,0 , x 6= 0 ; (3.72)

by a simple inspection, these terms are seen to correspond to the following choices
hi = 0 for all i ∈ {1, ..., d} ,
(3.73)
li ∈ {1, 2} for i ∈ {1, ..., d} and li = 2 for at least one i ∈ {2, ..., d} .

As an example, let us focus on one of the above terms; for any fixed p ∈ {2, ..., d},
we consider the one with

2 for i ∈ {2, ..., p}
hi = 0 for all i ∈ {1, ..., d} and li = . (3.74)
1 for i ∈ {p + 1, ..., d}

25
 (<)
With simple but long computations (9 ), the contribution of this term to ∂x1 y1 D1/2
can be expressed as
2 (−1)d−p
 2 2
(x ) +...+(xp )2

d+1 P− d+1 ≡ gp,l (x) . (3.75)
(4π T ) 2 2 T

Using the second relation in Eq. (3.34) and resorting to the asymptotic expansion
(see [29], p.178, Eq.8.7.3)
Γ(s, z) = Γ(s)(1 + O(z s )) for <s > 0, z → 0 , (3.76)
we readily infer, for x → 0,
2 (−1)d−p Γ( d+1
 
2
) 1 d+1
gp,l (x) = d+1 (1 + O(z )) √ . (3.77)
(4π) 2 z d+1 z= (x2 )2 +...+(xp )2

The above expression shows the non-integrable divergence of gp,l (for all p ∈ {2, ..., d});
the same conclusion holds as well for the terms corresponding to all the other choices
of (h, l) in Eq. (3.73). Let us stress that no cancellation of the divergent terms can
occur, since all the contributions with the same degree of divergence happen to have
the same sign; besides, due to the remainder estimate (3.54), no compensation of
these terms can either arise from the full series expansion (3.37).
The above comments prove the non-integrable behaviour of pren i (x) for x → 0 ; let
us remark that this fact is of utmost importance when attempting to evaluate the
total force acting on any side of the box, a topic we discuss in detail in subsection
3.8.
3.7 The total energy. First of all, let us recall from subsection 2.10 that the total
energy consists of the sum of both a bulk and a boundary contribution; since the
latter vanishes identically due to the Dirichlet conditions assumed on the boundary
(see the comments below Eq. (2.37)), we only have to discuss the bulk term.
Consider the representation (2.36) of the bulk energy (10 ):
κu
Z
u
E := dx D u−1 (x, x) .
2 Ω 2

9
In particular note that, with the present choices (3.73) (3.74), we have
p
X
bhl (x, y) = (xi )2 ,
y=x, x1 =0
i=2

δ l1
∂x1 bhl (x, y) = ∂y1 bhl (x, y) =0, ∂x1 y1 bhl (x, y) =− .
y=x, x1 =0 y=x, x1 =0 y=x, x1 =0 2
10
Let us mention that, in order to evaluate the bulk energy E u , we could proceed in an alternative
1−u
manner. Namely, we could consider the representation of E u in terms of the trace Tr A 2 (see
Eq. (2.36)) and determine the analytic continuation of the latter moving along the same lines we

26
Using the expression (3.24) for the Dirichlet kernel, we readily infer

E u = E u,(>) + E u,(<) where

κu
Z
(•) (3.78)
E u,(•) := dx1 ... dxd D u−1 (x, x) for • ∈ {>, <} .
2 (0,a1 )×...×(0,ad ) 2

In the next two paragraphs we discuss the analytic continuation of E u at u = 0,

giving the renormalized bulk energy. To this purpose we introduce series expansions
for the functions E u,(>) and E u,(<) , which can be used to build their analytic con-
tinuations; we also give fully quantitative remainder estimates for these series (for
more details, see Appendices B and C).
3.7.1 Series expansions and analytic continuations for E u,(>) , E u,(<) . Con-
sider the expression (3.78) for the regularized bulk energy E u ; hereafter we give series
expansions for the two addenda E u,(>) and E u,(<) , ultimately yielding the analytic
continuations of these functions to the whole complex plane.
(>)
On the one hand, inserting the expansion (3.30) for Ds into Eq. (3.78), we obtain
(see Appendix B)
κu
 
u,(>)
X
1−u u−1 2
E = ωn Γ , ωn T . (3.79)
2 Γ( u−1
2
) d
2
n∈N

The right-hand side of above equation can be proved to converge for all u ∈ C
(see the next paragraph and Appendix C); thus, Eq. (3.79) gives the analytic
continuation of E u,(>) to the whole complex plane, in particular at u = 0.
(<)
On the other hand, inserting the expansion (3.36) for Ds into Eq. (3.78), after
some effort we obtain (see again Appendix B)
u−1 d
κu T 2 (−1)d−p X aσ,p X
 
u,(<)
X Bσ,p (h)
E = d+1 u−1 P u−p−1 . (3.80)
2 Γ( 2 ) p=0 (d−p)!p! σ∈S (π T )p/2 h∈Zp 2 T
d

In the above Sd indicates the symmetric group with d elements and we have put
aσ,0 := 1 , Z0 := {0} , Bσ,0 (0) := 0 ,
p p
Y X (3.81)
aσ,p := aσ(i) , Bσ,p (h) := (aσ(i) hi )2 for σ ∈ Sd , p ∈ {1, ..., d} ;
i=1 i=1

note that the term with p = 0 in Eq. (3.80) is just (−1)d P u−1 (0) .
2

followed for the Dirichlet kernel, using the heat trace K(t) in place of the heat kernel K(t ; x, y).
Nonetheless, since the small t expansion of the heat trace K(t) for the present configuration involves
quite cumbersome expressions, we prefer to avoid this approach. Another advantage of the methods
proposed in this subsection is that, after minor variations, they also allow to evaluate the total
force on the boundary (see subsection 3.8).

27
Let us stress that the functions Ps in Eq. (3.80) must be evaluated according to Eq.
(3.33); in particular, recall that the first relation in Eq. (3.34) gives Ps (0) = 1/s .
Thus, for all p ∈ {0, ..., d}, the terms in the series (3.80) with Bσ,p (h) = 0, i.e.,
those with h = 0 (see Eq. (3.81)), are singular at u = p + 1 where they have a
simple pole. The series obtained from the right-hand side of Eq. (3.80) removing
the finitely many terms with h = 0 is (rapidly) convergent for all u ∈ C, a fact that
we discuss in the next paragraph and in Appendix C.
Because of the above considerations, the expression (3.80) gives the analytic contin-
uation of E u,(<) to a meromorphic function on the whole complex plane, with simple
poles at
u ∈ {1, 2, ..., d + 1} . (3.82)
Summing up, u = 0 is a regular point for the analytic continuations of both E u,(>)
and E u,(<) so that, according to the restricted zeta approach (see Eq. (2.38)), we
can put
E ren = E u,(>) + E u,(<) (3.83)
u=0 u=0
where the two addenda on the right-hand side simply indicate the expressions (3.79)
and (3.80) evaluated at u = 0.
3.7.2 Estimates for the series (3.79) (3.80). The scheme followed in this
paragraph closely resembles the one of subsection 3.4, where we dealt with the
(•)
series expansions for the Dirichlet functions Ds ; we retain here the same notations
introduced therein (see, in particular, Eq.s (3.45) (3.46)).
Let us consider the series in the right-hand side of Eq. (3.79), and the series in
the right-hand side of Eq. (3.80) after removing from it the finitely many singular
terms with h = 0. In Appendix C we prove the convergence of these series for all
u ∈ C, and derive the remainder estimates reported hereafter. For N ∈ (0, +∞),
let us write
u,(>) u,(>)
E u,(>) = EN + RN ,
u
 
u,(>) κ X
1−u u−1 2
EN := ωn Γ , ωn T ,
2 Γ( u−1
2
) 2 (3.84)
n∈Nd , |n|6N

κu
 
u,(>)
X
1−u u−1 2
RN := ωn Γ , ωn T ;
2 Γ( u−1
2
) d
2
n∈N , |n|>N

u,>
the remainder RN has the bound (Appendix C)

κ<u max(a<u−1 , A<u−1 ) (d) π 2 T <u−1

 
u,(>)
RN 6 HN α, 2 ; , 1−<u
2d+1 π <u−1 |Γ( u−1
2
)| A 2
r (3.85)
√ √ A (1−<u)
for either <u > 1, N > 2 d or <u < 1, N > 2 d + .
π 2αT

28
Again for N ∈ (0, +∞), let us put
u,(<) u,(<)
E u,(<) = EN + RN , (3.86)
u−1 d
κu T 2 (−1)d−p X aσ,p
 
u,(<)
X X Bσ,p (h)
EN := d+1 u−1 p P u−p−1 ,
2 Γ( 2 ) p=0 (d−p)!p! σ∈S (π T ) 2 2 T
d h∈Zp , |h|6N
u−1 d
κu T (−1)d−p X aσ,p
 
u,(<) 2 X X Bσ,p (h)
RN := d+1 u−1 p P u−p−1
2 Γ( 2 ) p=1 (d−p)!p! σ∈S (π T ) 2 p
2 T
d h∈Z , |h|>N
q
u,(<)
(where |h| := h21 + ... + h2p ); the remainder RN , which contains no singular term
with h = 0, fulfills (see, again, Appendix C)
d
u,(<) κ<u X max(a<u−p−1 , A<u−p−1 )
RN 6 ·
2d+1 |Γ( u−1
2
)| p=1
π p/2 (d−n)!n!

!
a2 p+1−<u
 
(3.87)
X (p)
· aσ,p HN α, ; , <u−p−1
σ∈Sd
T 2
r
√ √ 1 (<u−2)T
for either <u 6 2, N > 2 d or <u > d+1, N > 2 d+ .
a 2α
As sketched in Appendix C, we could give reminder estimates even for the case 2 <
<u 6 d + 1, excluded from (3.87); however, these would involve rather complicated
expressions. Taking into account that, in the sequel, we will be mainly interested in
the case u = 0, we prefer not to report these cumbersome expressions.
In subsections 4.1 and 4.2, moving along the same lines as for the VEV of stress-
u,(>) u,(<)
energy tensor, we will use the truncations EN and EN of a fixed sufficiently
large order N to obtain approximate expressions for the functions E u,(>) and E u,(<) ,
respectively; these will be used to evaluate the renormalized bulk energy (see Eq.
(3.78)), giving explicit errors estimates.
3.8 The total force on a side of the box. Let us consider the framework of
subsection 2.11; following the general scheme outlined therein for boundary forces,
we can in principle consider two alternative approaches to define the total force
acting on any side of the box.
As an example, let us focus on the force acting on π1,0 , i.e., the side contained in
the hyperplane {x1 = 0}; recall that the unit outer normal at points x interior to
π1,0 is n(x) = (−1, 0, ..., 0).
We first consider the regularized integrated force
Z Z
u i u
F1,0 := da(x) n (x) pi (x) = − da(x) pu1 (x) (3.88)
π1,0 π1,0

29
(compare with Eq. (2.39), here employed with O = π1,0 ), where pui (x) indicates the
regularized pressure (3.60). We will prove in the sequel that this can be analytically
continued up to u = 0, so that we can define the renormalized integrated force as

Fren u
1,0 := F1,0 . (3.89)
u=0

On the other hand, we have the alternative prescription (corresponding to Eq.

(2.41), with O = π1,0 )
Z Z
ren i ren
F1,0 := da(x) n (x) pi (x) = − da(x) pren
1 (x) , (3.90)
π1,0 π1,0

where pren
i (x) is the renormalized pressure, defined equivalently according to either
prescription (3.62) or (3.63).
As a matter of fact, we know from the previous subsection that the renormalized
pressure diverges in a non-integrable manner near the edges of the box; in conse-
quence of this, the prescription (3.90) gives an infinite value for the total force on
π1,0 . Since this result is patently physically unacceptable, in the following we only
consider the approach (3.88) (3.89).
Let us therefore consider the regularized expression (3.88); using Eq. (3.61) for the
regularized pressure on π1,0 (we are referring, in particular, to the representation in
the second line of the cited equation), we readily infer
u,(>) u,(<)
Fu1,0 = F1,0 + F1,0 where
u Z
(3.91)
u,(•) κ (•)
F1,0 := dx2 ... dxd ∂x1 y1 D u+1 (x, y) for • ∈ {>, <} .
4 (0,a2 )×...×(0,ad ) 2 y=x, x1 =0

In the next two paragraphs we introduce convergent series expansions for the func-
u,(>) u,(<)
tions F1,0 and F1,0 ; these expansions ultimately yield the analytic continuations
of these functions and of Fu1,0 at u = 0 (and so, they determine the renormalized
force on π1,0 according to Eq. (3.89)). We also give remainder estimates for these
series.
u,(>) u,(<)
3.8.1 Series expansions and analytic continuations for F1,0 , F1,0 . In-
(>)
serting into Eq. (3.91) the series (3.31) for ∂x1 y1 Ds (x, y) and integrating term by
term, we have

u,(>) κu X  n1 π 2 
u+1 2

−(u+1)
F1,0 = ωn Γ , ωn T . (3.92)
2a1 Γ( u+1
2
) d
a1 2
n∈N

The above expression can be proved to converge for all u ∈ C (by a simple variation
of the proof of the convergence of the expansion (3.79) for E u,(>) ; see paragraph C.1

30
 u,(>)
of Appendix C); thus, Eq. (3.92) gives the analytic continuation of F1,0 to the
whole complex plane, in particular at u = 0.
(<)
On the other hand, using the series (3.37) for ∂x1 y1 Ds (x, y) along with the defini-
tion (3.91), we can show with some effort that
u−3 d
u,(<) κu T 2 X (−1)d−p X aσ̄,p
F1,0 = − d+1 u+1 p ·
2 Γ( 2 ) p=1 (d−p)!(p−1)! (π T ) 2
σ̄∈S̄d
(3.93)
X 
Bσ̄,p (h)

T

Bσ̄,p (h)

2
· (a1 h1 ) P u−p−3 − P u−p−1 ;
h∈Zp
2 T 2 2 T

in the above we have put, for brevity,

S̄d := {σ̄ ∈ Sd | σ̄(1) = 1} , aσ̄,1 := 1 , Bσ̄,1 (h) := (a1 h1 )2 , (3.94)

p p
Y X
aσ̄,p := aσ̄(i) , Bσ̄,p (h) := (a1 h1 )2 + (aσ̄(i) hi )2 for σ̄ ∈ S̄d , p ∈ {2, ..., d} .
i=2 i=2

The last result is derived in a manner similar to expansion (3.80) for E u,(<) (see
Appendix B); besides, there hold considerations analogous to those below the cited
equation. More precisely: the terms in the right-hand side of Eq. (3.93) with
p ∈ {1, ..., d} and h = 0 have a simple pole at u = p + 1; after removing these
finitely many terms, the series in the right-hand side of Eq. (3.93) converges for
u,(<)
all u ∈ C. Summing up, Eq. (3.93) gives the analytic continuation of F1,0 to a
meromorphic function on the whole complex plane, with simple poles at

u ∈ {2, 3, ..., d, d + 1} . (3.95)

u,(>)
Summing up, u = 0 is a regular point for the analytic continuations of both F1,0
u,(<)
and F1,0 ; now, recalling the definition (3.89) for the renormalized total force on
π1,0 and comparing with Eq. (3.91), we get
u,(>) u,(<)
Fren
1,0 = F1,0 + F1,0 (3.96)
u=0 u=0

where the two addenda on the right-hand side simply indicate the expressions (3.92)
and (3.93) evaluated at u = 0.
3.8.2 Estimates for the series (3.92) (3.93). The arguments of paragraph
3.7.2 and Appendix C for the expansions (3.79) (3.80) of E u,(>) and E u,(<) can also
u,(>)
be adapted to deduce remainder estimates on the series (3.92) (3.93) for F1,0 and
u,(<)
F1,0 . Here we only report the final results, concerning truncation at a suitable
order N ; in our presentation we adopt the notations introduced in subsection 3.4
(see, in particular, Eq.s (3.45) (3.46)).

31
For N ∈ (0, +∞), let us write
u,(>) u,(>) u,(>)
F1,0 = F1,0,N + R1,0,N ,
u,(>) κu X  n1 π 2 
u+1 2

−(u+1)
F1,0,N := ωn Γ , ωn T ,
2a1 Γ( u+1
2
) a1 2 (3.97)
n∈Nd , |n|6N

u,(>) κu X  n1 π 2 
u+1 2

−(u+1)
R1,0,N := ωn Γ , ωn T ;
2a1 Γ( u+1
2
) d
a1 2
n∈N , |n|>N

u,(>)
the remainder R1,0,N has the bound

κ<u π 1−<u max(a<u+1, A<u+1 ) (d) π 2 T <u+1

 
u,(>)
R1,0,N 6 HN α, 2 ; , 1−<u
2d+1 a31 |Γ( u+1
2
)| A 2
r (3.98)
√ √ A (1−<u)
for either <u > 1, N > 2 d or <u < 1, N > 2 d + .
π 2αT
Moreover, let
u,(<) u,(<) u,(<)
F1,0 = F1,0,N + R1,0,N ,
u−3 d
u,(<) κu T 2 X (−1)d−p X aσ̄,p
F1,0,N := − d+1 u+1 p ·
2 Γ( 2 ) p=1 (d−p)!(p−1)! (π T ) 2
σ̄∈S̄d
    
X
2 Bσ̄,p (h) T Bσ̄,p (h)
· (a1 h1 ) P u−p−3 − P u−p−1 ,
p
2 T 2 2 T (3.99)
h∈Z , |h|6N
u−3 d
u,(<) κu T 2 X (−1)d−p X aσ̄,p
R1,0,N := − u+1 p ·
2d+1 Γ( 2 ) p=1 (d−p)!(p−1)! (π T ) 2
σ̄∈S̄d
    
X
2 Bσ̄,p (h) T Bσ̄,p (h)
· (a1 h1 ) P u−p−3 − P u−p−1
p
2 T 2 2 T
h∈Z , |h|>N

q
u,(<)
(recall that |h| := h21 + ... + h2p ); as for the remainder R1,0,(N ) (containing no
singular term with h = 0), there holds the estimate

32
 d
u,(<) κ<u X 1 X aσ̄,p
6 d+1
R1,0,N ·
2 |Γ( u+1 2
)| p=1 (d−p)!(p−1)! π p/2
σ̄∈S̄d
2
  
2 <u−p−3 <u−p−3 (p) a p+3−<u (3.100)
· a1 max(a ,A ) HN α, ; , <u−p−1 +
T 2
a2 p+1−<u
 
1 <u−p−1 <u−p−1 (p)
+ max(a ,A ) HN α, ; , <u−p−1
2 T 2
r
√ √ 1 (<u−2)T
for either <u 6 2, N > 2 d or <u > d+1, N > 2 d + .
a 2α
Concerning the case 2 < <u 6 d + 1, not taken into account in Eq. (3.100),
there hold considerations analogous to the ones below Eq. (3.87); also in this case
the corresponding reminder estimates would involve rather cumbersome expressions
which we choose not to discuss here in view of the fact that we will be interested
only in the case u = 0.
The evaluation of the force on π1,0 presented in the subsequent subsections 4.1 4.2
for d = 1, 2 will be based on the truncated expansions and on the related remainder
bounds discussed in this paragraph.
3.9 Scaling considerations. From Eq.s (3.55-3.58) and from the expressions for
the Dirichlet functions given in subsection 3.3, we easily infer the following relation
for each component of the stress-energy VEV (µ, ν ∈ {0, ..., d}):
u
h0|Tbµν (x)|0i = a1u−d−1 Tuµν (x? ; ρ) , (3.101)
where Tuµν is a suitable function and x? , ρ are, respectively, the d-tuple and the
(d − 1)-tuple with compoments
xi ai
xi? := ∈ (0, 1) for i ∈ {1, ..., d} , ρi := for i ∈ {2, ..., d} . (3.102)
ai a1
For d = 1, the variables ρi are not defined and Tuµν only depends on x1? = x1 /a1 (11 ).
Similarly, for the regularized pressure acting on any point x in the interior of the
side π1,0 , we deduce from Eq.s (3.60) (3.101) that
pui (x) = au−d−1
1 pui (x? ; ρ) for i ∈ {1, ..., d} (3.103)
11
It is apparent from Eq.s (3.8) (3.11) that the eigenfunctions Fk (x) := Fk1 (x1 )...Fkd (xd ) and
the corresponding eigenvalues ωk2 can be written in the form
−d ni π
Fk (x) = a12
ϕn,ρ (x? ) , ωk2 = a−2 2
1 λn,ρ for ki = (i ∈ {1, ..., d}) ,
ai
for some suitable functions ϕn,ρ , and some coefficients λn,ρ . Using the eigenfunction expansion
for the Dirichlet kernel (see Eq. (3.19) in Part I), we obtain
−(d−2)s
Ds (x, y) = a1 Ds,ρ (x? , y? ) ,

33
where pui are suitable functions and x? is defined as in Eq. (3.102) at points on the
boundary. Clearly, the same conclusions can be drawn for the pressure on any other
side πp,λ (p ∈ {1, ..., d}, λ ∈ {0, 1}).
Analogous considerations hold for the total energy and integrated force on the
boundary of the spatial domain. On the one hand, concerning the bulk energy,
from the expansions derived in subsection 3.7 we easily infer (indicating with Eu a
suitable function)
E u = au−1
1 Eu (ρ) . (3.104)
On the other hand, as for the total force on π1,0 (for example), from Eq.s (3.92)
(3.93) it follows that
Fu1,0 = au−2
1 Fu1,0 (ρ) (3.105)
for some suitable function Fu1,0 . Again, similar results hold for the total force on any
other side πp,λ .
By analytic continuation at u = 0, we obtain the renormalized counterparts of the
above relations: more precisely, we have
−(d+1) −(d+1)
h0|Tbµν (x)|0iren = a1 Tµν (x? ; ρ) , pren
i (x) = a1 pi (x? ; ρ) ,
(3.106)
E = a−1
ren
1 E(ρ) ,
ren −2
F1,0 = a1 F1,0 (ρ)

(where the right-hand sides of the above relations are obtained evaluating at u = 0
the functions in the right-hand sides of Eq.s (3.101-3.105).
Due to the remarks of this subsection, for any spatial dimension d the analysis of the
renormalized stress-energy VEV, total energy, pressure and of the integrated force
can always be reduced to the case a1 = 1; we will use this fact in the next section
on the cases d = 1 and d = 2.

4 The previous results in spatial dimension d ∈ {1, 2}

4.1 Case d = 1 (the segment). As a first application of our framework, let us
consider the case
d=1, Ω = (0, a1 ) (a1 > 0) . (4.1)
As a matter of fact, we already analysed this configuration in Section 6 of Part I;
the present Eq. (4.1) corresponds to Eq. (6.1) of Part I (with a = a1 ).
for some suitable function Ds,ρ ; from the above relation we can easily infer that, for any pair z, w
of spatial variables,
−(d−2)s−2
∂zw Ds (x, y) = a1 ∂z? w? Ds,ρ (x? , y? ) .
Using these results, one can easily deduce Eq. (3.101) and the subsequent statements of this
subsection.

34
In Part I we performed the exact computation for the renormalized VEV of the
stress-energy tensor and of the pressure for various types of boundary conditions.
Here we carry out an approximate evaluation of h0|Tbµν (x)|0iren , pren
1 (x)|x1 =0 and of
E ren for Dirichlet boundary conditions, truncating the series expansions for these
quantities derived in the present work for a box in arbitrary spatial dimension. Our
aim is just to check the validity of the general methods developed here; to this
purpose, we compare the results obtained for the renormalized stress-energy VEV,
pressure and total energy, respectively, with those reported in Eq.s (6.24), (6.26)
and (6.27) of Part I.
In our computations we only consider the case with
a1 = 1 , (4.2)
which cause no loss of generality due to the scaling considerations discussed in
subsection 3.9 (of course, due to Eq. (4.2), the rescaled variable x1? := x1 /a1 in fact
coincides with x1 ) (12 ).
Before proceeding to the evaluation of the renormalized VEVs h0|Tbµν |0iren and pren 1 ,
let us recall the representation (3.56-3.58) of the regularized stress-energy VEV in
terms of the (>) and (<) parts of the Dirichlet kernel; these parts depend on the
(>)
choice of a parameter T > 0, which however has no effect on the sum Ds = Ds +
(<) (>) (<)
Ds . After fixing T , we can approximate Ds and Ds truncating their series
expansions at some sufficiently large order N , giving estimates on the remainders as
well. Analogous considerations hold for the spatial integral of the diagonal Dirichlet
kernel, ultimately giving the renormalized bulk energy E ren (see Eq.s (2.36) (3.78)
and (3.83)).
Here we choose
T =1, N =5; (4.3)
the truncated sums are evaluated numerically, and the remainder estimates (3.49-
3.54) and (3.85) (3.87) are also taken into account, fixing (13 )
α = 0.03 . (4.4)
(>) (<)
Of course, the bounds obtained for the remainders associated to Ds and Ds
allow us, in turn, to infer error estimates for the approximate expressions of both
h0|Tbµν |0iren and pren
1 .

Let us first consider the renormalized stress-energy VEV h0|Tbµν |0iren ; according to
the analysis of subsection 3.5, this is obtained setting u = 0 in Eq.s (3.55-3.58).
12
Let us stress that, for a1 = 1, the quantities h0|Tbµν |0iren and pren
1 are, respectively, equal to
the rescaled functions Tren
µν and p ren
i (see Eq. (3.106)).
13
We make the choice (4.4) for α because it is close to the value minimizing the error esti-
mates cited above; a more precise evaluation of this optimal value for α would require a laborious
numerical analysis which we prefer to avoid here.

35
In reporting our results, we distinguish between the conformal and non-conformal
(♦)
parts of each component, which are respectively denoted as usual with h0|Tbµν |0iren
()
and h0|Tbµν |0iren (see Eq. (2.12)); notice that (since d = 1) Eq. (2.13) gives

ξ1 = 0 . (4.5)

The graphs of the conformal and noncomformal parts of each stress-energy com-
ponent are shown in Fig.s 1 and 2 (recall that we refer to the rescaled variable
x1? := x1 /a1 ≡ x1 ). Let us comment briefly on the above graphs. Apart from

(♦) ()
Figure 1: d = 1: graphs of h0|Tb00 |0iren and h0|Tb00 |0iren .

(♦) ()
Figure 2: d = 1: graphs of h0|Tb11 |0iren and h0|Tb11 |0iren .

()
h0|Tb00 |0iren , it appears that all the other components of the renormalized stress-
()
energy VEV are constants and h0|Tb11 |0iren is very small; indeed, our computations
with N = 5 (T = 1, α = 0.03) ensure, for all x ≡ x1? ∈ (0, 1),
(♦) (♦)
h0|Tb00 (x)|0iren = h0|Tb11 (x)|0iren = − 0.1308997 ± 8 · 10−7 ,
()
(4.6)
|h0|Tb11 (x)|0iren | 6 4 · 10−6 .

36
 ()
Concerning h0|Tb00 (x)|0iren we have, for example,

3.141593 ± 10−6 for x1? = 1/4


()
h0|T00 (x)|0iren =
b . (4.7)
1.570796 ± 10−6 for x1? = 1/2

These results are in agreement with the exact calculations of subsection 6.6 of Part
I, which gave the following outcomes (see Eq. (6.24) of the cited subsection with
a ≡ a1 = 1) (14 )
(♦) (♦) π
h0|Tb00 (x)|0iren = h0|Tb11 (x)|0iren = − ,
24
() π ()
(4.8)
h0|Tb00 (x)|0iren = 2 1
, h0|Tb11 (x)|0iren = 0 .
2 sin (πx )

Next, let us pass to the evaluation of

Fren (0) := pren

1 (x) ; (4.9)
x1 =0

this is nominally the “pressure” on the boundary point x1 = 0 but in fact coincides
with the force on this point, due to the zero dimensionality of the boundary.
Let us consider the prescriptions (3.62); computing the derivative of the functions
(>) (<)
D1/2 , D1/2 appearing therein with the choices a1 = 1, T = 1 and N = 5, we obtain

Fren (0) = 0.1308997 ± 3 · 10−7 (4.10)

(again, the error is obtained using the remainder estimates of subsection 3.8.2 with
π
α = 0.03). The above result is in agreement with the exact expression Fren (0) = 24
derived in subsection 6.6 of Part I (see Eq.s (6.24) (6.27), and set a ≡ a1 = 1
therein).
Finally, we consider the renormalized bulk energy E ren ; the series expansions (3.79)
(3.80) derived in subsection 3.7 (with the previous choices of a1 , T, N ) allow us to
infer
E ren = −0.1308996938996 ± 3 · 10−13 (4.11)
(where the error is obtained using the remainder estimates of subsection 3.7.2, again
π
with α = 0.03). The result (4.11) agrees with the exact computation E ren = − 24
obtained in subsection 6.6 of Part I (see Eq. (6.26), and set a ≡ a1 = 1 therein).
14
To make a comparison with Eq.s (4.6) (4.7), note that

3.14159265... for x1 = 1/4


π π
− = −0.13089969... , = .
24 2 sin2 (πx1 ) 1.57079633... for x1 = 1/2

37
Let us stress that our approximants by truncation, converge quite rapidly to the
exact results; in order to exemplify this statement, we notice that, by slightly in-
creasing the value of the truncation order N , we obtain a remarkable improvement
of the error estimates. For example (using again the estimates of subsection 3.7.2
with T = 1 and α = 0.03) the error ±3 · 10−13 in Eq. (4.11) (for N = 5) becomes
±2 · 10−46 for N = 10, 6 · 10−101 for N = 15 and 4 · 10−177 for N = 20.
4.2 Case d = 2. Let us now pass to the 2-dimensional case:
d=2, Ω = (0, a1 ) × (0, a2 ) (a1 , a2 > 0) . (4.12)
As in the previous subsection, we fix
a1 = 1 (4.13)
and consider different values of a2 ; let us repeat that the above choice does not
imply a loss of generality, due to the scaling properties of subsection 3.9. Moreover,
we present the final results in terms of the rescaled coordinates x1? := x1 /a1 ≡ x1 ,
x2? := x2 /a2 ∈ (0, 1), defined in Eq. (3.102) (15 ).
Again, the basic elements to compute the renormalized stress-energy VEV and the
(>) (<)
pressure are the Dirichlet functions Ds , Ds , along with their spatial derivatives,
for which we use the truncated expansions (3.49-3.54) and the remainder bounds of
Eq.s (3.49-3.54).
Needless to say, analogous considerations also hold for the renormalized bulk energy
and for the integrated boundary forces (see subsections 3.7 and 3.8, respectively).
Let us first consider the stress-energy VEV and the pressure; as examples, we com-
pute these observables for the two configurations with
a2 = 1 and a2 = 5 . (4.14)
In these cases, for the parameter T of the decomposition into (>) and (<) parts and
for the truncation order N , we make the following choices:
T =1, N =7 for a2 = 1 ;
(4.15)
T =1, N =9 for a2 = 5 .
The truncation errors in Eq.s (3.49-3.54) are evaluated making for the parameter α
therein the choice (16 )
α = 0.04 . (4.16)
15
Similarly to what we said in the footnote 12 on page 35, for a1 = 1, the quantities h0|Tbµν |0iren ,
pren
1 and Fren
1,0 (to be discussed hereafter) do in fact coincide with the rescaled analogues Tren µν ,
pi , Fren
ren
1,0 introduced in Eq.s (3.101) (3.103) (3.105)). Besides, the lenght a2 of the second side is
identified with the ratio ρ2 (see Eq. (3.102)).
16
This choice can be justified by considerations similar to the ones in the footnote 13 of page 35.

38
The renormalized stress-energy VEV h0|Tbµν |0iren is obtained setting u = 0 in Eq.s
(3.55-3.58). Again, we separate the conformal and nonconformal parts, respectively
indicated by the superscripts (♦) and (); recall that Eq. (2.13) gives, in the two-
dimensional case,
ξ2 = 1/8 . (4.17)
(♦) ()
In the following we present the graphs for h0|Tbµν |0iren and h0|Tbµν |0iren obtained
from the previous truncated expansions; more precisely, Fig.s 3-5 and Fig.s 6-9 show,
respectively, the results obtained for the configurations with a2 = 1 and a2 = 5 .
In the cited figures we refer to the variables xi? := xi /ai ∈ (0, 1) and, keeping into
account some obvious symmetry considerations (17 ), we only show the graphs for

xi? ∈ (0, 1/2) for i ∈ {1, 2} ; (4.18)

moreover, in the case of a square box with a1 = a2 = 1 we do not report the graphs
for the conformal and non-conformal parts of h0|Tb22 (x)|0iren , since these are equal
to the corresponding parts of h0|Tb11 (x)|0iren .

(♦) ()
Figure 3: d = 2: graphs of h0|Tb00 (x)|0iren and h0|Tb00 (x)|0iren for a2 = 1 .

17
Indeed, every component of the stress-energy VEV can be shown to be symmetric under the
exchange xi ↔ ai − xi (or xi? ↔ 1 − xi? ) for i ∈ {1, 2} .

39
 (♦) ()
Figure 4: d = 2: graphs of h0|Tb11 (x)|0iren and h0|Tb11 (x)|0iren for a2 = 1 .

(♦) ()
Figure 5: d = 2: graphs of h0|Tb12 (x)|0iren and h0|Tb12 (x)|0iren for a2 = 1 .

(♦) ()
Figure 6: d = 2: graphs of h0|Tb00 (x)|0iren and h0|Tb00 (x)|0iren for a2 = 5 .

40
 (♦) ()
Figure 7: d = 2: graphs of h0|Tb11 (x)|0iren and h0|Tb11 (x)|0iren for a2 = 5 .

(♦) ()
Figure 8: d = 2: graphs of h0|Tb12 (x)|0iren and h0|Tb12 (x)|0iren for a2 = 5 .

(♦) ()
Figure 9: d = 2: graphs of h0|Tb22 (x)|0iren and h0|Tb22 (x)|0iren for a2 = 5 .

41
Concerning the error estimates, for µ, ν ∈ {0, 1, 2} and • ∈ {♦, }, let us introduce
the following notation:

(•) remainder corresponding to our approximation

Eµν := (•) (4.19)
by truncation of h0|Tbµν |0iren .

For a2 = 1, our choice N = 7 (T = 1, α = 0.04) yields the uniform bounds

(♦) ()
|E00 | 6 2 · 10−12 , |E00 | 6 2 · 10−11 ;
(♦) ()
(4.20)
|Eij | 6 6 · 10−12 , |Eij | 6 3 · 10−11 for i, j ∈ {1, 2} .

For a2 = 5, our choice N = 9 (T = 1, α = 0.04) ensures

(♦) ()
|E00 | 6 5 · 10−12 , |E00 | 6 4 · 10−11 ;
(♦) ()
(4.21)
|Eij | 6 2 · 10−11 , |Eij | 6 8 · 10−11 for i, j ∈ {1, 2} .

Now, let us evaluate the pressure preni (x) at points x of one side; as in the construc-
tion of the general theory we consider, as an example, the points x ≡ (0, x2 ) in the
interior of the side π1,0 , making reference to the prescription (3.62).
Fig. 10 shows the graphs obtained for pren 2 2
1 (x) as a function of x? := x /a2 (again,
choosing T = 1 and truncating the related expansions to order N = 7, for a2 = 1,
and N = 9, for a2 = 5).

Figure 10: d = 2: graphs of pren

1 for a2 = 1 (left) and a2 = 5 (right).

As for the error, indicating with 1 the remainder associated to our approximation
by truncation of pren
1 , we obtain the following uniform estimates (setting α = 0.04):

|1 | 6 2 · 10−12 for a2 = 1 ;

(4.22)
|1 | 6 5 · 10−12 for a2 = 5 .

42
Before moving on, let us briefly comment on the behaviour of the renormalized
pressure pren
1 near the edge x = 0. Indeed, specializing to the present d = 2 case
the considerations of paragraph 3.6.2 one can prove that, for all a2 > 0,
1
pren
1 (x) = + O((x2 )2 ) for x = (0, x2 ) and x2 → 0+ . (4.23)
32π(x2 )3
Now, let us pass to the computation of the bulk energy and of the integrated force.
For each one of these two observables we consider several configurations, corre-
sponding to different values of a2 ; for any one of these values, we consider the
decomposition into (>) and (<) parts, and choose the truncation order N of the
related expansions so as to obtain error estimates all of approximatively the same
order of magnitude, fixing again
α = 0.04 . (4.24)
In order to obtain the renormalized bulk energy E ren , we first consider its reg-
ularized version (3.78), along with the series expansions (3.79) (3.80), giving the
analytic continuations of the functions E u,(>) and E u,(<) , respectively. Due to the
considerations of subsection 3.7, we can simply put u = 0 in these expansions, since
no singularity arises there; next, we truncate the corresponding series at a suitable
order N (see the comments above) and use the remainder estimates (3.85) (3.87).
In this way we obtain, for example, the following results: for a1 = 1 (and with the
choice T = 1, as usual)
E ren = − 1.73691776 ± 4 · 10−8 for a2 = 0.1 (N = 40) ;
E ren = + 0.03524178 ± 3 · 10−8 for a2 = 0.5 (N = 8) ;
E ren = + 0.04104060 ± 9 · 10−8 for a2 =1 (N = 4) ; (4.25)
E ren = − 0.05412096 ± 2 · 10−8 for a2 =5 (N = 7) ;
E ren = − 0.17369178 ± 1 · 10−8 for a2 = 10 (N = 15) .
Again for a1 = 1, one can plot the renormalized bulk energy E ren as a function
of a2 ; the graph in Fig. 11 has been obtained using the truncation order N = 50
(with T = 1; with the choices made for the parameters N, T, a1 , we know that the
remainder is smaller than 2 · 10−3 for a2 ∈ [0.05, 10)).
Let us discuss some facts regarding the function a2 7→ E ren (a2 ), which can be read
from the graph in Fig. 11 (the results reported hereafter are obtained using standard
numerical methods, implemented in Mathematica) (18 ).
18
The derivation of the results in items i)-iv), including the error estimates, should be accounted
for; as an example, let us give some details on Eq. (4.26). To obtain this result we have used for
E ren an approximant by truncation at order N = 50 that, according to the reminder estimates
(3.85) and (3.87), gives E ren up to an error 6 10−272 for a1 = 1 and 0.5 6 a2 6 1. This large order
approximant of E ren has been maximized numerically with respect to a2 via Mathematica, asking
for a precision of order 10−15 on the maximum point. The final results have been prudentially
truncated to 8 digits, yielding Eq. (4.26); in view of the previous considerations, they are very
likely even though not certified.

43
 Figure 11: d = 2: graph of E ren as function of a2 .

i) There is only one point of maximum amax

2 ; our approximation by truncation at
order N = 50 gives

amax
2 = 0.72719110 ± 10−8 ,
(4.26)
E ren (amax
2 ) = 0.04472675 ± 10−8 .
(1) (2)
ii) E ren vanishes for two values ā2 < ā2 of a2 ; these are found to be
(1)
ā2 = 0.36538151 ± 10−8 ,
(2)
(4.27)
ā2 = 2.73686534 ± 10−8 .
(1) (2)
E ren is positive for ā2 < a2 < ā2 and negative elsewhere. This feature was also
(2) (1)
pointed out in [22]; therein it is stated that ā2 = (ā2 )−1 , a relation (approximately)
verified by the numerical values in Eq. (4.27).
iii) For a2 → 0+ , E ren has the asymptotic behaviour
e0  
E ren (a2 ) = 1 + O(a 2 ) with e0 = −0.02391 ± 10−5 . (4.28)
(a2 )2

iv) There are indications that E ren approaches an asymptote for a2 → +∞; taking
into account values of the abscissa up to a2 = 100, we find that this asymptote is
the straight line

mE = − 0.02391416 ± 10−8

y = mE a2 + qE with . (4.29)
qE = + 0.06544985 ± 10−8

Finally, let discuss the renormalized total force Fren

1,0 acting on the side π1,0 ; following
the analysis of subsection 3.8, we consider the expression (3.96) and represent the
u,(>) u,(<)
functions F1,0 , F1,0 appearing therein using the series expansions (3.92) (3.93).
Fren
1,0 is obtained setting u = 0 in these series expansions; for the actual calculation,

44
these can be truncated at a fixed, sufficiently large order N using the remainder
estimates (3.98) (3.100).
In this way we obtain, for example, the following results: for a1 = 1 (and with the
choice T = 1, as usual)
Fren
1,0 = + 2.3914162 ± 4 · 10−7 for a2 = 0.1 (N = 50) ;
Fren
1,0 = + 0.0956400 ± 2 · 10−7 for a2 = 0.5 (N = 9) ;
Fren
1,0 = + 0.020520 ± 6 · 10−6 for a2 =1 (N = 4) ; (4.30)
Fren
1,0 = − 0.173692 ± 4 · 10−6 for a2 =5 (N = 7) ;
Fren
1,0 = − 0.412833 ± 3 · 10−6 for a2 = 10 (N = 15) .
Again for a1 = 1, one can plot the renormalized force Fren
1,0 as a function of a2 ; the
graph in Fig. 12 has been obtained using the truncation order N = 60 (with T = 1;
with the choices made for the parameters N, T, a1 , we know that the remainder is
smaller than 2 · 10−3 for a2 ∈ [0.07, 10)).

Figure 12: d = 2: graph of Fren

1,0 as function of a2 .

In the following we briefly discuss a number of facts concerning the function a2 7→

Fren
1,0 (a2 ), which appear in the above graph.
i) The function under analysis is strictly decreasing for a2 ∈ (0, +∞) .
ii) There is only one value of a2 , which we indicate with ā2 , where Fren
1,0 vanishes;
our approximation by truncation at order N = 60 allows us to infer
ā2 = 1.3751543 ± 10−7 . (4.31)
iii) For a2 → 0+ , Fren 1,0 has the asymptotic behaviour

f0  
Fren
1,0 (a 2 ) = 1 + O(a 2 ) with f0 = 0.02391 ± 10−5 . (4.32)
(a2 )2
iv) It appears that Fren
1,0 approaches an asymptote for a2 → +∞; considering again
values of the abscissa up to a2 = 100, the equation of this asymptote is found to be
m = − 0.0478283 ± 10−7

y = m a2 + q with . (4.33)
q = + 0.0654498 ± 10−7

45
Let us mention that, in agreement with the general considerations of subsection 4.4
in Part I, the above results for the renormalized total force on the side π1,0 of the
box could be equivalently derived by differentiating the renormalized total energy
E ren with respect to the lenght a2 of the edge of the box perpendicular to π1,0 .
To conclude, let us compare the previous results about E ren , Fren1,0 with the calcula-
tions of Bordag et al. [5]. The authors of [5] derive (both by Abel-Plana formula
and by zeta regularization) series expansions different from ours for the renormalized
bulk energy and for the force on one side; besides, they give no remainder estimates
for these expansions (19 ).
The numerical values of E ren , Fren
1,0 given by our previous analysis are in good agree-
ment with those arising from the expansions in [5], a fact strongly indicating the
equivalence between our approach and [5]. Let us also mention that the results of
[5] about E ren or Fren
1,0 are equivalent to the ones of [3, 11, 16].

Acknowledgments. This work was partly supported by INdAM, INFN and by

MIUR, PRIN 2010 Research Project “Geometric and analytic theory of Hamiltonian
systems in finite and infinite dimensions”.

19
In fact, a not so trivial analysis (which we do not report here for brevity) allows to conclude
that the series expansions given in [5] also converge with exponential speed.

46
A Appendix. Absolute convergence and remain-
der bounds for the series of subsections 3.3.
Let us consider the general framework of the subsections mentioned in the title.
(>)
Note that the representation (3.30) for the function Ds (x, y) contains the series
X
ωn−2s Γ(s, ωn2 T ) Cn (x, y) . (A.1)
n∈Nd

(<)
On the other hand, in the expansion (3.36) for Ds (x, y) there appears a series
over h ∈ Zd and l ∈ {1, 2}d , that we have reexpressed in the equivalent form (3.39).
We pointed out in paragraph 3.3.2 that the terms in this series with bhl (x, y) = 0
become singular at s = d/2 ; these have either |h| = 0 or |h| = 1 (see Eq. (3.22))
and, after removing all terms with |h| = 0, 1 we are left with the series
 
X bhl (x, y)
δl Ps− d . (A.2)
d
2 T
h∈Z , |h|>1
l∈{1,2}d

Note that the absolute convergence of the series (A.2) is equivalent to the absolute
convergence of the series in the second line of Eq. (3.39).
In the next subsections we prove the convergence of the series (A.1) (A.2) for all
s ∈ C , deriving as well remainder estimates for both of them; the statements in Eq.s
(>) (<)
(3.30) (3.50) (related to Ds (x, y)) and in Eq.s (3.36) (3.53) (related to Ds (x, y))
follow easily from these results.
The arguments presented in this appendix can be easily generalized to derive the
analogous conclusions for the series in Eq.s (3.31) (3.51) and (3.37) (3.54) related to
(>) (<)
the derivatives ∂zw Ds (x, y) and ∂zw Ds (x, y); we will briefly return on this topic
in the sequel (see subsections A.2 and A.3).
In the final subsection of this appendix we also derive the asymptotic (3.47) for the
(d)
function HN in our remainder estimates.
Before proceeding, let us recall the following definitions (see Eq. (3.45)):
d
X
a := min {ai } , A := max {ai } , |z|2 := zi2 for z = n ∈ Nd or z = h ∈ Zd .
i∈{1,...,d} i∈{1,...,d}
i=1

A.1 Preliminary estimates. First of all, let us consider the upper incomplete
gamma function (see Eq. (3.29))
Z +∞
Γ(s, z) := dw e−w ws−1 (s ∈ C, z ∈ (0, +∞))
z

47
and note that the above definition implies the following:
|Γ(s, z2 )| 6 Γ(<s, z1 ) for all s ∈ C, z1 , z2 ∈ (0, +∞) with 0 < z1 6 z2 ; (A.3)
|Γ(s, z)| 6 (1−α)−<s e−αz Γ(<s, (1−α)z) for all s ∈ C, z ∈ (0, +∞), α ∈ (0, 1) . (A.4)
Hereafter we are going to prove that
X (d)
|h|ρ |Γ(s, β|h|2 )| 6 HN (α, β; <s, ρ)
h∈Zd , |h|>N

for all s ∈ C, N, ρ ∈ R, β ∈ (0, +∞), α ∈ (0, 1) such r

that (A.5)
√ √ ρ
either ρ 6 0 and N > 2 d or ρ > 0 and N > 2 d + ,
2αβ
(d)
where HN is defined according to Eq. (3.46), i.e. (for σ ∈ R),
(d)
HN (α, β; σ, ρ) :=
√ !d−1
π d/2 √ 2
 
N− d 2 d+ρ
d+ρ
√ Γ(σ, (1−α)βN ) Γ ; α β(N −2 d) .
(1−α)σ (α β) 2 Γ( d2 ) N −2 d 2
√
In the sequel we always assume s ∈ C, ρ ∈ R, N ∈ (2 d, +∞), β ∈ (0, +∞), α ∈ (0, 1).
In order to derive the estimate (A.5) we first point out that, due to Eq.s (A.3) (A.4),
there holds
2
|Γ(s, β|h|2 )| 6 (1−α)−<s e−αβ|h| Γ(<s, (1−α)βN 2 )
(A.6)
for all h ∈ Zd with |h| > N ;
this in turn implies
2
X X
|h|ρ |Γ(s, β|h|2 )| 6 (1−α)−<s Γ(<s, (1−α)βN 2 ) |h|ρ e−αβ|h| . (A.7)
h∈Zd , |h|>N h∈Zd , |h|>N
√
To proceed, let us note that, for any non-increasing function F : [N −2 d, +∞) →
[0, +∞), there holds the following estimate (see, e.g., [28], p. 691):
X 2π d/2 +∞
Z √ d−1
F (|h|) 6 d √
dτ (τ + d) F (τ ) ; (A.8)
d
Γ( 2
) N −2 d
h∈Z , |h|>N

√  √   √  √
noting that τ + d = 1 + τd τ 6 N−2 N− √d
d
τ (for τ > N − 2 d > 0), the above
inequality yields
√ !d−1 Z +∞
X 2π d/2 N − d
F (|h|) 6 d
√ √
dτ τ d−1 F (τ ) . (A.9)
d
Γ( 2
) N −2 d N −2 d
h∈Z , |h|>N

48
We want to employ the above bound to estimate the sum in the right-hand side of
Eq. (A.7); this case involves the function
2
F (τ ) = τ ρ e−αβ τ , (A.10)
√
which is decreasing on [N −2 d, +∞) for
√ √ r
ρ
either ρ 6 0 and N > 2 d (as before) or ρ > 0 and N > 2 d + . (A.11)
2αβ

In each one√of the above two situations, making the change of variable v := α β τ 2 ∈
[α β(N − 2 d)2 , +∞) and recalling the definition (3.29) of the upper incomplete
gamma function, we easily obtain
Z +∞ √ 2
 
d+ρ−1 −αβ τ 2 1 d+ρ
√
dτ τ e = d+ρ Γ ; α β(N −2 d) , (A.12)
N −2 d 2(α β) 2 2

which, along with Eq.s (A.7) (A.9) and (A.10), proves Eq. (A.5).
In passing we point out that Eq. (A.5) implies the analogous estimate (20 )
X 1 (d)
|n|ρ |Γ(s, β|n|2 )| 6 H (α, β; <s, ρ)
2d N
n∈Nd , |n|>N

for all s ∈ C, N, ρ ∈ R, β ∈ (0, +∞), α ∈ (0, 1) such r

that (A.13)
√ √ ρ
either ρ 6 0 and N > 2 d or ρ > 0 and N > 2 d + .
2αβ
A.2 The series (A.1); connections with the expansions (3.30) (3.31) for
Ds(>) and its derivatives. For s ∈ C and N ∈ (0, +∞), let us define
(>)
X
Rs,N (x, y) := ωn−2s Γ(s, ωn2 T ) Cn (x, y) ; (A.14)
n∈Nd , |n|>N

(>)
of course, the series (A.1) is absolutely convergent if and only if Rs,N (x, y) < +∞
for some N . The rest of this subsection will be mostly dedicated to evaluating
(>)
Rs,N (x, y) .
First of all, note that from the definitions (3.11) of Cn , ωn2 and (3.45) of a, A it easily
follows, for all n ∈ Nd ,
|Cn (x, y)| 6 1 , (A.15)
20
Indeed, it sufficies to observe that for any function F : (0, +∞) → [0, +∞) there holds
X X
F (|h|) > 2d F (|n|) .
h∈Zd , |h|>N n∈Nd , |n|>N

49
 π2 2 2 π2 2
|n| 6 ωn 6 |n| . (A.16)
A2 a2
Using Eq. (A.16) we infer, for all s ∈ C, n ∈ Nd ,

max(a2<s , A2<s ) π2 T
 
|ωn−2s | 6 2<s
|n|−2<s , |Γ(s, ωn2 T )| 6 Γ <s, 2 |n|2
(A.17)
π A

(to deduce the second inequality we also employ Eq. (A.3)). Returning to the
definition (A.14), we obtain

max(a2<s , A2<s ) π2 T
X  
(>) −2<s 2
Rs,N (x, y) 6 |n| Γ <s, 2 |n| , (A.18)
π 2<s d
A
n∈N , |n|>N

and using the estimate (A.13) we conclude the following, for any α ∈ (0, 1):

max(a2<s , A2<s ) (d) π2 T

 
(>)
Rs,N (x, y) 6 HN α, 2 ; <s, −2<s
2d π 2<s A
r (A.19)
√ √ A |<s|
for either <s > 0, N > 2 d or <s < 0, N > 2 d + .
π αT
(>)
As noted before, the finiteness of Rs,N (x, y) proves the absolute convergence for all
(>)
s ∈ C of the series (A.1), contained in Eq. (3.30) for Ds (x, y); the bound (3.50)
(>)
for the remainder Rs,N (x, y) of Eq. (3.49) follows straightforwardly from Eq. (A.19)
since we have
(>) 2d (>)
|Rs,N (x, y)| 6 Rs,N (x, y) . (A.20)
a1 ...ad |Γ(s)|
(>)
Let us mention that the series in Eq. (3.31) for ∂zw Ds can be discussed similarly;
(>)
in place of Rs,N (x, y), we use the quantity

(>)
X
(z,w) Rs,N (x, y) := ωn−2s Γ(s, ωn2 T ) ∂zw Cn (x, y) (A.21)
n∈Nd , |n|>N

(>)
and we estimate it similarly to what we did with Rs,N (x, y), noting that

π2 2
|∂zw Cn (x, y)| 6 |n| . (A.22)
a2
The conclusions of this analysis are the absolute convergence for all s ∈ C of the
series (3.31) and the remainder bound (3.51).

50
A.3 The series (A.2): connections with the expansions (3.36) (3.37) for
Ds(<) and its derivatives. Keeping in mind the definitions (3.16), (3.18) and
(3.33) of bhl , δl and Ps (and noting that |δl | = 1), let us put
 
(<)
X bhl (x, y)
Rs,N (x, y) := Ps− d , (A.23)
d
2 T
h∈Z , |h|>N
l∈{1,2}d

for s ∈ C and N ∈ (0, +∞) . Let us consider the series (A.2): clearly, this converges
(<)
absolutely if and only if Rs,N (x, y) < +∞ for some N .
(<)
Most of the sequel will be dedicated to evaluating Rs,N (x, y) . To this purpose we
√
first note that, for l ∈ {1, 2}d , h ∈ Zd and |h| > N > d, (21 )
√ !2 √ !2
d d
a2 1− |h|2 6 bhl (x, y) 6 A2 1 + |h|2 . (A.24)
N N

Using these bounds we easily infer, for all s ∈ C,

(d)
|bhl (x, y)s | 6 Ca,A (<s, N ) |h|2<s (A.25)

where, as in Eq. (3.48), we have put (for σ ∈ R)

 √ !!2σ √ !!2σ 
(d) d d
Ca,A (σ, N ) := max  a 1− , A 1+ .
N N

Next, notice that the identity (3.34) implies

|Ps (β)| 6 |β|<s Γ(−<s, β) for all s ∈ C, β ∈ (0, +∞) ; (A.26)

21
For example, in order to prove the first inequality in Eq. (A.24), recall the definition (3.18) of
bhl ; moreover, let us put U l (x, y) := (Uli (xi , y i ))i=1,...,d and note that
d d
X
2 X
|U l (x, y)|2 := Uli (xi , y i ) 6 1=d.
i=1 i=1

Then, for all i ∈ {1, ..., d}, we have

d
X
bkl (x, y) = a2i (hi − Uli (xi , y i ))2 > a2 |h − U l (x, y)|2 >
i=1
√ !2
2
 2
2
 √ 2 2 d
>a |h| − |U l (x, y)| >a |h|− d > a 1 − |h|2 .
N

51
in particular, from the above relation and Eq.s (A.3) (A.24) (A.25), we deduce
√
(d)
!
2 d 2
b 
hl C a,A (<s, N ) a (1− )
Ps 6 |h|2<s Γ −<s, N
|h|2 . (A.27)
T T <s T
(<)
Inserting the previous results into the definition (A.23) of Rs,N (x, y) we get (22 )
√
d (d)
!
d 2 d 2
(<) 2 C a,A (<s − 2
, N ) X d a (1− )
Rs,N (x, y) 6 |h|2<s−d Γ −<s, N
|h|2 ; (A.28)
T <s−d/2 d
2 T
h∈Z , |h|>N

now, using Eq. (A.13) we conclude the following, for any s ∈ C and any α ∈ (0, 1):
√
(d)
!
(<) 2d Ca,A (<s − d2 , N ) (d) a2 (1− Nd )2 d
Rs,N (x, y) 6 HN α, ; −<s, 2<s−d
T <s−d/2 T 2
s (A.29)
d √ d √ 1 (<s− d
)T
2
for either <s 6 , N > 2 d or <s > , N > 3 d + .
2 2 a α
(<)
As anticipated above, the finiteness of Rs,N (x, y) implies the absolute convergence
(<)
of the series (A.2), which appears in the expansion (3.36) for Ds (x, y); besides,
(<)
the bound (3.53) for the remainder Rs,N (x, y) in Eq. (3.52) follows easily from Eq.
(A.29) noting that
d
(<) T <s− 2 (<)
|Rs,N (x, y)| 6 d/2
Rs,N (x, y) . (A.30)
(4π) |Γ(s)|
A similar analysis can be developed for the series in the right-hand side of Eq. (3.37)
(<) (<)
for ∂zw Ds ; namely, in place of Rs,N (x, y) we consider
(<)
(z,w) Rs,N (x, y) :=
     
X bhl ∂z bhl ∂w bhl bhl ∂zw bhl
Ps− d −2 − Ps− d −1 (x, y) (A.31)
2 T T2 2 T T
h∈Zd , |h|>N
l∈{1,2}d

and estimate this quantity using Eq. (A.27), along with the following relations:
√ !
d 1
|∂zi bhl (x, y)| 6 A 1 + |h| , |∂zi wj bhl (x, y)| = δij
N 2 (A.32)
for z, w = x or y, i, j ∈ {1, ..., d} .
The final result proves, in this case, the absolute convergence of the series in the
right-hand side of Eq. (3.37) as well as the remainder estimate (3.54).
22
Note that, since the estimate obtained no longer depends on l, the sum over l ∈ {1, 2}d just
yields a multiplicative factor 2d .

52
 (d)
A.4 Asymptotics for HN . The upper incomplete gamma function is known to
possess the following asymptotic expansion (see [29], p.179, Eq.8.11.2):
Γ(s, z) = e−z z s−1 (1+O(z −1 )) for z ∈ R, z → +∞, and all s ∈ C . (A.33)
(d)
Comparing with the definition (3.46) of HN (for any σ ∈ R), we readily infer
(d) π d/2 β σ−2 e−4α βd −βN (N −4αβ √d) 2σ+ρ+d−4
HN (α, β; σ, ρ) = d
e N (1 + O(N −1 ))
α(1−α) Γ( 2 )
for N → +∞ ,
which is the result stated in Eq. (3.47).

B Appendix. The expansions of subsection 3.7

for the bulk energy
Consider the general framework of subsection 3.7, where we discuss the regularized
total energy that, due to the vanishing of the boundary contributions, coincides with
the regularized bulk energy E u . For the latter we have (see Eq. (3.78))
E u = E u,(>) + E u,(<) where
u Z
κ (•)
E u,(•) := dx1 ... dxd D u−1 (x, x) for • ∈ {>, <} .
2 (0,a1 )×...×(0,ad ) 2

Hereafter we show in several steps how to obtain the series expansions (3.79) and
(3.80) for E u,(>) and E u,(<) , respectively (23 ).
B.1 Derivation of the expansion (3.79). Let us first consider the function
(>)
E u,(>) ; using the expression (3.30) for the diagonal Dirichlet kernel D u−1 (x, x) ap-
2
pearing in the definition (3.78), we readily obtain
E u,(>) =
2d−1 κu
 Z
X
1−u u−1 2 (B.1)
ωn Γ , ωn T dx1 ... dxd Cn (x, x) .
a1 ...ad Γ( u−1
2
) d
2 (0,a1 )×...×(0,ad )
n∈N

Recalling the definition (3.11) of Cn (x, y), the integral in the above equation can be
straightforwardly evaluated to yield
Z d Z ai  
2 ni π i a1 ... ad
Y
1 d i
dx ... dx Cn (x, x) = dx sin x = , (B.2)
(0,a1 )×...×(0,ad ) i=1 0
ai 2d

which, along with Eq. (B.1), allows us to infer Eq. (3.79).

23
Concerning the interchange of certain sums with integrals or derivatives, recall the comments
of footnote 7 on page 15.

53
B.2 Derivation of the expansion (3.80). Let us now discuss the term E u,(<) ;
(<)
to this purpose, we consider the expression (3.36) for the Dirichlet kernel D u−1 (x, y).
2
When evaluated along the diagonal y = x, this expression reduces to
u−d−1  
(<) T 2 X bhl (x)
D u−1 (x, y) = δl P u−d−1 ; (B.3)
2 y=x (4π)d/2 Γ( u−1
2
) d d
2 T
h∈Z , l∈{1,2}

here and in the remainder of this appendix, for all h ∈ Zd , l ∈ {1, 2}d , we use the
short-hand notation (compare with Eq. (3.18))
d
X
bhl (x) := bhl (x, x) = a2i (hi − Uli (xi , xi ))2 . (B.4)
i=1

Eq.s (3.78) (B.3) allow us to infer for E u,(<) the representation

u−d−1
u,(<) κu T 2 (d,T )
E = d/2 u−1 Q u−d−1 (B.5)
2 (4π) Γ( 2 ) 2

where, for suitable s ∈ C and for all T ∈ (0, +∞), we have introduced the function
Z  
(d,T )
X
1 d bhl (x)
Qs := δl dx ... dx Ps . (B.6)
d d (0,a1 )×...×(0,ad ) T
h∈Z , l∈{1,2}

In the next subsection we show that this function can be expressed as

d √
π T )d−p X
 
(d,T )
X (− X Bσ,p (h)
Qs = aσ,p Ps+ d−p (B.7)
p=0
(d−p)!p! σ∈S h∈Zp
2 T
d

where the coefficients aσ,p , Bσ,p (h) are as in Eq. (3.81). Once (B.7) is proved, this
equation and (B.5) give the thesis (3.80).
B.3 Concluding the previous argument: derivation of Eq. (B.7). First of
(d,T )
all, notice that the definitions (B.6) and (3.33) of Qs and Ps give
Z 1 Z
1 bhl (x)
X
Qs(d,T )
= dτ τ s−1
δl dx1 ... dxd e− τ T ; (B.8)
0 (0,a1 )×...×(0,ad )
h∈Zd , l∈{1,2}d

this implies, recalling the definition (B.4) of bhl (x),

Z 1 d Z ai
Y X a2
i i i 2
Q(d,T
s
)
= dτ τ s−1
δli dxi e− τ T (hi −Uli (x ,x )) . (B.9)
0 i=1 hi ∈Z, li ∈{1,2} 0

54
Let us now focus on the general term in the product over i ∈ {1, ..., d}; explicitating
the sum over li ∈ {1, 2} and noting that U1 (xi , xi ) = 0, U2 (xi , xi ) = xi /ai (see Eq.
(3.16)), we get
Z ai
a2
i − τ iT (hi −Uli (xi ,xi ))2
X
δli dx e =
hi ∈Z, li ∈{1,2} 0

X Z ai Z ai  (B.10)
(ai hi )2
i − i − τ1T (ai hi −xi )2
= dx e τT − dx e .
hi ∈Z 0 0

The first integral in the square brackets above is trivial, since the integrand function
is constant; moreover
X Z ai 1 i 2
dxi e− τ T (ai hi −x ) =
hi ∈Z 0

X Z ai hi Z +∞ √ (B.11)
(z i )2 (z i )2
i − i −
= dz e τT = dz e τT = πτ T ,
hi ∈Z ai (hi −1) −∞

where in the second passage we have performed the change of variable z i := ai hi −xi .
Summing up, Eq. (B.10) yields
!
X Z ai
a2i i i 2
X (ai hi )2 √
δl i dxi e− τ T (hi −Uli (x ,x )) = ai e− τ T − π τ T . (B.12)
hi ∈Z, li ∈{1,2} 0 hi ∈Z

In the following we use the notations introduced in Eq. (3.81). Let us return to Eq.
(B.9) and consider the product over i ∈ {1, ..., d} therein; Eq. (B.12) allow us to
infer (24 )
d d
" ! #
Y X Z ai
a2i i i 2
Y X (ai hi )2 √
δli dxi e− τ T (hi −Uli (x ,x )) = ai e− τ T − πτ T =
i=1 hi ∈Z, li ∈{1,2} 0 i=1 hi ∈Z

d √
X (− π τ T )d−p X X 1
= aσ,p e− τ T Bσ,p (h) . (B.13)
p=0
(d−p)!p! σ∈S h∈Zp d

24
This result depends on the identity (already pointed out in footnote 10 of Part II of this series)

d d p
! d 
Y X 1 X Y Y
(ap + bp ) = aσ(i)  bσ(j) 
p=1 p=0
(d−p)!p! i=1 j=p+1 σ∈Sd

holding for any d ∈ {1, 2, 3, ...}, ap , bp ∈ R (p ∈ {1, 2, ..., d}), where by convention we intend
Q0 Qd
i=1 aσ(i) := j=d+1 bσ(j) := 1 .

55
Summing up, Eq.s (B.9-B.13) imply
d √
X (− π T )d−p X XZ 1 d−p 1 Bσ,p (h)
Q(d,T
s
)
= aσ,p dτ τ s+ 2 −1 e− τ T . (B.14)
p=0
(d−p)!p! σ∈S h∈Zp 0
d

To conclude and obtain Eq. (B.7), just note that all the integrals over τ ∈ (0, 1)
in Eq. (B.14) can be evaluated according to Eq. (3.33) to give the functions
Ps+ d−p ( Bσ,pT (h) ) .
2

C Appendix. Absolute convergence and remain-

der bounds for the series of subsection 3.7.
Let us recall that the expansion (3.79) for E u,(>) contains the series
 
X
1−u u−1 2
ωn Γ , ωn T . (C.1)
d
2
n∈N

On the other hand, the right-hand side of Eq. (3.80) for E u,(<) contains series in
h ∈ Zp , for p ∈ {0, ..., d}. Let us consider one of these series; after removing the
term with h = 0, which becomes singular at u = p + 1, we are left with
 
X Bσ,p (h)
P u−p−1 . (C.2)
h∈Zp , h6=0
2 T

In the following subsections we prove the absolute convergence of the series (C.1)
(C.2), for all u ∈ C ; moreover, we derive remainder estimates for both series, which
justify statements (3.85) (3.87). Throughout this appendix we adopt systematically
the notations introduced in Appendix A and the results obtained therein (see, in
particular, subsection A.1).
C.1 The series (C.1); connections with the expansion (3.79) of E u,(>) .
Let us define, for u ∈ C and N ∈ (0, +∞),
 
u,(>)
X
1−u u−1 2
RN := ωn Γ , ωn T . (C.3)
d
2
n∈N , |n|>N

u,(>)
If we can show that RN < +∞ for some N , of course the series (C.1) converges
absolutely.

56
 u,(>)
Hereafter we derive quantitative estimates for RN . Using the inequalities (A.17),
we infer from definition (C.3) that
max(a<u−1 , A<u−1 ) <u−1 π 2 T
X  
u,(>) 1−<u
RN 6 |n| Γ , |n|2 ; (C.4)
π <u−1 2 A2
n∈Nd , |n|>N

now, using the result (A.13) we obtain the following, for any α ∈ (0, 1) :
max(a<u−1 , A<u−1 ) (d) π 2 T <u−1
 
u,(>)
RN 6 HN α, 2 ; , 1−<u
2d π <u−1 A 2
r (C.5)
√ √ A |<u − 1|
for either <u > 1, N > 2 d or <u < 1, N > 2 d + .
π 2αT
u,(>)
Finally, let us point out that the remainder bound (3.85) for RN follows easily
from the above inequality noting that
u,(>) κ<u u,(>)
RN 6 u−1 RN . (C.6)
2 |Γ( 2 )|
C.2 The series (C.2); connections with the expansion (3.80) of E u,(<) .
Let us define the following functions, for u ∈ C, N ∈ (0, +∞), p ∈ {1, ..., d} and
σ ∈ Sd :  
u,(<)
X Bσ,p (h)
RN (σ, p) := P u−p−1 (C.7)
p
2 T
h∈Z , |h|>N

where Ps and Bσ,p are as in Eq.s (3.33) and (3.81), respectively.

u,(<)
Hereafter, we show that RN (σ, p) < +∞ for some suitable N ; this implies the
absolute convergence of the series (C.2).
u,(<)
In the rest of this paragraph we derive explicit estimates for the functions RN (σ, p),
for all u ∈ C, p ∈ {1, ..., d}, σ ∈ Sd and some suitable N . First of all, we note that,
for all h ∈ Zd , from the definition of Bσ,p (h) in Eq. (3.81) it follows (compare with
Eq. (A.24))
a2 |h|2 6 |Bσ,p (h)| 6 A2 |h|2 ; (C.8)
using these inequalitites we easily infer (for all u ∈ C)
u−1
Bσ,p (h) 2 6 max(a<u−1 , A<u−1 ) |h|<u−1 . (C.9)

Similarly to the derivation of Eq. (A.27), using the above bound along with Eq.s
(A.3) (A.26) we obtain
 
Bσ,p (h)
P u−p−1 6
2 T
(C.10)
max(a<u−p−1 , A<u−p−1 ) p+1−<u a2 2
 
<u−p−1
<u−p−1 |h| Γ , |h| .
T 2 2 T

57
 (<)
Inserting the previous results into the definition (C.7) of Ru,N (σ, n), we get
u,(<)
RN (σ, p) 6
<u−p−1 <u−p−1
p+1−<u a2 2
 
max(a ,A ) X
<u−p−1 (C.11)
<u−p−1 |h| Γ , |h| ;
T 2 2 T
h∈Zp , |h|>N

using Eq. (A.5) (with d replaced by n), we conclude the following, for any α ∈ (0, 1):

max(a<u−p−1 , A<u−p−1 ) (p) a2 p+1−<u

 
u,(<)
RN (σ, p) 6 <u−p−1 HN α, ; , <u−p−1 (C.12)
T 2 T 2
r
√ √ 1 |<u−p−1|T
for either <u 6 p+1, N > 2 p or <u > p+1, N > 2 p + .
a 2α
u,(<)
The above relation proves the finiteness of RN (σ, p) for all u ∈ C, p ∈ {1, ..., d}
and all σ ∈ Sd , which implies the absolute convergence of the series (C.2); moreover,
u,(<)
the remainder bound (3.87) for RN follows straightforwardly from Eq. (C.12),
noting that
<u−1 d
u,(<) κ<u T 2 X 1 X aσ,p u,(<)
RN 6 d+1 u−1 p/2
RN (σ, p) . (C.13)
2 |Γ( 2 )| p=1 (d−p)!p! σ∈S (π T )
d

58
References
[1] A.A. Actor, Local analysis of a quantum field confined within a rectangular
cavity, Ann.Phys. 230, 303–320 (1994).
[2] A.A. Actor, Scalar quantum fields confined by rectangular boundaries,
Fortsch.Phys. 43, 141–205 (1995).
[3] J. Ambjørn, S. Wolfram, Properties of the vacuum I. Mechanical and thermo-
dynamic, Ann.Phys. 147(1), 1–32 (1983).
[4] J. Ambjørn, S. Wolfram, Properties of the vacuum II. Electrodynamic,
Ann.Phys. 147(1), 33–56 (1983).
[5] M. Bordag, G.L. Klimchitskaya, U. Mohideen, V.M. Mostepanenko, “Ad-
vances in the Casimir Effect”, Oxford University Press (2009).
[6] O. Calin, D.-C. Chang, K. Furutani, C. Iwasaki, “Heat Kernels for Elliptic
and Sub-elliptic Operators: Methods and Techniques”, Springer Science &
Business Media (2010).
[7] B. Deconinck, M. Heil, A. Bobenko, M. van Hoeij, M. Schmies, Computing
Riemann Theta Functions, Math.Comput. 73, 1417–1442 (2004).
[8] A. Edery, Multidimensional cut-off technique, odd-dimensional Epstein zeta
functions and Casimir energy of massless scalar fields, J.Phys.A:Math.Gen.
39, 685-712 (2006).
[9] A. Edery, Casimir piston for massless scalar fields in three dimensions,
Phys.Rev.D 75, 105012 (2007).
[10] E. Elizalde, S.D. Odintsov, A. Romeo, A.A. Bytsenko, S. Zerbini, “Zeta reg-
ularization techniques with applications”, World Scientific (1994).
[11] E. Elizalde, “Ten Physical Applications of Spectral Zeta Functions”, Springer
Science & Business Media (1995).
[12] D. Fermi, L. Pizzocchero, Local zeta regularization and the scalar Casimir
effect I. A general approach based on integral kernels, arXiv:1505.00711 (2015).
[13] D. Fermi, L. Pizzocchero, Local zeta regularization and the scalar Casimir
effect II. Some explicitly solvable cases, arXiv:1505.01044 (2015).
[14] D. Fermi, L. Pizzocchero, Local zeta regularization and the scalar Casimir
effect III. The case with a background harmonic potential. arXiv:1505.01651
(2015).

59
[15] R. Estrada, S.A. Fulling, L. Kaplan, K. Kirsten, Z.H. Liu, K.A. Milton, Vac-
uum stressenergy density and its gravitational implications, J. Phys. A: Math.
Theor. 41 (2008), 164055 (11pp)

[16] S.A. Fulling, L. Kaplan, K. Kirsten, Z.H. Liu, K.A. Milton, Vacuum Stress
and Closed Paths in Rectangles, Pistons, and Pistols, J.Phys.A 42, 155402
(2009).

[17] L. Greengard, P. Lin, A fast algorithm for the evaluation of heat potentials,
Comm.Pure App.Math. 43(8), 949-963 (1990).

[18] L. Greengard, P. Lin, Spectral Approximation of the Free-Space Heat Kernel,

App.Comp.Harm.Analysis 9, 83-97 (2000).

[19] X. Li, H. Cheng, J. Li, X. Zhai, Attractive or repulsive nature of the Casimir
force for rectangular cavity, Phys.Rev.D 56(4), 2155–2162 (1997).

[20] W. Lukosz, Electromagnetic Zero-Point Energy and Radiation Pressure for a

Rectangular Cavity, Physica 56, 109–120 (1971).

[21] W. Lukosz, Electromagnetic Zero-Point Energy Shift Induced by Conducting

Surfaces II. The Infinite Wedge and the Rectangular Cavity, Z.Physik 262,
327–348 (1973).

[22] S.G. Mamaev, N.N. Trunov, Vacuum means of energy-momentum tensor of

quantized fields on manifolds of different topology and geometry. I, Sov.Phys.J.
22(7), 766-770 (1979).

[23] S.G. Mamaev, N.N. Trunov, Vacuum averages of the energy-momentum ten-
sor of quantized fields on manifolds of various topology and geometry. II,
Sov.Phys.J. 22(9), 966-969 (1979).

[24] S.G. Mamaev, N.N. Trunov, Vacuum expectation values of the energy-
momentum tensor of quantized fields on manifolds with different topologies
and gometry. III, Sov.Phys.J. 23(7), 551-554 (1980).

[25] S.G. Mamaev, N.N. Trunov, Vacuum expectation values of the energy-
momentum tensor of quantized fields on manifolds of different topology and
geometry. IV, Sov.Phys.J. 24(2), 171-174 (1981).

[26] F. D. Mera, S. A. Fulling, Vacuum energy density and pressure of a massive

scalar field, J. Phys. A: Math. Theor. 48, 245402 (18pp) (2015).

[27] S. Minakshisundaram, A generalization of Epstein zeta functions,

Canad.J.Math. 1, 320–327 (1949).

60
[28] C. Morosi, L. Pizzocchero, On approximate solutions of semilinear evolution
equations II. Generalizations, and applications to Navier-Stokes equations,
Rev.Math.Phys. 20(6), 625–706 (2008).

[29] F.W.J Olver, D.W. Lozier, R.F. Boisvert, C.W. Clark, “NIST Handbook of
mathematical functions”, Cambridge University Press (2010).

[30] R.B. Rodrigues, N.F. Svaiter, R.D.M. De Paola, Vacuum Stress Tensor of a
Scalar Field in a Rectangular Waveguide, arXiv:hep-th/0110290 (2001).

[31] J.R. Ruggiero, A. Villani, A.H. Zimerman, Application of Analytic Regular-

ization to the Casimir Forces, Rev.Bras.Fis. 7(3), 663–687 (1977).

[32] J.R. Ruggiero, A. Villani, A.H. Zimerman, Some comments on the application
of analytic regularisation to the Casimir forces, J.Phys.A:Math.Gen. 13, 761–
766 (1980).

61